A Reference Functional Architecture for Network Digital Twins in 6G Systems (Extended Version)

This document is an extended white paper accompanying the manuscript: A. Zaki-hindi et al., A Reference Functional Architecture for Network Digital Twins in 6G Systems, submitted for publication, 2025. This document has not been peer reviewed. For citation purposes, please cite the journal manuscript. 1 A Reference Functional Architecture for Network Digital Twins in 6G Systems (Extended Version) Ayat Zaki-Hindi ∗ , Paola Soto † , German Castellanos ‡ , Touhid Hossain Pritom § , Gaetano Volpe ¶ , Iskander Zellagui ∥ , Burkhard Hensel § , Julian Jimenez † , Michele Marvulli ¶ , Luís Santos ∗∗ , Ayse Sayin †† , Jean-Sébastien Sottet ∗ , Wasim Ali¶, Ines EL-Korbi∥, Ion Turcanu∗, Andrey Belogaev†, André Duarte∗∗, Ultan Kelly‡‡, Sumit Kumar∗, Georgy Myagkov x , Stephen Parker ‡ , Mario Franke § , Shajjad Hossain ∥ , Rajarshi Sanyal xi , Nida Shafi x , Christoph Sommer § , Miguel Camelo†, Julien Baudouinxi , Régis Decormexii , Maria Pia Fanti¶, Ramin Fuladi††, Chris Murphyx, Ana Pereira∗∗, Sidi Mohammed Senouci∥, Simon Pryor‡, Sébastien Faye∗ ∗Luxembourg Institute of Science and Technology (LIST), Luxembourg †IDLab, University of Antwerp - imec, Belgium ‡Accelleran, Belgium §TU Dresden, Faculty of Computer Science, Germany ¶Politecnico di Bari, Italy ∥ Université de Bourgogne, France ∗∗ Ubiwhere, Portugal †† Ericsson Ara¸stırma Geli¸stirme ve Bili¸sim Hizmetleri A.¸S., Turkey ‡‡Viavi Solutions Ireland Ltd, Ireland xViavi Solutions UK Ltd, UK xi Proximus Luxembourg S.A., Luxembourg xii R2M Solution, France Corresponding author: [email protected] Abstract—AI-native, programmable and disaggregated 6G networks will be highly dynamic and distributed, demanding tools that can explain, predict, and safely optimize behavior across the edge–cloud continuum. Network Digital Twins (NDTs) promise this capability, yet current efforts in research and industry are fragmented and lack widely accepted formal definitions and architectural guidelines. This paper proposes a structured framework for NDTs in 6G, addressing these gaps by refining the conceptual foundations of NDTs, introducing a functional architecture, inherited from the 6G-TWIN EU consortium, and clarifying key components such as AI-driven workflows, the place of simulation, data management, and orchestration. Concrete examples illustrate how these components enable network automation, optimization, and predictive analytics. The paper proceeds by reviewing related work and standardization efforts, specifying functional and non-functional requirements, presenting the architecture and its various domains, and detailing lifecycle management across cloud to edge. We then report early implementations and evaluation results, and discuss security, privacy, and governance considerations, concluding with directions for validation and uptake. The key objective is to offer a cohesive reference model that guides the community in shaping NDT development, ensuring interoperability, scalability, adaptability, and seamless integration into AI-native 6G networks for improved intelligence and efficiency. Index Terms—6G, AI-Native Network Orchestration, ClosedLoop Network Automation, Edge–Cloud Continuum Management, Federated Co-Simulation, Network Digital Twin I. INTRODUCTION Future 6G networks will operate under unprecedented levels of complexity, driven by ultra-dense deployments, heterogeneous access technologies, stringent latency and reliability constraints, and pervasive intelligence at the network edge. The number of connected applications and services will far exceed those of previous generations, with traffic demand expected to be nearly ten times higher than when 5G was first introduced in 2018 [ 1 ]. These trends will also introduce new constraints from verticals such as connected mobility, digital health, and public protection and disaster relief. Each of these domains will impose distinct requirements on reliability, latency, and distributed intelligence across the edge-cloud continuum, while ensuring energy efficiency and seamless operation across diverse and dynamic network environments. To manage such systems, network operators will require continuous visibility, predictive control, and automated assurance across the entire service lifecycle, from design and planning to real-time operation and optimization. Conventional management frameworks, largely static and rule-based, can no longer cope with the dynamics, heterogeneity, and scale of next-generation networks. Instead, networks must evolve toward dataand model-driven operation, combining predictive intelligence (i.e., anticipating behaviors, trends, and anomalies beyond the limits of traditional simulation) with real-time adaptive control (i.e., ensuring continuous optimization and resilience under rapidly changing conditions). In this context, Network Digital Twins (NDTs) are emerging as a cornerstone of next-generation network management. An NDT is a virtual, continuously synchronized replica of the physical network that captures its topology, configuration, performance, and environment. Through bidirectional data exchange, the NDT mirrors the state of the live network, enabling real-time monitoring, predictive analytics, “what-if” experimentation, and closed-loop optimization. NDTs thus act as cognitive mirrors and safe sandboxes of the network to test policies, train Artificial Intelligence (AI) models, and anticipate failures before they affect live services. The integration of NDTs into 6G architectures is therefore essential. The 6G vision extends beyond connectivity to create intelligent, sustainable, and self-adaptive infrastructures spanning the edge-cloud continuum. Achieving this vision requires a management paradigm in which the network continuously reasons about its own behavior, learns from historical and contextual data, and refines its operation autonomously. NDTs provide the means to achieve this: they support closed-loop 2 learning and optimization at multiple timescales, facilitate the safe introduction of AI-based control, and underpin key functions such as network planning, slice assurance, and energy optimization. Incorporating NDTs into the operational lifecycle of 6G, including design, deployment, operation, and assurance, ensures that the network remains both agile and trustworthy as conditions evolve. However, orchestrating NDTs efficiently across distributed 6G infrastructures introduces significant challenges [ 2 ]. The heterogeneity of data sources, compute nodes, and administrative domains demands federated, context-aware orchestration and semantic interoperability [ 3 ]. From a literature perspective, existing efforts remain fragmented: while several standardization bodies (e.g., ITU-T SG13, 3GPP, ETSI) and research works have proposed concepts or domain-specific digital twins, proposals are either too abstract for implementation or too narrow to generalize beyond a particular segment (e.g., RAN, core, or transport). Critically, there is no widely accepted, formal scope and lifecycle for NDTss, limited integration between AI training and simulation pipelines, and insufficient coordination across the edge-cloud continuum. This fragmentation underscores the need for an architecture that explicitly governs orchestration and lifecycle management across the entire stack, from cloud to edge. This paper addresses these gaps by proposing a comprehensive functional architecture for NDTs in 6G systems, designed to enable scalable, interoperable, and AI-driven orchestration from cloud to edge. Building upon our previous work [ 4 ], which introduced the first conceptual framework from the Horizon Europe 6G-TWIN project [ 5 ], [ 6 ], [ 7 ], [ 8 ], [ 9 ], [ 10 ], [ 11 ], we extend and consolidate that vision through four key advancements: • We formalize a working definition, scope, and lifecycle for NDTss in 6G, consolidating fragmented literature/industry efforts and deriving functional and non-functional requirements to guide design and evaluation. • We present a reference functional architecture, decomposed into application, physical, digital, and management domains, with clear interfaces and a harmonized data pipeline. • We specify lifecycle-aware procedures for NDT instances (instantiation, synchronization, model selection/update, federation, and safe actuation) that operate coherently across the full cloud–edge continuum. • We integrate AI and simulation via coordinated MLOps and co-simulation workflows, distinguishing basic and functional models to support explainable “what-if” experimentation and closed-loop optimization. • We validate the approach through early implementations, showcasing different but complementary assets of the NDTs while discussing security, privacy, and data governance. Beyond the proposed functional architecture, the paper also introduces preliminary implementation results demonstrating early NDT deployments that focus on specific architectural aspects such as data harmonization, AI-based modeling, and simulation-driven optimization. Furthermore, we outline deployment considerations critical to ensuring secure, privacypreserving, and sustainable integration of NDTs into operational networks. The remainder of the paper is structured as follows. Section II reviews related work and existing architectural proposals. Section III presents the functional and non-functional requirements of the proposed NDT architecture. Section IV introduces our proposal for an AI-native functional architecture that integrates NDT as a central element. Sections V to VIII detail the application, physical, digital, and management domains. Section IX discusses NDT lifecycle management and mapping to representative 6G scenarios. Section X discusses implementation details and evaluation results. Section XI concludes the paper by addressing security, privacy, governance, and future research directions. II. RELATED WORK This section reviews the current state of NDT architectures as defined and explored by various technology-shaping entities, including standardization bodies, academic research, and industrial initiatives. The discussion highlights each sector’s key contributions, conceptual orientations, and distinguishing features. A. Standard Development Organizations (SDOs) The International Telecommunication Union - Telecommunication Standardization Sector (ITU-T), through its recommendation Y.3090 [ 12 ], has established one of the baseline reference architectures for NDTs, as abstracted in Figure 1. The framework is structured into three layers: the application layer, the Digital Twin (DT) layer, and the physical network layer. The physical layer includes all real and virtualized elements of the network, such as gNodeB (gNB), User Equipments (UEs), spectrum resources, and core network functions. The DT layer defines the essential NDT functionalities, hosting the data, models, and management entities necessary to represent and control the network. It integrates components such as the Unified Data Repository (UDR), Unified Data Model repository (UDM), and DT entity management. The application layer operates above, guiding the creation of NDT instances in response to application-specific requirements and transmitting control commands toward the physical network via the twin. The ITU-T identifies four key enablers underpinning the architecture: data, mapping, models, and interfaces. Data constitutes the foundation, providing accurate and continuously updated information within a unified repository. Mapping mechanisms ensure real-time synchronization between the digital and physical domains, distinguishing NDTs from traditional simulation systems. Models within the DT embody both basic and functional aspects of the physical network entities. Finally, standardized interfaces ensure scalability and interoperability, where southbound interfaces connect the DT to the physical network, and northbound interfaces link it to external applications and management systems. Other standardization entities have also advanced NDTrelated initiatives. The 3rd Generation Partnership Project (3GPP), within the technical report 28.915 [ 13 ], investigates 3 Network application Physical network Network Digital Twin Visualization Optimization … UE Infrastructure Computing resources NDT management Unified data models Unified Data Repository Capability exposure Intent input Data collection Control Functional models Basic models Security mgmt Topology mgmt Model mgmt Figure 1. Baseline reference architecture from ITU-T how NDTs can integrate with network automation and management functions. This includes the role of NDTs in enabling predictive and closed-loop automation and in supporting specific use cases relevant to future 6G networks. Likewise, the Open RAN Alliance (O-RAN) Alliance [ 14 ] recognizes NDTs as a pivotal enabler for next-generation Radio Access Network (RAN) intelligence and automation, emphasizing their application to RAN optimization and Lifecycle Management (LCM). The European Telecommunications Standards Institute (ETSI) [ 15 ] and the Internet Engineering Task Force (IETF) [ 16 ] provide complementary architectural viewpoints consistent with ITU-T’s recommendations. ETSI outlines how NDTs can be incorporated into the Zero-touch Service and network Management (ZSM) framework, highlighting their role in enabling automated, closed-loop service orchestration. It details the necessary functional components, standardized interfaces, and architectural capabilities to integrate NDTs within ZSM-based management systems. Conversely, the IETF proposes a reference architecture that delineates NDT functional components, their interactions, and the corresponding interfaces to ensure seamless coupling with live network infrastructure. Its approach emphasizes the use of NDTs for real-time modeling, analytics, simulation, and feedback mechanisms to enhance decision-making, assurance, and LCM across network operations. B. Industry The telecommunications industry has increasingly embraced the concept of NDTs as a key enabler for managing the growing complexity of network infrastructures. Industrial stakeholders recognize NDTs as tools that enhance operational efficiency, support predictive maintenance, and facilitate data-driven network planning and automation. The following examples illustrate how leading vendors and operators are advancing NDT-related technologies. A prominent contribution from the industrial domain is presented in Spirent’s white paper [ 17 ], which conceptualizes NDTs as combined software and hardware emulations of real 5G networks. This approach enables iterative prototyping, validation, and performance assurance of network configurations. Spirent’s proposed architecture emphasizes modularity, allowing flexible testing and integration of different network components and functions. Similarly, ZTE’s white paper [ 18 ] introduces an evolved NDT architecture incorporating an additional service layer that aggregates functionalities into microservices. This design enhances scalability and maintainability by structuring nine independent microservices, each encapsulating distinct capabilities, within a framework that remains agnostic to programming languages, data storage formats, or underlying architectures. Ericsson’s work explores the convergence of NDTs and AI/Machine Learning (ML) technologies within 5G and future 6G RANs [ 19 ]. The company emphasizes the importance of standardization, particularly through collaboration with 3GPP, to define common interfaces, data flows, and control mechanisms for AI/ML-enabled functions. Such efforts aim to ensure interoperability across vendors and to pave the way for AI-native cellular networks capable of autonomous optimization. Ericsson has also demonstrated practical advancements in collaboration with Deutsche Telekom and Google Cloud, deploying a cloud-native 5G Core on Google Distributed Cloud Edge to improve deployment speed, efficiency, and compliance with EU data protection regulations [20]. Nokia’s vision for AI-native 6G integrates AI/ML directly into the air interface, enabling adaptive radio systems that dynamically adjust to network and environmental conditions [ 21 ], [ 22 ]. Its Dynamic Digital Twin initiative exemplifies this approach, creating continuously updated virtual representations of networks to support predictive operations, fault prevention, and adaptive model optimization [ 23 ]. These real-time capabilities enhance both network resilience and management efficiency. Huawei similarly advocates for AI-centric architectures in the evolution toward 6G networks. Its “Intelligent World 2030” strategy highlights the role of DTs in achieving selflearning, predictive, and self-optimizing networks [ 24 ], [ 25 ]. By embedding AI throughout the network stack: from perception to control, Huawei envisions intelligent systems capable of autonomous management and sustainable energy-efficient operation, further reinforcing the convergence of NDT and AI-native paradigms in future communication networks. C. Academia Academic research began exploring the application of AI/ML techniques to communication networks well before the emergence of the NDT concept. As early as 2015, Wang et al. [ 26 ] published a comprehensive survey detailing how AI-based methods can address challenges in management, optimization, and maintenance of heterogeneous and fastevolving mobile networks. The use of the term “Digital Twin (DT)” in the context of networks appeared later, with Dong et al. [ 27 ] introducing a framework that employs Deep Neural Networks (DNNs) to enhance energy efficiency in Mobile Edge Computing systems, targeting both latency-critical and delay-tolerant services. Subsequent academic studies have proposed alternative perspectives on NDTs compared to those defined by SDOs. For example, Zhou et al. [ 28 ] introduced a hierarchical DT framework for satellite communication networks. Their 4 design deploys edge-DTs co-located with ground stations, each maintaining a model of its associated physical entity, comprising satellites, terminals, and wireless links. These edge-DTs perform real-time tasks such as beam scheduling, fault diagnosis, resource allocation, and data processing, while exchanging information with central-DTs hosted in a centralized control facility. Central-DTs oversee global network management and can create multiple, isolated virtual instances for verification, optimization, traffic engineering, and slicing control. By ensuring that each physical asset maps to a single edge-DT, the architecture effectively minimizes resource redundancy and system overhead. Several studies have also concentrated on RAN-centric NDT architectures. Vilà et al. [ 29 ] proposed a RAN-focused framework aligned with both O-RAN Alliance and IETF guidelines [ 16 ]. Their architecture is structured around three main components: a data repository, a suite of service mapping models (including both basic and functional models), and a twin management entity. The design allows each application to operate its own RAN NDT instance, customized according to specific requirements such as active service mapping models, target Key Performance Indicators (KPIs), data selection rules, and visualization or emulation capabilities. Kuruvatti et al. [ 30 ] offered a comprehensive survey summarizing requirements, challenges, and use cases for NDTs. Their work synthesizes recommendations from both the IETF and ITU-T, identifying four fundamental components of a DT: data, mapping, model, and interface. They propose a three-layer, three-domain, and dual closed-loop architecture consistent with the principles outlined in ITU-T Recommendation Y.3090 [ 12 ]. D. Differences with the related work The work on NDT architectures and their interaction with the physical network has been mainly led by prominent SDOs such as ITU-T [ 12 ] and IETF [ 16 ]. While agreeing on the main building blocks and NDT capabilities such as mapping, models, and data, both works lack in detailing the interaction between the physical and the digital network, how the data should be collected, how AI and non-AI models should be created, and what the role of simulators is within the NDT. Industry and most research work in academia focus on the application of NDTs within the RAN domain [ 31 ], [ 29 ], [ 32 ], as this represents one of the most intricate and resource-intensive components of 6G infrastructure. In general, we can see that existing works lack a unified, practical framework, often being too abstract or too narrow. Key aspects like AI integration, automation, and scalability remain insufficiently addressed, motivating our proposal for a more concrete and adaptable NDT architecture. Some studies [ 33 ], [ 34 ], [ 35 ] propose AI-native architectures for future communication networks and acknowledge the potential role of NDTs in supporting network operations; however, they do not provide detailed methodologies for their integration or explicitly incorporate NDTs within the proposed architectural frameworks. Finally, our previous work presents an initial AI-Native network architecture that incorporates NDTs to optimize, manage, and control future networks in real-time [ 36 ] and its updated version [ 4 ]. While the initial architecture was heavily inspired by the ITU-T, the updated version included four key pillars: (i) a data collection framework for dynamic data acquisition and harmonization, (ii) ZSM for AI-driven automation, (iii) Federated Management for decentralized network control, and (iv) the simulation framework for predictive modeling. The architecture presented in this paper, as it will be shown in Section IV, further refines the architecture to introduce an NDT MANagement and Orchestration (MANO) layer as a unifying vertical layer that integrates all the aforementioned domains. This layer not only incorporates the MANO functions of both the physical and digital domains, including federated management, simulation, and AI workflows, but also establishes a clear structural boundary between the core NDT components and the instantiated NDT subcomponents, thereby defining the operational interface between the digital and physical domains. III. NDT FUNCTIONAL AND NON-FUNCTIONAL REQUIREMENTS When creating a software tool, one of the first things to do is to list the Functional Requirements (FRs) and NonFunctional Requirements (NFRs). This is not different for an NDT since it lays the foundation for aligning the NDT with the needs of users, stakeholders, and the technical challenges posed by 6G environments, making this phase extremely important. To ensure effective implementation, we propose both FR (as outlined in Table I) and NFR (as detailed in Table II) for an NDT architecture. Both tables reflect the FRs and NFRs for building an NDT and the requirements to build the blocks in the NDT MANO layer. The key requirements for building our architecture focus on ensuring a high-fidelity, interoperable, and AI-driven NDT ecosystem capable of real-time network management and optimization. Functionally, the NDT must accurately model and interact with the physical network through standardized interfaces, support both basic and functional models, and integrate AI/ML capabilities for autonomous operation. It should enable analytical and controlling modes, operate in federated environments, and rely on a robust data collection framework ensuring harmonization, security, and interoperability across domains. The MANO layer, enhanced with ZSM, must orchestrate NDT lifecycle processes including creation, deployment, updates, and deletion, while supporting AI-based workflows and standardized Application Programming Interfaces (APIs). Similarly, the simulation framework must synchronize heterogeneous simulators, manage configurations, and maintain closed-loop interoperability with other system components. Non-functionally, the architecture must guarantee scalability, low latency, reliability, and security, with mechanisms for redundancy, access control, and fault tolerance. Its modular, service-based design should ensure maintainability and compatibility across technologies. Moreover, it must integrate Developer Operations (DevOps), Machine Learning Model Operationalization Management (MLOps), and cloud-native principles to achieve efficient, secure, and flexible orchestration 5 Table I FUNCTIONAL REQUIREMENTS OF THE NDT ARCHITECTURE FR ID: Description FR.NDT.01: An NDT should be able to model and represent the physical network with high fidelity, building accurate network models that can operate in real-time, to represent network elements and topologies and models for network analysis, prediction and assurance. FR.NDT.02: The NDT should interact in real-time with the physical network. Southbound interfaces are required for information exchange between the physical and the DT. Northbound interfaces are required for information exchange between NDT and network applications. FR.NDT.03: The NDT should support basic and functional models for network applications in the form of unified data models. FR.NDT.04: The NDT should support an efficient and unified data collection and repository. FR.NDT.05: The NDT should have a management system able to define, instantiate, update, and delete NDT instances. FR.NDT.06: The NDT should support the integration with AI/ML models. FR.NDT.07: The NDT should be able to work as an analytical NDT or a controlling NDT. An analytical NDT primarily focuses on understanding, evaluating, diagnosing, and predicting network behaviour and performance without directly enacting changes on the physical network. A controlling NDT goes beyond analysis to directly influence, manage, and optimize the physical network, often through a closed-loop system. FR.NDT.08: An NDT instance should be able to work together with other NDT instances in a federated manner. FR.DC.01: The Data Collection Framework ensures the integration of data from various sources and across various domains, extending its capabilities towards the Cloud-to-FarEdge continuum. FR.DC.02: The Data Collection Framework provides harmonisation mechanisms for different data formats and standards, ensuring a distributed model for securely sharing and storing data across multiple locations, while also guaranteeing efficiency in data processing, analysis, and scalability. FR.DC.03: The Data Collection Framework enforces robust access policies that support multiple communication protocols and ensure privacy and security among the connected devices, including current and legacy protocols. FR.ZSM.01: The ZSM allows a fully automated network management system, discovering and integrating different data points while automatically monitoring the real-time status of the NDT FR.ZSM.02: The ZSM enables the management and orchestration of network resources, using AI-based NFs/NSs to optimize their allocation, automating CI/CD mechanisms, and supporting programmable interfaces for NDT control. FR.ZSM.03: The ZSM ensures interoperability between new and existing services, protecting APIs and network resources using necessary Authorisation/Authentication/Accounting (AAA) protocols, allowing for the NDT to have secure and reliable communications. FR.MANO.01: The F-MANO supports AI-based NFs/NSs to allow for cross-domain and multi-time scale network Lifecycle Management (LCM), managing the deployment and orchestration of workloads. FR.MANO.02: The F-MANO provides a standardized way to offer MLOps support to the created AI models, handling the relevant CI/CD practices. FR.MANO.03: The F-MANO exposes a comprehensive and standardized API for secure and reliable communications across the NDT ecosystem, including LCM mechanisms and cross-domain AI-based NFs/NSs. FR.SIM.01: The Simulation framework employs specifications that describe communication protocols and data structures between itself and the different simulators. Relevant simulation services, such as semantic description of data structures, time management services, and exception handling, are also handled by the Simulation Framework. FR.SIM.02: The Simulation Framework integrates the necessary basic and functional models on the simulators as they are available, custom-built according to model parameters. FR.SIM.03: The Simulation Framework tailors the models with machine-readable descriptions of computing, storage, and functionality requirements, such as model deployment and metric gathering, according to the simulation goals, FR.SIM.04: The Simulation Framework establishes a protocol for the closed-loop framework of the NDT, allowing for the transfer of important simulation information regarding configurations and results. FR.SIM.05: The Simulation Framework produces abstract specifications and configurations for interoperability. It also establishes a synchronisation for time management, event distribution and global state for different working simulators. of distributed AI-enabled network functions and simulations in real time. IV. NDT ARCHITECTURE OVERVIEW Building upon the requirements outlined in Section III, Figure 2 represents the proposed functional architecture, designed as an integrated framework that unifies NDTs with AI-driven functionalities and simulation components under Table II NON-FUNCTIONAL REQUIREMENTS OF THE NDT ARCHITECTURE NFR ID: Description NFR.NDT.01: The NDT should support the modelling of large, distributed networks, and managing high data volumes and device counts, including lightweight twin instances for edge computing. NFR.NDT.02: The NDT should enable low-latency processing, real-time responsiveness, and high-throughput data ingestion; ensures fast operation and response to network changes. NFR.NDT.03: The NDT should ensure fault-tolerant data collection and model execution with redundancy and high availability mechanisms. NFR.NDT.04: The NDT system should implement encryption, access control, authentication, data masking, and integrity checks; secures data collection, transmission, storage, and model behaviour. NFR.NDT.05: The NDT system should be modular, leveraging the service-based architecture for easy updates and agile maintenance. NFR.DC.01: The Data Collection Framework should optimise system resource usage in order to ensure optimal performance of network operations, real-time data analysis, while ensuring minimal delay. NFR.DC.02: The Data Collection Framework should be compatible with all devices and technologies, ensuring interoperability and seamless connectivity, improving the efficiency and flexibility of data collection. NFR.DC.03: The Data Collection Framework should guarantee strong security and privacy measures to protect data from unauthorized access. NFR.ZSM.01: The ZSM can integrate multiple NDTs for real-time data analytics and automated decision-making across multiple domains and time scales, guaranteeing standardised APIs and interfaces for interoperability between devices and domains, as well as the ability to handle failures. NFR.ZSM.02: The ZSM will integrate Network analysis, planning, management, and control operations, by using AI-based functions and DevOps principles on running workloads, minimizing latency and optimising energy efficiency. NFR.ZSM.03: The ZSM shall comply with NDT requirements regarding a secure and private environment across domains. NFR.MANO.01: The F-MANO can integrate multiple NDTs for real-time data analytics and decision-making across multiple domains, seamlessly integrating various network elements, resorting to relevant industry standards and regulations. NFR.MANO.02: The F-MANO will use AI-based NFs/NSs for data analytics and decision-making purposes, maintaining network operations by adhering to GitOps, DevOps, and MLOps principles to ensure efficient management and orchestration support to the NDT. NFR.MANO.03: The F-MANO will support multiple NDT instances and network elements across multiple network domains, efficiently scaling AI model training according to demand and following a cloud-native approach to manage workload LCM. NFR.SIM.01: The Simulator Framework establishes a platform-independent solution through multiple programming languages, using Open-Source software interfaces NFR.SIM.02: The Simulation Framework federation overhead allows for secure communication between components and flexible simulations. NFR.SIM.03: The Simulation Framework will parameterise simulations, to guarantee an efficient verification and validation process, as well as an efficient output of metrics and results. a unified management layer. This functional architecture is designed to operate as a closed-loop system, where data flows continuously between the real-world network, its digital representation, and the management functions that synchronize them. To structure this interaction, the architecture is organized into four interconnected domains: application,physical,digital, and management. Each domain fulfills a distinct role while contributing to the end-to-end integration of the NDT. The application domain contains the dashboard equipped with the interfaces through which humans interact with the NDT. The dashboard serves as the human-in-the-loop entry point, enabling stakeholders to visualize network state, configure NDT uses, and monitor NDT operations in real time. The physical domain comprises the infrastructure network, including RAN, core, transport, and edge/cloud computing domains. This domain has a dual function: 1) it provides realworld telemetry and context data to the digital domain, 2) receives insights, optimization decisions, and controls actions generated by the NDT. The digital domain contains the digital representation of the physical network and the mechanisms needed to analyze 6 NDT Instance Network applications Data collection: telemetry collector, databases, data buses, … Actuation and execution Physical Network Digital Network - aka NDT - Human world Functional models: Energy savings in dense deployments Teleoperated driving Real-world representation User equipment Environment Radio access network Core network Network MANO with ZSM capabilities Simulator 1 Simulation model 1 Simulation model 2 Simulator 2 Simulation model 3 Simulation model 4 Interface layer Use-case instantiation & results Unified data repository: structured and harmonised data … Results, inc. linked KPIs Push/request data to/from the real network Push/request data, metrics (i.e., predicted), and models Effect changes Open and secured NDT modelling and simulation framework RIC Tester Real CU/DU/RU:ACC, PX, … NDT Instance Link with the reality RIC Core: Free5GC, OAI, … Ideas and scenarios: Snapshot & requirements pushed to the framework „Scenario builder“ interface inc. NDT functionality „marketplace“ Real-time monitoring Complete, persistent view “What-if” representation for testing & data generation - simulation Incomplete, on-demand view (snapshot) AI training / optimizer (functional models) Unified data model repository Training data from simulation or real network Trained model Unified dashboard AI training methods Optimization methods Comparison of what-if simulation results Basic models: representations of network components and behaviour Functional models:Traffic analysis & generation Network planning Network management & control 6G-TWIN Management and Orchestration Layer Physical Network MANO Simulation Management NDT Management: mapping between scenario requests and NDT‘s data/models/simul ation; life-cycle management. Data Management NDT Model Management NDT Translation LayerResource ManagementNDT Orchestrator NDT Monitor AI Workflow Management Federated Management Management and Orchestration Instructions Figure 2. 6G NDT functional architecture and predict its behavior. It integrates the data-driven digital representation of the physical network, simulation engines, and AI training pipelines. This domain ensures that the NDT continuously reflects the state of the physical network while enabling predictive modeling, what-if analysis, and the creation of deployable functional models that are validated in the NDT before being instantiated in the physical domain through the NDT MANO to enable real-time optimization and control. The management domain oversees the instantiation, synchronization, and lifecycle of NDT components. It orchestrates interactions between domains, ensuring that data flows are harmonized, models remain consistent, and simulations or AI workflows are executed efficiently. By coordinating telemetry ingestion, model updates, and decision deployment, the management domain is what makes the NDT “alive,” transforming static models into an adaptive, self-evolving system. An important aspect of the proposed architecture is the distinction between the NDT and an NDT instance, a differentiation that is rarely emphasized in the existing literature. The NDT architecture covers the full set of components, representations, and functionalities that, in principle, could replicate the physical network in all its detail. However, maintaining such a global and continuously accurate digital replica is both costly and unnecessary for most applications. In practice, the NDT is instantiated to create an NDT instance, designed to perform a specific task for a given time duration. For example, an application may require an NDT instance dedicated to planning new gNB deployments within a certain geographic area under specific Quality of Service (QoS) constraints. In such cases, only a subset of models and functionalities is activated to serve the task. This distinction is essential: treating the NDT as a single, monolithic instance reduces its value as a flexible, general-purpose architecture, whereas recognizing instances allows the architecture to support multiple, concurrent objectives efficiently. The following subsections detail the building blocks within each domain and their role in enabling scalable, adaptive, and AI-driven NDTs. V. APPLICATION DOMAIN The unified dashboard serves as the primary interface between NDT stakeholders and the system. It provides users with the means to interact with both the NDT and, indirectly, the physical network. The range of functionalities exposed through the dashboard depends on the autonomy level of the NDT [ 37 ]. In low-autonomy settings, the NDT operates only in the digital domain, and the dashboard is limited to monitoring and analysis tasks. At higher levels of autonomy, the NDT can issue control actions to the physical network, making the dashboard a critical channel for decision-making and oversight. At its core, the dashboard supports the visualization of the current network state, enabling users to query specific parameters or network elements. More advanced features extend this role by offering predefined models and commands that can be configured or personalized. A further step is the integration of natural-language interfaces, for example, through Large Language Model (LLM)-based chatbots pre-trained on domainspecific tasks. In this case, users can describe their goals in 7 natural language, and the chatbot translates these objectives into actionable tasks within the NDT. Similar approaches have already emerged in the broader network management domain, such as the LLM connector introduced in [ 38 ], which demonstrates the potential of conversational AI to simplify network operations. To ensure reliability, such interfaces can incorporate a verification step, where the chatbot presents its interpretation and planned actions for user confirmation before execution. The dashboard also acts as the NDT output channel. For tasks requiring control over the physical network, different levels of automation are possible. In a conservative mode, the NDT provides recommended solutions, and users retain responsibility for applying them, thus ensuring compliance with organizational policies or regulatory constraints. In fully autonomous settings, the NDT may apply changes directly, while still presenting a summary of actions to the user. Hybrid approaches are also possible, where changes are executed only after explicit user approval, although this may limit applicability in real-time scenarios. Interfaces between the dashboard and the NDT are realized through APIs, which are defined and managed within the NDT MANO framework. VI. PHYSICAL DOMAIN The physical domain constitutes the tangible environment on which the NDT operates and provides feedback and control. As illustrated in Figure 3, this domain is composed of three main components. The first is the physical network, representing the real infrastructure, including its relevant entities, interfaces, and resources that the NDT is designed to replicate, monitor, and optimize. The second is the data collection framework, which enables the NDT to monitor the state of the physical network by continuously acquiring, processing, and harmonizing operational data to maintain synchronization between the physical and digital worlds. The third component is the actuation and execution framework, responsible for translating the insights, recommendations, or control policies generated by the NDT into deployable configurations or algorithms within the live network. Together, these components establish the foundation for closed-loop operation, where the NDT continuously performs a cycle of sensing (observing the physical network), thinking (analyzing and deciding within the NDT), and acting (executing optimized actions in the physical domain), followed by renewed sensing to assess outcomes and refine subsequent decisions. These components are further detailed in the following subsections. A. Physical network To construct the NDT, it is first necessary to define the network domains that the twin must represent. Since the focus of this work is on next-generation systems, where 6G is considered a natural evolution of 5G, the NDT must capture the components that remain fundamental to these networks while accommodating emerging architectural extensions. As illustrated in Figure 4, the network can be divided into several Telemetry Data Layer Telemetry Data Storage Telemetry data ingestion Telemetry Data Stream Processor Network planning, management and control User equipment Environment Radio access network Core network Telemetry Data Stream Output Act Sense Figure 3. Structure of the physical domain illustrating its components: physical network, data collection, and actuation User Equipment (UE) Radio Access Network (RAN) Core Network (CN) Data Network (DN) Transport Network (TN) Edge and Cloud Computing Figure 4. 5G and 6G physical network domains main domains that collectively form the foundation of end-toend communication systems. The RAN and the Core Network (CN) constitute the structural backbone of contemporary 5G deployments and are expected to retain this central role in 6G [ 39 ]. These domains are typically managed by Mobile Network Operators (MNOs), who oversee most of the optimization and orchestration activities. Network planning, mobility management, traffic steering, and resource allocation are primarily performed within these domains. Complementing them, the Transport Network (TN) interconnects the RAN and CN, ensuring reliable data forwarding and synchronization. Although the TN is not always the primary target of optimization, it plays a vital role in supporting end-to-end capabilities such as network slicing and latency-sensitive service delivery [40]. The edge and cloud computing domains are also essential components of next-generation architectures. As in 5G, these domains enable the processing of large volumes of data closer to where it is generated, reducing latency and improving service efficiency. Their integration is tightly coupled with QoS and Quality of Experience guarantees, which are key drivers of future intelligent communication systems. Consequently, they must be represented within the NDT to ensure accurate modeling of distributed computation and data offloading behaviors. At the final stage of the communication chain lies the Data Network (DN), which provides connectivity to external application servers and the broader Internet. Since the DN generally falls outside the administrative control of MNOs, a full-scale digital replication may not be feasible. Nonetheless, aggregated application-level statistics exposed by the CN can serve as proxies to capture the DN’s impact on end-to-end performance metrics. Finally, the UE must be included as an integral component 8 of the physical network representation. Many NDT applications aim to enhance user-centric performance, whether at the individual or group level, through adaptive resource management and context-aware optimization. The inclusion of the UE within the physical domain therefore ensures that the NDT can accurately reproduce and predict real network behavior from both the infrastructure and user perspectives. Driven by the principles of network programmability [ 41 ] and ZSM [ 42 ], the physical network adopts an AI-native architecture that enables data-driven automation and dynamic, closed-loop optimization. It exposes the telemetry data, programmability hooks, and orchestration interfaces needed for the NDT to operate as an analytical and decision-support service or to evolve into a fully controlling component in highly autonomous networks [ 43 ]. In this role, the NDT serves as a safe, isolated sandbox where decisions, configurations, and experiments can be evaluated before deployment on the live network, ensuring no impact on its real-time operation or performance. This capability supports the training and validation of AI and nonAI algorithms, while enabling the continuous refinement of decision-making mechanisms. Ultimately, it allows network controllers to dynamically adapt to changing conditions through programmatic, feedback-driven loops. B. Data collection The data collection framework is considered as one of the essential requirements to build and maintain an accurate digital replica, as specified by several standardization bodies [ 12 ], [ 44 ]. Data collection must be driven by specific operational objectives rather than performed indiscriminately, as excessive data acquisition increases resource consumption and reduces system efficiency. To address this, operators can define adaptive data collection parameters that respond to network conditions or predefined events, ensuring that only relevant and actionable data is gathered—an essential feature in large-scale networks where bandwidth and storage are limited. Our approach adopts a multi-layer design comprising a Telemetry Data Layer (TDL) in the physical domain and a Harmonization Data Layer (HDL) in the digital domain. The TDL captures and pre-processes raw telemetry streams from network interfaces. The HDL, on the other hand, aggregates, harmonizes, and structures these pre-processed datasets into standardized smart-data models 1 for use in modeling, simulation, and decision-making. This separation enables telemetry data to be leveraged independently for network performance monitoring, without invoking the full NDT processing chain. Details of the HDL and its integration into the UDR are further discussed in Section VII-A 1. The two layers are interconnected via Data Communication Buss (DCBs), which ensure reliable and adaptive data flow between domains through dynamic routing, prioritization, and integrity checking mechanisms. Focusing on the physical domain, the TDL, as shown in Figure 3, is subdivided into four key functional elements. The telemetry data ingestion collects raw telemetry data from various sources in real-time, forming the initial entry point into the system. This raw data is then passed to the 1https://smartdatamodels.org/ telemetry data stream processor, which processes and analyzes the ingested data streams, performing tasks such as filtering, aggregation, and enrichment to generate actionable insights. The processed telemetry data is then stored on the telemetry data storage, enabling historical analysis, reporting, and use in future decision-making processes. Finally, the telemetry data stream output facilitates the delivery of processed telemetry data to the next stage, which includes preparation for network planning, management, and control. The data collected by the TDL undergoes an initial phase of pre-processing, filtering, aggregation, and storage to prepare it for further analysis. This process ensures the data is consistent and usable by external applications or different layers within the NDT architecture. A benefit of this pre-processing is that it can reduce communication overhead by eliminating the need to transmit raw data. The software component responsible for data extraction, normalization, and exposure in the NDTs context is a Digital Twin Connector. This connector is a composition of protocols, communication buses, and specific instances of the TDL’s ingestion, processing, and output functions. This layer collects realtime data using various protocols, depending on the network domain. For instance, in the management plane, the collected data is used for performance metrics, logs, warnings, and state data, interfacing with network management systems and using protocols such as Simple Network Management Protocol [ 45 ], Network Configuration Protocol (NETCONF) [ 46 ], among others. In the control plane, the collected data is used to monitor the health of network control protocols, e.g., routing protocols, to aid in real-time issue detection and optimization using protocols such as sFlow [ 47 ], [ 48 ] and Border Gateway Protocol-Monitoring Protocol [ 49 ]. Finally, the data plane telemetry involves extracting and analyzing data from user packets to ensure efficient network operations, balancing collection overhead with traffic processing using protocols such as Alternate Marking Technology [ 50 ], In Situ OAM [ 51 ], or Packet Sampling [52]. In this sense, the TDL enables the “sense” phase of the Sense–Act cycle by providing continuous telemetry to the upper digital and management layers, which transform this information into actionable insights and decisions. The complementary “act” phase, where these decisions are applied back to the physical network, is described in the following subsection VI-C. C. Actuation and execution Once the NDT has been instantiated and validated within the digital domain, its outputs such as optimized configurations, predictive models, or control strategies can be deployed back into the physical network through the NDT MANO. These outputs represent the actionable intelligence derived from the NDT and are used to enhance network operation and management across domains. The process of translating NDT insights into deployable actions, however, differs depending on the characteristics of each network domain defined in Section VI-A . 6G systems are expected to integrate multi-vendor infrastructure through O-RAN and rely heavily on AI/ML-based 9 automation, achieving seamless interoperability and coordinated operation requires advanced Service Management and Orchestration (SMO) capabilities. The following subsections describe how NDT-derived actions and models are applied within each domain, including the RAN, TN, CN, and multidomain SMO, through their respective control and management mechanisms. It is important to note that while UEs are key elements of network performance optimization, they are not direct targets of actuation. Instead, the NDT actions focus on optimizing the network itself to improve the overall UE experience. 1) RAN actuation: The O-RAN architecture introduces a software-based control entity known as the RAN Intelligent Controller (RIC), which enables intelligent, flexible, and vendoragnostic management of the radio access network [ 53 ]. The RIC is responsible for monitoring, controlling, and optimizing multiple aspects of RAN operation, including radio resource allocation, interference coordination, power control, and mobility management. It provides standardized APIs that allow developers to deploy specialized intelligence modules, known as applications, that dynamically enhance RAN performance and adaptability. The RIC is composed of two hierarchical components that operate over distinct timescales: • Non-real-time RIC: Located within the SMO layer, this component operates at timescales above one second. It performs long-term network optimization, policy management, and model training. The applications developed for this layer, known as rApps, are responsible for non time-critical decision-making tasks such as configuration management, radio resource management policy generation, and performance forecasting. • Near-real-time RIC: Deployed closer to the network edge, this component handles operations with timescales between 10 ms and 1 s. It executes time-sensitive control and optimization tasks, such as handover management, scheduling, and real-time interference mitigation. The applications running on this controller, referred to as xApps, interact directly with the underlying RAN nodes via standardized interfaces to apply these actions in real time. In the context of the proposed NDT architecture, the RIC provides the interface through which decisions derived from the digital domain are translated into actionable network configurations. Once functional models within the NDT generate optimized control policies or parameter adjustments, these outputs are conveyed to the non -real-time or near-real-time RIC depending on the required response timescale. The RIC then deploys these actions via corresponding rApps or xApps, ensuring that NDT-driven intelligence is effectively realized in the live network. Through this mechanism, the NDT and RIC together establish a closed feedback loop where the RIC executes decisions generated by the NDT, while telemetry from the physical network continuously updates the digital replica, maintaining synchronization and enabling ongoing optimization. 2) TN actuation: Within the transport domain, SoftwareDefined Network (SDN) introduces a clear separation between the control and data planes, enabling flexible, programmable, and efficient management of network resources. This separation allows the network to dynamically adapt to changing service requirements, supporting on-demand configuration and reconfiguration of paths, bandwidth, and quality of service. One of the most important capabilities enabled by SDN is network slicing, which allows the creation of multiple logically isolated virtual networks over the same physical infrastructure. Each slice can be tailored to meet the performance and reliability needs of specific service types such as Enhanced Mobile Broadband, Ultra-Reliable Low-Latency Communications, and massive Machine-Type Communications, thereby ensuring service differentiation and isolation. To support this capability, the ETSI has defined a comprehensive slicing management framework composed of four key functions that coordinate the orchestration and lifecycle of slices across domains [54]: • Communication Service Management Function: Converts high-level service requirements provided by tenants or applications into slice specifications that can be interpreted by lower-level management entities. • Network Slice Management Function (NSMF): Decomposes the overall slice request into sub-slices corresponding to different domains (e.g., RAN, TN, CN) and manages end-to-end slice orchestration. • Network Slice Subnet Management Function (NSSMF): Oversees the instantiation, configuration, and optimization of slice subnets within a specific domain, ensuring that allocated resources meet the service-level objectives. • Network Function Management Function: Handles the lifecycle operations of individual network functions, including deployment, scaling, monitoring, and termination. These components collectively enable coordinated, end-toend orchestration across the RAN, TN, and CN, ensuring consistency and performance isolation between slices. In the context of the proposed NDT architecture, the NDT augments this framework by introducing predictive and adaptive control over slicing and transport management. The NDT continuously analyzes telemetry from the TN to anticipate congestion, optimize routing, and predict resource bottlenecks. It can then generate optimized configurations or reallocation strategies, which are translated into actionable commands and deployed through the SDN controller. The outputs of functional models, such as traffic forecasting or anomaly detection, can be directly mapped to slice adjustment operations via the NSMF or NSSMF interfaces. Through this integration, the NDT transforms the TN from a statically configured backbone into an intelligent, selfoptimizing substrate, closing the loop between observation, prediction, and actuation across domains. 3) CN actuation: Within the CN, the orchestration and management of Virtual Network Functions (VNFs) represent essential processes that ensure flexibility, scalability, and resilience in modern cloud-native infrastructures. The Network Function Virtualization (NFV)-MANO framework, standardized by the ETSI, defines the architecture for deploying, scaling, and maintaining VNFs across distributed cloud environments [ 55 ]. The orchestration task aims to minimize operational expenditure and resource fragmentation while meeting service-level 16 Table III 6G NDT FUNCTIONAL MODELS TAXONOMY. Functionality Specific Focus / Problem to Solve Model Type Operation Mode Computing Element Network Domain Reference Network Management and Control VNF MANO Analytical Reactive Edge, Cloud Core [89] VNF MANO Deterministic Hybrid Cloud Core [90] VNF MANO AI-based Proactive Edge, Cloud Core [ 57 ], [ 91 ], [ 92 ], [40], [93] VNF MANO Hybrid Proactive Edge, Cloud Core [94] Federated VNF MANO AI-based Proactive Edge, Cloud Multidomain [95] Federated VNF MANO Deterministic Reactive Edge, Cloud Multidomain [96], [97] Beam management Analytical Proactive Edge RAN [98], [99] Beam management AI-based Proactive Edge RAN [ 100 ], [ 101 ], [102] BS sleep management AI-based Proactive Edge, Cloud RAN [103] Frequency (RAT) mapping / Positioning AI-based Proactive Edge RAN [47] Traffic steering / Near-real time RIC AI-based Reactive Edge, Cloud RAN [104] Network Planning VNF/CNF / SFC Placement Hybrid / Deterministic / Analytical Reactive Edge, Cloud RAN, Core, Transport, Multidomain [ 105 ], [ 106 ], [107], [108] VNF/CNF / SFC Placement AI-based Proactive Edge, Cloud – [ 109 ], [ 110 ], [111], [112] SFC composition Analytical Reactive Edge, Cloud RAN, Core, Transport, Multidomain [113] SFC / Service Composition AI-based Proactive Cloud RAN, Core, Transport [114], [115] Slice Design / Creation AI-based Proactive Cloud RAN, Core [116], [117] Slice Design / Creation AI-based Proactive Edge, Cloud RAN, Core, Transport [118] PCI planning Analytical Proactive Edge, Cloud RAN [119] Slice monitoring / Visualization Analytical / Deterministic Proactive / Reactive Edge, Cloud RAN, Core [120] Slice monitoring / Visualization AI-based / Hybrid Proactive Edge, Cloud RAN, Core [121] Network Diagnosis and Security Anomaly / Intrusion Detection / Attack Prevention AI-based, Hybrid Proactive, Reactive Cloud Core [122] Anomaly / Intrusion Detection / Attack Prevention AI-based / Analytical Proactive Edge, Cloud, Federated RAN, Core [123] Anomaly / Intrusion Detection / Attack Prevention AI-based Proactive Cloud Core [124] Anomaly / Intrusion Detection / Attack Prevention AI-based Proactive Edge, Cloud RAN, Core, Transport [118] Encryption / Authentication Analytical Reactive Cloud Core / RAN [125] Encryption / Authentication AI-based Proactive, Reactive Cloud Core [126] cation follows the functional model types and characterization criteria defined earlier in this work. Functional models are first grouped according to their primary functionality, which determines the overarching category of each model. The second column specifies the intended purpose or problem addressed (e.g., RIC optimization, VNF management and orchestration, network slicing, scheduling, or multidomain coordination). The subsequent columns describe the model type, operational mode, computing element, and corresponding deployment domain. Finally, the last column lists representative studies from the literature associated with each category. This table is not intended to be exhaustive but rather to provide a structured overview of illustrative examples that capture key research directions and integration possibilities within the NDT framework. 4) Unified data model repository: The UDM serves as a central repository for storing metadata of functional models that have been trained, calibrated, or parameterized using an NDT instance. This includes AI-based models as well as analytical and probabilistic models that have been adjusted for specific scenarios. Each functional model in the repository can have multiple versions, corresponding to diverse operational conditions such as indoor or outdoor environments, urban or rural areas, specific geographic zones, or distinct QoS requirements. These versioned models are systematically labeled following the functional model taxonomy introduced in Section VII-A 3b, and are linked to the corresponding basic models used during their training when relevant. Once a model has been validated and tested within the NDT environment, it can either be deployed directly to the physical network or stored for future evaluation and deployment. The model management component within the MANO framework (cf. Section VIII-A ) is responsible for maintaining this repository. It oversees the versioning, labeling, and application procedures of functional models, ensuring that the correct model variant is selected for a given scenario and that the NDT remains synchronized with the deployed network. This unified repository enables systematic model reuse, facilitates rapid deployment, and supports consistent performance across 17 heterogeneous network environments. B. Simulation environment Simulation is a core feature of NDTs, providing the ability to study what-if scenarios, i.e., to analyze the consequences of network configuration changes before applying them to the physical infrastructure. This capability relies on the availability of an accurate and reliable virtual representation of the real network, capturing both its structural and behavioral characteristics. Simulators are complementary to the graph-based basic models introduced in Section VII-A 2. While graph-based models define the static topology, components, and relationships of the network, simulators endow these representations with dynamics, enabling the generation of realistic network behaviors that closely mirror those of the physical system. The simulation environment thus serves as an experimental ground for testing, evaluating, comparing, and training functional models under controlled and reproducible conditions. In this context, the complete AI training pipeline of functional models (further detailed in Section VII-C ) can integrate simulators as part of the real-time or online training process. Given the scarcity and sensitivity of real operational data, simulators also serve as a critical source of synthetic data, supporting the expansion of training datasets and the exploration of edge-case scenarios that may be rare or infeasible to reproduce in real networks. As 6G networks are expected to exceed the complexity and heterogeneity of current systems, no single existing simulator can encompass all the features, protocols, and technologies required to address the challenges of future networks. To overcome this limitation, our approach adopts a federated discrete-event simulation framework, designed to interconnect and orchestrate multiple domain-specific simulators. This federated approach allows for modularity and extensibility, enabling the coupling of simulators that model different network segments (e.g., RAN, TN, CN), radio environments, or service types, while ensuring synchronization and consistency across the simulation instances. The interaction and coordination of the simulation framework are managed by the NDT management layer (detailed in Section VIII). This layer retrieves the required models, configurations, and parameters from the UDM and the AI training module, instantiates and supervises the execution of the simulation instances, and collects performance and behavioral metrics from the simulation outcomes. The resulting insights are then reintegrated into the AI training process for continuous model refinement and optimization. Finally, the optimized parameters and validated functional models can be leveraged by the NDT management layer to update or reconfigure the physical network, thus closing the loop between simulation, prediction, and real-world operation. 1) Simulation ecosystem: The simulation ecosystem includes a diverse range of tools and frameworks that support the development, testing, and optimization of complex systems across multiple domains. Each simulator type provides distinct capabilities, enabling accurate modeling, real-time experimentation, and validation of network and physical processes. Collectively, they form the foundation of the NDT’s ability to emulate realistic behaviors and evaluate system performance prior to deployment. The following sections outline the main categories of simulators relevant to NDTs. a) Real-time simulation: Real-time simulation executes models at the pace of actual time, providing immediate feedback for testing, control validation, and operator training. It is widely applied in Hardwareand Software-in-the-Loop environments to validate interactions between real and simulated components before deployment. Some frameworks also operate faster than real time, enabling accelerated scenario exploration and longterm performance evaluation. Simulator coupling is an effective way to achieve real-time capabilities. For instance, Veins [ 127 ] integrates OMNeT++ and SUMO for bidirectional interaction between mobility and communication layers, enabling realistic assessments of vehicular networks. Similarly, [ 128 ] extends this approach by federating additional simulators for heterogeneous real-time testing of connected vehicle scenarios. Several studies have explored the integration of different simulators to achieve real-time or near real-time simulation performance across vehicular communication systems. In [ 127 ], the authors present Veins, the first bidirectional coupling between the network simulator OMNeT++ and the traffic simulator SUMO. This integration enables real-time interaction between vehicular mobility and communication processes, allowing the study of Vehicular Ad hoc Network (VANET) protocols under realistic traffic conditions. The framework supports dynamic scenario adjustments, facilitating comprehensive evaluations of network performance in response to changing vehicular behaviours. In [ 128 ], the authors integrate Hardware-in-the-Loop, Automotive Simulation Models, Ego Vehicle Interface, and VANET Veins and SUMO. This work demonstrates how a Discrete-Event Simulator (DES) can be leveraged to simulate real-time systems, effectively enabling heterogeneous federation of real-time components. Similarly, the study in [ 129 ] combines a VANET simulator with a 3D driving simulator to streamline real-time Advanced Driver Assistance System (ADAS) simulation. This setup supports large-scale, city-level testing of VANET applications with real user interaction, highlighting another example of heterogeneous federation in real-time simulation environments. In [ 130 ], the authors identify the challenges of using DES for real-time system simulation and propose an optimization method focusing only on simulating the Region of Interest to enhance real-time performance. Complementing this, the work in [ 131 ] demonstrates that asynchronous parallelism – executing isolated computations in the background using multicore architectures – can accelerate DES to operate in real time. Finally, [132] introduces a framework that merges a network simulator, based on an analytical model, with the INTEGRATION traffic simulator. The coupling operates at variable intervals, allowing dynamic two-way communication and significantly improving scalability. This design enables the system to run faster than real time while maintaining synchronization accuracy, representing an integrated simulation framework with real-time support. 18 b) Simulation with High Level Architecture (HLA): High Level Architecture (HLA) provides a standard framework for achieving interoperability and synchronization among distributed simulators. It enables heterogeneous simulation systems to communicate and operate collectively through standardized data exchange and time management mechanisms. HLA-based and HLA-inspired co-simulation environments are particularly valuable for integrating network, mobility, and physical-layer simulators within a cohesive ecosystem. This interoperability allows researchers to conduct holistic, cross-domain experiments that maintain temporal and semantic consistency across simulators. Several works have explored heterogeneous federations of simulators to enhance the realism and interoperability of vehicular communication and mobility simulations. In [ 133 ], the authors present the VSimRTI framework, inspired by the HLA, which integrates SUMO for traffic simulation, JiST/SWANS for network modeling, and eWorld for environmental context. This combination enables a cohesive environment for simulating realistic V2X communication scenarios with strong coupling between mobility, network, and environmental dynamics. Building upon this, [ 134 ] enhances VSimRTI by coupling it with OMNeT++ and CCMSim to create a detailed co-simulation framework. Within this setup, OMNeT++ handles higher communication layers, while CCMSim models the physical and radio channel layers, allowing for comprehensive testing and evaluation of network protocols and signal transmission under diverse conditions. Further extending the integration capabilities of VSimRTI, [ 135 ] employs SUMO for traffic modeling, OMNeT++ for cellular network simulation, and VSimRTI_App for applicationlevel interactions. This trace-driven simulation approach enables emulation of real-world cellular communication for V2X applications, bridging mobility, networking, and application perspectives. Similarly, the work in [ 136 ] integrates Carla, MOSAIC, and SUMO to form a flexible co-simulation environment for autonomous and cooperative driving systems. This extensible design supports both individual autonomous vehicles testing and collective behavior studies, contributing to advancements in cooperative driving automation research. Branching out, [ 137 ] combines MATSim for traffic scenario calibration, MOSAIC for co-simulation management, and SUMO for simulation execution. This multi-federate framework demonstrates an end-to-end process for evaluating smart mobility applications at large scale, underlining the growing importance of heterogeneous federation approaches in realistic and scalable simulation ecosystems. Finally, more flexible approaches to HLA-inspired coupling based on Veins and SUMO have recently started to strive towards improving the scalability of this approach while retaining modularity for domain-specific simulators like road vehicle, Unmanned Aerial Vehicle (UAV), and Low Earth Orbit satellite network cosimulation [138]. c) Domain-specific simulation: Domain-specific simulators are crucial in studying complex systems such as UAVs, robotic systems, and vehicular networks, providing realistic virtual environments for testing, training, and optimization. For UAVs, frameworks such as AirMobiSim [ 139 ], SkyRoute [ 140 ], and UTSim [ 141 ] integrate aerial vehicle dynamics, communication systems, and traffic interactions, enabling flexible simulation of large numbers of UAVs and their interactions with ground vehicles in smart city environments. These tools support the evaluation of cellular connectivity, sense-andavoid mechanisms, communication protocols, navigation, and UAV-aided networks for applications ranging from emergency networks to energy-efficient aerial communications [ 142 ], [143], [144]. Robotic simulators, including RoboNetSim [ 145 ] and USARSim [ 146 ], combine robotics engines with network simulators to study multi-robot coordination, sensor-actuator behavior, and autonomous vehicle dynamics, providing an immersive environment for testing collaborative tasks, autonomous driving, and ADAS systems [147]. Vehicular and V2X simulation frameworks extensively couple traffic, network, and control components to enable realistic evaluation of connected and automated vehicle systems. Cosimulation platforms integrate traffic simulators such as SUMO or VISSIM with network simulators like OMNeT++, NS3, MATLAB, iCS, or CarMaker [ 127 ], [ 148 ], [ 149 ], [ 150 ], [ 151 ], [ 152 ], [ 153 ], supporting development and testing of vehicle platoons, autonomous driving algorithms, ADAS, energyefficient vehicles, and cooperative driving systems [ 154 ], [ 155 ]. Cellular and C-V2X communications are evaluated using frameworks such as Veins LTE [ 156 ], Artery-C [ 157 ], OpenCV2X [ 158 ], and the Vienna 5G System Level Simulator [ 159 ], which allow large-scale simulation of vehicular networks with realistic traffic modeling, network dynamics, and 5G-enabled communication features. Satellite connectivity is evaluated using frameworks such as space_Veins [ 160 ], which provides satellite movement based on Two-Line Element (TLE) data. Collectively, these domain-specific simulators provide modular, extensible, and heterogeneous environments, enabling detailed and scalable study of aerial, robotic, and vehicular systems and their interactions in complex operational scenarios. 2) Simulation tools: Simulation tools are essential for network planning and optimization in both industrial and research contexts. Industrial tools, typically licensed, address real-world deployment challenges such as coverage analysis, capacity planning, and KPI support, ensuring efficient network design and operation. Research-focused tools, often opensource, provide flexible environments for detailed modeling, protocol exploration, and performance evaluation, enabling innovation in telecommunications technologies [161], [162]. a) Industrial-focused tools: Industrial tools include comprehensive 5G planners such as the ASSET suite [ 161 ], which supports 5G New Radio (NR) modeling with advanced propagation models, complex antenna arrays, 3D coverage and capacity simulations, and human exposure analysis. The Atoll 5G NR module [ 162 ] and Capgemini’s 5G planning solution [ 163 ] provide multi-RAN modelling, ray tracing for mmWave networks, and KPI-driven continuous optimization. Cell Designer [ 164 ] and Hamina [ 165 ] offer coverage analysis for 5G and Wi-Fi, including indoor/outdoor scenarios and cloud-based planning. Hardware vendors like Huawei provide precision propagation, automatic site planning, and novel service simulations [ 166 ]. Tools like iBwave [ 167 ] 19 and LSTelecom [ 168 ] focus on indoor network planning, small cells, interference assessment, and capacity optimization. NetSim [ 169 ] simulates end-to-end 5G NR networks, including aerial networks, while the Terragraph project [ 170 ], [ 171 ] provides open-source planning for 60 GHz mesh networks with SDN/SDR capabilities. b) Research-focused tools: Research-focused simulators enable protocol-level, system-level, and network performance studies. The 5G Air simulator [ 172 ] and 5G K-Simsys [ 173 ] model key 5G NR features, including MaMIMO, beamforming, Narrowband-IoT, and mmWave scenarios. CGA Simulation [ 174 ] uses digital twin representations for urban 5G planning and packet error analysis. Discrete-event simulators like OMNeT++ [ 175 ], [ 176 ], NS-3 [ 177 ], and OPNET [ 178 ], with extensions such as 4GSim, 5GSim, Simu5G, and 5GLENA, provide detailed stack-level simulations including RRA, V2X, and multi-tier heterogeneous networks. OpenAirInterface [ 179 ] integrates RAN and Core network for 3GPPcompliant simulations, while the Vienna 5G Simulator [ 159 ], [ 180 ] supports large-scale, system-level simulations with 3D channel models, MaMIMO, and mmWave abstraction for thousands of nodes. 3) Interaction between simulation and functional models: The integration between the simulation framework and the NDT components is established through a structured data translation process. Specifically, the data contained in the basic models are converted into simulator-readable formats to build the virtual environment in which the simulation is executed. This translation ensures that the virtual network accurately reflects the structure and parameters of the real system, enabling consistent testing and validation of models. It is, however, essential to distinguish between simulation models and functional models, as they are closely related but conceptually distinct. Both play a key role in enabling the operation and intelligence of the NDT, yet they differ in scope and level of abstraction. A simulation model refers to a machine-readable description of a specific scenario that can be interpreted and executed by a simulator. The simulator itself is the software responsible for loading and running this model, generating results that describe the system’s behavior under the given conditions. The term simulation therefore denotes the actual execution of the simulation model within the simulator, producing synthetic data or performance indicators that reflect the system’s response. In contrast, a functional model is designed to analyze, predict, or optimize network behavior. It typically combines an analytic, stochastic, or AI-based component with an optimization or training algorithm, while also embedding a simulation model that defines the operational scenario. In other words, the simulation model provides the environment and dynamic context in which the functional model operates. The functional model extends beyond mere simulation by incorporating intelligence and decision-making capabilities that enable continuous adaptation, learning, and performance optimization. The interaction between both functional and simulation models forms the foundation of closed-loop experimentation and model refinement in the NDT. The process of training functional models through simulation will be detailed in Section VII-C , while the execution and federation of simulations for model evaluation will be further discussed in Section VIII-B. C. AI training/optimizer The AI training and optimization component within the NDT architecture is responsible for preparing functional models for deployment, refined for the application and scenario requirements. This module enables models to learn from data, predict network behavior, adapt to changing conditions, and continuously improve their performance through iterative training and validation cycles[181], [182]. While all functional models are subject to optimization, their training requirements differ depending on their nature. Analytical and deterministic models are expressed through explicit mathematical formulations or algorithmic rules and therefore do not require a dedicated training pipeline [ 183 ]. Once parameterized with suitable static values, these models are directly stored in the UDM and can be invoked by the NDT management layer when needed. In contrast, AI-based and probabilistic models require a structured training process. This involves computing model parameters, such as neural network weights or probability distributions, based on available data [ 86 ]. The following subsections outline the training pipelines applied to each class of models. 1) Deep learning and probabilistic model training: For DL and probabilistic models, the training process begins with the preparation of datasets that define the relationship between input and output variables [ 184 ]. These datasets may originate from the physical network or be synthetically generated through the simulation framework. In both cases, data preprocessing is critical in ensuring convergence and accuracy. This includes data cleaning to handle missing or outlier values, and feature selection or dimensionality reduction techniques such as ANOVA, Lasso, or principal component analysis to retain only the most informative features. Once preprocessed, the data are fed into the learning stage, where model parameters are iteratively optimized according to a defined loss function, e.g., mean square error for regression, cross-entropy for classification [ 185 ]. Gradient-based optimization adjusts the model weights until a target accuracy or convergence criterion is achieved. The trained model, along with its metadata, e.g., training conditions, performance metrics, version number, is then stored in the UDM to ensure traceability and future reuse. Functional models can be retrained when new data becomes available or as part of continuous learning procedures. In the case of probabilistic models, the pipeline follows a similar structure. After preprocessing, the algorithm learns the underlying probability distributions from the data using techniques such as Bayesian Networks, MDPs, or Monte Carlo simulations. The resulting probability functions and their associated metadata are also stored in the UDM for future inference or integration into hybrid models. 2) Reinforcement learning training: Reinforcement Learning (RL) follows a distinct paradigm in which the model, or 20 agent, learns optimal decision-making strategies by interacting with an environment [ 186 ]. Unlike DL or probabilistic models, RL does not rely on a pre-existing dataset; instead, it generates training data dynamically through exploration and exploitation. The learning process is typically modeled as an MDP defined by a state space (environmental conditions), an action space (possible interventions), and a reward function (feedback signal after each action) [187]. For small state-action spaces, tabular Q-learning methods can store explicit values for each possible pair. However, as complexity increases, RL benefits from neural network approximations of the value function, resulting in DRL [ 188 ]. During training, the agent iteratively explores its environment, balancing exploration of new strategies with exploitation of known ones using techniques such as ϵ -greedy policies or Boltzmann exploration. The training can occur in real-time within an NDT instance or be conducted offline using simulated environments built from the simulation framework [189] Once training converges, the learned policy or Q-function weights are stored in the UDM repository. These trained models can subsequently be deployed to monitor live network states and autonomously execute control actions based on learned strategies. 3) Unified training and optimization framework: The proposed architecture in this work provides a unified framework for the training and optimization of all functional model types: AI-based, probabilistic, analytical, deterministic, and hybrid. While AI-based models rely on supervised, unsupervised, or reinforcement learning paradigms, other models undergo calibration and parameter fitting to align theoretical assumptions with empirical observations. Hybrid models combine these approaches, requiring joint optimization procedures that bridge analytical and data-driven methods. Figure 7 illustrates the pipelines for DL and RL training. The outcome of each training or optimization process, whether it is a set of trained AI weights, a probabilistic distribution, or optimized analytical parameters, is systematically stored in the UDM. Each version is labeled with metadata corresponding to the model taxonomy defined in Section VII-A 3, including its functionality, type, operational mode, and network domain. This structured metadata enables efficient orchestration, reuse, and adaptation of models within the NDT management layer. In summary, the AI training and optimization pipeline ensures that every functional model within the NDT is continuously improved and ready for deployment. The interaction between the training component, simulation framework, and management layer creates a closed learning loop in which models evolve alongside the network itself. The detailed mechanisms of this interaction are further elaborated in Section VIII-B. VIII. MANAGEMENT DOMAIN The efficient operation of NDTs requires a comprehensive management framework capable of coordinating activities across multiple domains, including the physical network, simulation environments, data repositories, and various network models. To achieve this, the proposed architecture introduces Figure 7. Functional models training pipeline a unifying vertical layer, referred to as the NDT MANO framework, illustrated in Figure 2. The NDT MANO framework acts as the central control entity that governs the lifecycle and interaction of all NDT components. It ensures synchronization between the physical and digital domains, supervises the instantiation and execution of NDT instances, and coordinates data collection, simulation workflows, and AI-driven model training. By providing crossdomain orchestration, it enables adaptive management of network resources and continuous optimization of network performance in response to changing operational conditions. The NDT MANO consists of several functional subcomponents: NDT management, simulation management, AI workflow management, data management, federated management, and physical network MANO. Each subcomponent plays a distinct role in maintaining consistency, automation, and scalability within the NDT ecosystem. The following subsections introduce these subcomponents in detail, outlining their main responsibilities and interrelations within the overall architecture. A. NDT management ETSI, within the ZSM suggests that an NDT should be use case specific [ 15 ], where specific models should be created to deliver the expected NDT capabilities. We call this specific combination of basic and functional models a running NDT instance (cf. Figure 2). The NDT instance relies on several key components to function correctly, as illustrated in the figure. These include an NDT Orchestrator, which coordinates all internal processes related to the creation and execution of an NDT instance; a Model Manager, which governs the interaction between basic and functional models and ensures their performance, triggering corrective actions when necessary; a Resource Manager, responsible for overseeing the computing resources used by the NDT instance; and an NDT Monitor, which continuously tracks the status and behavior of the NDT, issuing alerts in case of anomalies. Additionally, an NDT Translation Layer is required to interpret human inputs for the NDT and convert its outputs into human-readable information. The NDT Management block, within the Management domain, is a higher-level management responsible for the operation, coordination, and the LCM of NDT instances and their subcomponents [ 190 ], [ 191 ]. This block serves as an 21 intelligence layer, synthesizing consumer requirements, network context, and available resources. Among its key responsibilities are (i) overseeing the instantiation, configuration, and orchestration of NDT components based on application and operational requirements; (ii) monitoring the resource utilization and ensuring optimal performance of twin instance; (iii) managing the lifecycle of NDTs, including creation, adaptation, and termination, aligned with the evolution of network services and topologies. Furthermore, the NDT MANO layer governs the integration and operation of all components within the NDT architecture, ensuring synchronized interactions between the physical network and the NDT elements, including data, models, simulation framework, and AI training mechanisms. The creation and operation of an NDT involve several stages [ 192 ], [ 193 ], [ 194 ], [ 29 ]; thus, the NDT management block is seen as the mastermind within the NDT management domain. It begins with requirements gathering and planning, where the NDT scope is defined and the relevant assets and systems to be modeled are identified. This is followed by data collection, which involves collecting and integrating information from diverse sources to build a reliable foundation for the twin. Once the data is in place, a virtual representation is created using appropriate modeling techniques tailored to the system’s complexity. Once the virtual entities composing the NDT are created, the next step involves federation, integration, and ensuring interoperability by connecting the NDT with existing systems and adhering to established industry standards and communication protocols. The deployment phase then places the NDT in the desired environment, whether in the cloud, on-premises, or at the edge, while establishing seamless data exchange with its physical counterpart. The twin is then used for simulation and analysis, enabling the prediction of potential issues, performance optimization, and decision support. Ongoing monitoring ensures that the twin stays accurate by continuously updating it with real-time data from the physical system. This continuous refinement is key to maintaining model fidelity. In parallel, collaborative and visualization tools allow users to interact with the NDT, share insights, and make informed decisions. Furthermore, the DT facilitates network optimization by enabling scenario exploration and hypothesis testing, improving operational efficiency, and reducing costs. Finally, maintenance and LCM ensure the NDT remains relevant over time through regular updates, enhancements, and alignment with evolving system conditions. Section IX provides a more detailed examination of the above-mentioned steps. B. Simulation management Within the functional architecture, simulations primarily support what-if analyses—examining how network behavior changes under different parameters, without the need of having to deploy them in real-life. This process involves creating simulation models, parameterizing them, executing runs, and delivering results to other components, such as the AI training block. The simulation framework coordinates these processes, synchronizing multiple simulators and facilitating data exchange during runtime. Dedicated Scenario Manager 1 Dedicated Scenario Manager 2 Dedicated Scenario Manager 3 Master Scenario Manager Co-Sim Server Co-Sim Framework Protocol Translator 1 OMNeT++ Network Simulation Co-Sim Federation Protocol 2 Protocol Translator 2 AirMobiSim UAV Simulation Co-Sim Federation Protocol 2 Protocol Translator 3 SUMO Traffic Simulation Co-Sim Federation Protocol 2 NDT MANO Launch Controller Client Co-Sim Launch Protocol 1 Figure 8. Co-simulation framework architecture. 1) Simulation creation, evaluation, and exploitation: Creating a simulation model relies on both basic and functional models, often using subsets of the full network for feasibility, in case the full network cannot be simulated. Information about basic and functional models might be needed when configuring simulators. For example, the network properties of a basic model are used while configuring an OMNeT++ simulation while UEs mobility traces might be used for configuring SUMO. In AI workflows (cf. Section VIII-C ), simulations are frequently reused and adapted across training loops. Once executed, simulation results are collected and passed to relevant entities, such as the AI module, dashboard, or the NDT MANO, for evaluation or real-world actuation. 2) Simulation framework: A co-simulation framework is critical for analyzing complex, multi-domain network behavior within the NDT environment. It enables synthetic data generation, scenario testing, and what-if analyses to assess network performance under dynamic conditions. To support these capabilities, the simulation framework orchestrates the lifecycle of simulators, or federates, that together form a cohesive simulation environment. As illustrated in Figure 8, it adopts a federated architecture [ 138 ] where multiple specialized simulators, or federates, represent distinct domains (e.g., road traffic, UAV traffic, or communication). Federates interact through the co-simulation framework, ensuring synchronized and cooperative execution. The framework adopts a modular, technology-agnostic design that allows substituting individual simulators such as OMNeT++ (network) and SUMO (road traffic) with alternative tools, supporting extensibility and interoperability across diverse experimental setups. Such composability has proven essential for accurately capturing effects across multiple domains in closed-loop optimization [195]. For conducting simulation studies, the NDT communicates with the co-simulation framework using a standardized google Remote Procedure Call (gRPC)-based interface and a protocol referred to as Co-Sim Launch Protocol 1 (Protocol-1), which 22 NDT MANO UAV Simulation Framework Network Simulation Protocol-1 Protocol-2 Protocol-2 Protocol-2 Protocol-2 Protocol-2 Protocol-2 Protocol-1 Figure 9. Co-simulation framework interaction. manages simulation control, configuration, and result exchange. Via this protocol, it interacts with a dedicated component in the framework, the co-sim server that sets up simulation runs. Running simulations are driven by a master scenario manager which interacts with dedicated scenario managers, one per specialized simulator, to advance time and exchange information between simulators. In order to keep the architecture modular, these share a common API, which also uses gRPC and a shared protocol, Co-Sim Federation Protocol 2 (Protocol-2), allowing seamless data exchange and synchronization across distributed environments. This is made possible by pairing each specialized simulator with a simulator-specific protocol translator. Together, Protocol-1 and Protocol-2 establish a unified communication structure that supports scalable, highfidelity simulations across heterogeneous domains. Protocol-2 provides a set of services that must be implemented by the protocol translator of a specific simulator to successfully couple with the simulation framework and participate in the federation. All services offered by the CoSim frameworks are described below: Start and Finish services start and end execution of a corresponding simulator. The ExecuteOneTimeStep service advances the local time of a simulator by the update interval, which is pre-configured in the configuration file of the Co-Sim framework. The framework also provides a QueryRequest service, which is used by the Co-Sim framework to proactively query a particular simulator to submit any data exchange request after each ExecuteOneTimeStep service is called. These requests usually contain a data exchange request from another simulator or a command to influence their behavior. On the other hand, the GetManagedHosts service is used by the Co-Sim framework for synchronization purposes. After all federates are advanced by a fixed update interval, this service keeps all the federates in sync by exchanging the current internal state of hosts managed by each simulator. Furthermore, the framework also provides the SetAttribute and GetAttribute services to process the requests received from the QueryRequest service. Furthermore, two more services, InsertHost and DeleteHost, are present to dynamically control the number of simulation entities being simulated by a simulator. Upon completion, simulation outputs are transmitted back to the NDT through Protocol-1 for further use in AI-based optimization or visualization (Figure 9). A persistent session between the NDT and the framework allows iterative experimentation, scenario refinement, and adaptive model calibration without reinitialization. C. AI workflow management Within the proposed architecture, the AI workflow management component is vital, which enables data-driven NDT operations, and the integration between the NDT and the AI block adapts three training processes to the characteristics of the underlying models: DL: for tasks such as traffic classification or radio prediction. Probabilistic models: like Gaussian Process Regression for spatial interpolation, and RL, including DRL, for closed-loop control scenarios like resource allocation or handover optimization. These pipelines communicate with NDT via a unified data infrastructure where telemetry data is first ingested, harmonized, and stored in the UDR using standard formats, for example, smart data models aligned with 3GPP/ORAN. The AI pipeline uses this data, whether real-time or historical, to train, validate, and deploy models. For RL-based functions, the AI block dynamically interacts with the simulation framework (e.g., OMNeT++, ns-3) to create synthetic environments for trial-and-error learning, allowing for safe policy training prior to deployment in live networks. These workflows are automated and scaled using modern MLOps technologies. For example, the project demonstrates the use of Perfect as an orchestration engine to manage training pipelines in Kubernetes environments, ensuring that they are versioned, reusable, and resource-efficient. Model metadata, such as performance metrics, training conditions, and version IDs, as well as weights, are stored in the UDM to simplify the process of updating and tracing models. The future of 6G networks will be significantly influenced by the implementation of AI workflow management. Initially, it facilitates real-time adaptation at a large scale. Secondly, the architecture facilitates distributed and federated learning scenarios by decoupling data ingestion from model training and inference. Lastly, the modular design enables the system to manage the sheer volume and velocity of 6G telemetry (e.g., sub-second KPIs from gNBs, UEs, and core functions) while preserving low-latency decision-making, which is crucial for ultra-reliable, latency-sensitive applications such as industrial automation or teleoperated driving. One of the abilities of the NDT is to create productionready models that are later deployed in the physical network as controllers and orchestrators, or eventually be used as functional models within the same NDT operation. To achieve this ability, effective workflows for AI LCM are needed and must be aligned with MLOps frameworks, to automate and operationalize ML processes to ensure the delivery of production-ready software [196]. The AI Workflow Management component manages the full AI/ML lifecycle, including data preprocessing and versioning, 23 training, validation, deployment, and inference. It coordinates interactions between the physical network and the NDT elements, including data, models, the simulation framework, and AI training mechanisms. The MLOps framework includes Model Assessment, which monitors the accuracy and consistency of active AI/ML models surveyed by the NDT, automatically triggering pipelines to retrain or switch models based on inference quality values. The workflow should generally be modeland platformagnostic, supporting both supervised learning and RL paradigms. Future AI-native orchestration systems must enable autonomous workflow management, allowing model selection and validation beyond simple accuracy metrics and evaluating models based on network-level performance outcomes and task-specific objectives. D. Data management The evolution of 6G networks introduces significant complexity and diversity, making efficient and secure data management crucial. In the NDT-enabled 6G architectures, effective data management supports real-time decision-making, optimization, and orchestration, addressing scalability, reliability, and security challenges within the NDT MANO layer. Data management ensures data integrity, availability, and usability across systems, especially in dispersed cloud-based environments. The data management provides the real-time network data feeds to the basic and functional models, enabling a continuous refinement of models, simulations, and AI-driven functionalities. Protecting sensitive data involves controlling access through role-based access control and multi-factor authentication, while preserving data integrity. Additionally, privacy regulations must be considered, which require data masking or anonymization operations. The vulnerabilities that cause malicious exploitation are increasing as the system expands. Therefore, it is crucial to have key requirements, including maintaining data consistency across nodes, preventing the sharing of sensitive data, and providing secure data management throughout the system. Our architecture proposes an integrated data management system that extends a data exposure and collection architecture, where data is moved from the physical world to the digital and human world via diverse interfaces, as explained in Section VI. Based on the IETF RFC9232 [ 197 ], our approach includes the generation,collection,processing, and consumption data modules, mapped in the architecture via the telemetry data layer. These modules support data MANO services, such as security, scalability, and federated distributed parallelization. Additionally, the HDL acts as a bridge between the telemetry layer and the NDT instance. It ensures that the raw telemetry data is transformed into standardized formats compatible with structured smart data models, which serve as the foundation for defining and operating the entire architecture. E. Federated management Modern and future communication networks are increasingly composed of multiple autonomous systems, each governed by distinct administrative domains that independently define and enforce their operational policies. This decentralization is driven by diverse business objectives, technological capabilities, and governance models, leading to highly heterogeneous network environments [ 198 ]. While such autonomy promotes flexibility and innovation, it also introduces substantial challenges for achieving end-to-end management, coordination, and optimization across domains, particularly when supporting automated network control. Federation in communication systems can be analyzed from both the physical and digital perspectives. From the physical network viewpoint, ensuring end-to-end service stability and reliability requires decomposing a given service into resource slices that span multiple administrative domains. To enable this, recent 3GPP releases have introduced hierarchical architectures supporting cross-domain orchestration. For instance, 3GPP Release 16 [ 61 ] defines three fundamental management entities. This hierarchical model enables scalable, modular orchestration and aligns with other emerging frameworks such as ITU-T Y.3061 and ETSI ZSM, which also advocate federated, multidomain service management [43], [199]. While hierarchical orchestration provides a structured foundation for multi-domain coordination, the integration of AI-based network functions and services introduces new challenges that go beyond the assumptions of traditional static models. AIenabled functions often exhibit context awareness, adaptivity, and continuous learning, characteristics that may conflict with the deterministic and pre-defined workflows of current management systems. Achieving consistent performance and service quality requires negotiating orchestration decisions across domains with varying trust levels, policy frameworks, and infrastructure constraints. Moreover, AI-driven functions often depend on real-time data exchange, model retraining, and crosslayer feedback loops, demanding more flexible and responsive orchestration mechanisms. These requirements challenge the predictability and determinism of conventional frameworks such as those defined in 3GPP standards, necessitating new forms of coordination that blend automation with autonomy. To address these limitations, federation at the digital level can complement physical orchestration by leveraging interconnected NDTs. NDTs provide an abstraction layer that enables privacy-preserving, cross-domain coordination while maintaining the autonomy of each administrative entity. NDTs can operate at multiple levels of abstraction [ 15 ]; Node-level twins represent individual network elements or functions, capturing fine-grained behavior and local KPIs. Domain-level twins model broader segments such as the RAN, CN, edge, or cloud domains, aggregating insights from multiple nodes. Service-level twins span multiple domains, offering a holistic end-to-end view of service delivery and performance. These levels can be hierarchically coordinated, where lowerlevel twins (e.g., node-level NDTs) feed insights into higherlevel models (e.g., domain-level NDTs), enabling multi-scale analysis and decision support. This hierarchical composition forms the foundation of federated NDT management, where each administrative domain retains control of its data and models but shares standardized, anonymized insights to support cross-domain optimization. Such an approach also allows for pre-deployment validation—testing new controllers or 24 AI policies in a federated virtual environment before their release to the physical network—thus preventing potential conflicts between administrative domains and ensuring policy compliance. In large-scale or federated environments, multiple NDT instances may operate concurrently across geographically and administratively distributed infrastructures. The federated management component ensures the synchronization, coordination, and state consistency mechanisms among NDTs. Together with other blocks in the management layer, the federated management provides a policy-driven orchestration that regulates data exchange, event propagation, and lifecycle management across the NDT federation to prevent performance degradation or conflicting decisions. Additionally, depending on the use case, federated NDTs may need to interact with external simulators, emulators, or digital platforms to extend their modeling capabilities [ 200 ]. In such cases, a simulation framework, as described in Section VIII-B , plays a key role in maintaining temporal and semantic synchronization among distributed components, enabling accurate cross-domain experimentation and validation. F. Physical network management Our proposed architecture introduces a physical network management block, an intelligent orchestration and automation component bridging the physical network and its digital twin. This component integrates traditional LCM mechanisms with AI/ML-driven decision-making and closed-loop control. Its design focuses on two key capabilities, (i) increasing network programmability to enable dynamic, software-driven control and (ii) establishing efficient data collection pipelines to maintain consistency between the physical and digital layers. Moreover, the physical layer supports self-monitoring, selfanalysis, and self-adjustment mechanisms, key enablers for achieving zero-touch network operation in 6G environments. Central to the physical network operation are two interrelated closed-loop processes that enable adaptive and autonomous control. The internal closed-loop manages the creation and configuration of NDT instances, through the NDT management, coordinating interfaces between the physical network, simulation frameworks, and AI training modules. The external closedloop governs the runtime operations of the NDT, leveraging continuous telemetry to diagnose performance deviations, optimize configurations, and reapply decisions to the physical network in real time. Together, these feedback loops ensure that network management evolves from static, rule-based operation to a self-optimizing, AI-native paradigm. Furthermore, the MANO/ZSM layer supports auto-discovery and secure onboarding of physical devices, translating network states into actionable intents and operational policies. In this way, the physical network becomes a programmable, adaptive, and autonomous substrate for intelligent network management in the 6G era. IX. LIFECYCLE MANAGEMENT OF NDT INSTANCES Building upon the concepts and components described in the previous sections, this section focuses on the dynamic interactions between the various building blocks and layers of an NDT. These interactions are essential for orchestrating the critical stages of the NDT lifecycle, enabling the system to meet the previously outlined functional and non-functional requirements in Section III. To capture these complex interdependencies, we adopt a process-oriented view that describes how the system behaves over time. To that end, Unified Modeling Language (UML) sequence diagrams represent these workflows, as they provide a precise and intuitive mechanism for illustrating the chronological order of message exchanges among system entities. While the procedures described herein are intended to remain generic and broadly applicable, it is important to note that customized sequences should be developed for each specific use case to reflect domain-specific interactions and constraints. The following subsections introduce and briefly explain the key sequence diagrams underpinning the LCM of an NDT. A. Creation Figure 10 illustrates the creation of an NDT, a task typically initiated on demand due to the potential computational intensity associated with running high-fidelity simulations and performing “what-if” analyses. In our architecture, this process is orchestrated by the NDT MANO layer, specifically through its dedicated NDT Management component. Upon receiving a creation request, the NDT Manager interprets the operational and analytical requirements and translates them into a formal NDT descriptor [ 199 ]. This descriptor encapsulates all essential information needed to instantiate and execute the NDT. The NDT descriptor includes information related to the initial scenario deployment, including the conditions that should be evaluated, and initialization of parameters, including simulation setup; the basic and functional models to be used by the NDT, monitoring pipelines, performance bounds, and learning metrics to manage their lifecycle effectively; data to be used and where to retrieve it; the desired output format, which may range from user-facing suggestions (in open-loop configurations), to direct network actuation (in closed-loop scenarios) or predictive alerts; software and runtime dependencies, such as required simulators or emulators; interconnection of components to form a service, including any constraints on network characteristics like bandwidth and latency. The NDT descriptor can be built in co-creation with the end user, where the NDT system engages in a refinement process from high-level requests into executable specifications. This process can be done through external interfaces such as dashboards. Generating a precise NDT descriptor ensures context-aware instantiation and reliable support for real-time network analysis, optimization, and decision-making. Once the request for an NDT is received, the NDT Manager initiates the process by generating a mock-up of a generic NDT instance, in which some foundational blocks are preconfigured, as illustrated in Figure 10. This preliminary structure acts as a template, which the system refines based on the specific requirements outlined in the NDT descriptor. The orchestrator takes over by analyzing this descriptor and coordinating with the resource manager to verify whether the underlying infrastructure can support the deployment. The resource manager 25 NDT Instance Real World Representation List of available resources Simulation Manager Unified Data Model Model Manager Resource Manager Orchestrator NDT ID Get available resources Deploy NDT RequestBasic and Functional Model(s) Basic and Functional Model(s) IDs Simulations NDT Descriptor Basic and Functional Models IDs 6G-TWIN MANO Request NDT Creation Request Models Send simulation data Request simulation data NDT Creation acknowledgment Simulation Framework Simulation Running Instance Simulator Simulation Finished NDT Manager Unified Data Repository Simulation finished Start Simulationfor NDT Creation Human World Human Operator/ Network Applications Figure 10. NDT creation sequence diagram evaluates the availability of computing resources, such as CPUs or GPUs, necessary to run the NDT components. If sufficient resources are available, the orchestrator consults the model manager to ensure that all required basic and functional models are already created and accessible. These models are expected to be stored in the unified data model repository; if any models are missing, the model manager may initiate auxiliary processes to generate or retrieve them. Additionally, the system may need to pull relevant input data, either historical or real-time, from the UDRto feed into the basic models. At this stage, any simulation tools required for the NDT’s operation are instantiated by engaging the simulation manager, which interfaces with the appropriate simulation frameworks. These simulations may also need live or recorded data from the physical network, again accessed through the UDR by the simulation manager and sent to the simulation framework when starting a simulation. Once all necessary components, including models, data, and simulators, are properly instantiated and configured, the orchestrator finalizes the deployment of the NDT instance and returns an acknowledgment to the original requester, signaling that the NDT is operational and ready for use. B. Update The NDT is designed to evolve continuously by updating its models based on new data from the physical network to ensure a high-fidelity representation. The basic models are assumed to be a mere representation of the information coming from the physical network. In that sense, basic models are updated on the same basis as the data collection process. On the other hand, the functional models can be updated in at least two ways: (i) their performance is continuously monitored, and an update is triggered when there is a deviation from the performance bounds declared in the NDT descriptor; or (ii) the NDT orchestrator regularly programs an automated update. In the remaining section, we assume that the update is reactive rather than scheduled (first case). Depending on the type of functional model(s) the NDT is based on, several scenarios may occur when updating it. NDT Instance Real World Representation Unified Data Model Model Manager Resource Manager Orchestrator NDT Deployed Deploy NDT Monitor Loop Request quality Metrics Get quality Metrics Sent Quality Alert Request Functional model update Request latest version functional model Send latest version Stop Quality Alert Service not available due to update Dashboard NDT descriptor updated NDT Descriptor Models ready for update Figure 11. Update analytical functional models Note also that the NDT service continues to operate under the previous settings during this update. The user should be notified when an updated NDT version is available, e.g., by displaying a message in the dashboard. 1) Analytical functional models: As seen in Figure 11, the NDT monitoring system triggers the model update when a mismatch between the expected quality metrics and the measured ones is detected, sending a quality alert to the orchestrator. The NDT orchestrator then asks the model manager to update the functional model. The model manager communicates with the UDM repository to get the latest version of the functional model. A new model must be created if the latest version is the same as the current version. If a new version exists, the orchestrator will be notified and make the respective change. Additionally, it is the responsibility of the orchestrator to update the NDT descriptor to include the new version of the model and to stop the alert triggered by the monitoring system. Notice that a request to update an analytical model may happen when a physical principle has changed, e.g., the propagation mode is changed or a transmission technology has changed, and therefore, their parameters must adapt accordingly. 2) AI-based functional models: As mentioned in Section VII-C , if the functional model is based on AI/ML techniques, different pipelines are followed depending on the 32 of the mobile network conditions as they vary temporally and spatially. The functional model trained within the AI workflow management, then stored and managed within the UDM, demonstrates how RL creates intelligent, adaptive networking solutions that respond to the dynamic requirements of next-generation connected vehicle applications. 1) Basic and functional Models: The proposed framework includes Basic models and Functional models and is organized into two key components: • Basic models — representing the 6G mobile network infrastructure, including gNB entities for radio resource management, handover, connection establishment, and the vehicular infrastructure, including cars, roads and topography information. Normally, these basic models are provided by the different simulators for the first work. As well as, constructing a clustered graph representation of another city based on [ 201 ] for road networks, and [ 205 ] for stochastic latency and bandwidth parameters modeled using normal distributions to capture temporal network variability in [210] • Functional models — encompassing AI-based modules for both coverage prediction and optimal vehicle routing, while [ 210 ] focuses specifically on optimal vehicle routing through DRL. The optimal vehicle routing functional model formulates the route optimization problem as a MDP, where States encode vehicle position, destination coordinates, associated network performance metrics (SINR, or alternatively bandwidth and latency), and accumulated path metrics. The Action space consists of basic navigation maneuvers (turn left,turn right,go straight) [ 211 ], or selecting adjacent intersections, with action masking applied to enforce graph connectivity constraints and prevent invalid moves. The routing policy minimizes path length while maintaining strong network coverage, ensuring safe teleoperation under dynamic network conditions. 2) DRL-based functional models: We propose two approaches for the route optimization problem. The first approach employs a Dueling Deep Q-Network (DDQN) architecture where its Reward function balances competing objectives through a weighted linear combination: R(s, a)=wd·Rd(s, a)+wb·Rb(s, a)+wl·Rl(s, a)(1) where the normalized components are defined as Rd= −d(e)/dmax for the distance penalty, Rb= (b(e, τ)− bmin)/(bmax −bmin) for the bandwidth reward, and Rl= −l(e, τ)/lmax for the latency penalty. Through systematic experimentation, we determined the optimal weights wd= 0.5 , and wb=wl= 0.25 . The workflow of the DDQN algorithm is illustrated in Figure 22. The second approach uses the Proximal Policy Optimization (PPO) [ 212 ], selected for its stability and efficiency in continuous control problems. The multi-objective reward function integrates several components: rtot(t) =αrshortest(t)+βrcoverage(t) +γrpenalty(t)+robjective +rillegal (2) Figure 22. Dueling DQN algorithm flowchart showing dual-stream architecture and action selection process with exploration-exploitation balance. where rshortest incentivizes progress toward the destination, rcoverage encourages higher network coverage, rpenalty penalizes insufficient coverage, robjective rewards successful task completion, and rillegal penalizes unlawful actions. The weights α , β , and γ allow tuning the balance between efficiency, connectivity, and safety. This formulation balances progress toward the destination, network coverage quality, and penalties for unsafe or disconnected states. The algorithm supports three operational scenarios: • Coverage Maximization – prioritizes maintaining high SINR values. • Shortest Path Maximization – minimizes travel distance while keeping coverage above threshold. • Balanced Optimization – achieves a trade-off between path length and connectivity, ensuring stable communication and efficient routing. 3) Case study and results: To demonstrate the effectiveness of the proposed approach, a comprehensive case study was conducted, using a realistic city map where a teleoperated vehicle navigates between multiple gNBs. In the first implementation, the Coverage Prediction model is needed to map the vehicle positioning with network-related metrics such as SINR, delay, or bandwidth. Specifically, the coverage prediction model was developed using a Random Forest (RF) regression algorithm [ 213 ] trained on SINR data collected along simulated vehicle trajectories in the city center of Bari, Italy. This experiment provides a concrete validation of the proposed NDT-based AI workflow under realistic conditions. The trained RF model achieves R2= 0.9993 and RMSE = 0.3274, demonstrating outstanding predictive performance. The resulting SINR heatmaps (see Figure 23) provide a spatially continuous estimation of coverage quality, which informs the 33 Figure 23. Heatmap of SINR values over the city of Bari generated by the Random Forest coverage prediction model. Figure 24. Learning performance of the PPO agent in the balanced optimization scenario, showing convergence and stable reward accumulation. routing agent with accurate connectivity information. Regarding the optimal vehicle routing model, Figure 24 depicts the learning outcomes of the PPO model, demonstrating convergence within a few hundred steps and an average cumulative reward exceeding 1400, indicating fast and stable convergence of the PPO agent in all configurations, validating its robustness for multi-objective optimization in stochastic urban network environments. More details of the evaluation and usage of this functional model can be found in Paparella et al. [211]. In addition to the PPO, we evaluated the DDQN model [ 210 ] implemented on the city center of Nevers, France. The urban environment was extracted from OpenStreetMap [ 201 ] and transformed into a directed graph through clustering. The network metrics used were latency and bandwidth assigned to each edge based on proximity to 5G base stations from the OpenCelliD dataset [ 205 ], with stochastic temporal variations Figure 25. Combined performance metric (normalized KPI respect rate + normalized inverse distance) demonstrating overall routing effectiveness across multiple objectives. modeled using normal distributions. The DDQN model was evaluated by comparing it against classical routing algorithms: Shortest Path,Lowest Latency, and Highest Bandwidth. To enable a holistic assessment of multi-objective performance, we designed a combined evaluation metric that normalizes and aggregates both the operational requirement compliance rate and the route efficiency. Under this unified metric, our method achieved the highest overall score, as illustrated in Figure 25. These results, obtained through the NDT framework, confirm that the proposed reinforcement learning approach effectively balances competing objectives, achieving a level of performance unattainable by any single-objective optimization strategy. This demonstrates the NDT’s capability to serve as a comprehensive testbed for validating functional models before deployment in physical networks. E. Co-simulation framework As mentioned in the previous section, several simulators can be coupled to enhance the capabilities of the NDT in predictive modeling. The co-simulation framework of the architecture (cf. Figure 2) is responsible for the coordination and sequential execution of separate simulators and supports data exchange among the simulators, as mentioned in Section VII-B and Section VIII-B . For this purpose, protocol translators are used to map the generic protocol-2 services to specific simulators. This has already been implemented for some of the most well-known simulators in the field of mobile communication. This section presents a reference implementation of the cosimulation framework, with some of the first features already tested successfully. 1) Implemented components: One of these simulators is OMNeT++, a widely used network simulator in communication studies. Our protocol translator implementation for OMNeT++ supports starting and stopping OMNeT++ as well as telling OMNeT++ the current position of UEs in the simulation. Additionally, the protocol translator is able to make OMNeT++ continue its simulation until a given point in (simulated) time. Inserting and deleting UEs to and from the simulation are also 34 supported at runtime. OMNeT++ is started with simulation configuration files that contain further information about the simulated tasks, e.g., the simulation time and transmission power. A second supported simulator is SUMO, the most wellknown mobility simulator for vehicles. Our protocol translator prototype allows starting and stopping SUMO based on given configuration files, processing until a given time, as well as calling the position of all vehicles at runtime. Also, added and removed vehicles are communicated to the co-simulation framework via the protocol translator of SUMO. Third, a protocol translator prototype for AirMobiSim has been implemented. AirMobiSim computes the position of UAVs based on path information that is given to the simulator at the start of the simulation. The protocol translator allows starting and stopping AirMobiSim and retrieval of the current position of UAVs. Information about newly added and deleted UAVs is sent to the co-simulation framework as well. In order to make a fully coupled simulation system, also a prototype of the co-simulation framework itself has been implemented. This coordinates the creation and deletion of all necessary simulator instances based on a configuration file. It tells the simulators which configuration files they shall use and starts their execution. Then, it coordinates the execution of each simulator by telling them to simulate to a certain time instance. When the simulators reach this synchronization point, data from the simulators is collected and sent to the relevant other simulators. For example, the number and position of all vehicles is collected from SUMO via its protocol translator and sent to OMNeT++ via its protocol translator. Similarly, the UAV position data from AirMobiSim is collected by the co-simulation framework at each synchronization point and sent to OMNeT++. Furthermore, during run-time the cosimulation framework also provides a mechanism such that one simulator may influence the behavior of another simulator by sending commands like changing the route of a vehicle or changing its speed at certain points in time. This is done at the synchronization phase by pro-actively asking each simulation to submit any requests it wishes to perform before moving to the next step of the simulation. At the end of the simulation, all simulators are stopped and their instances are disposed. 2) Simulation results: The implemented components have been tested in several configurations. It has been shown that running the co-simulation framework with OMNeT++ and SUMO is working, as well as an extended simulation containing OMNeT++, SUMO, and AirMobiSim. This can simply be realized by activating AirMobiSim in the co-simulations’ configuration file, provided that each simulator has a fitting configuration file that is loaded when starting each simulator. The performance of the co-simulation framework has been measured in order to make sure that the overhead due to the introduction of co-simulation middleware is reasonable in comparison to the execution time of the tightly coupled simulation framework. More information on the performance results can be found in [ 138 ]. A screenshot of the running simulation is presented in Figure 26, where on the left, OMNeT++ is shown, and on the right SUMO. The positions of the vehicles in OMNeT++ are taken from SUMO via the coFigure 26. Screenshot of the running co-simulation framework. simulation framework. The position of the UAV is taken from AirMobiSim via the co-simulation framework. AirMobiSim has no own GUI. The blue dotted lines represent packet transmissions, showing that OMNeT++ correctly uses the vehicle data it gets from SUMO and AirMobiSim, respectively. There is no direct communication between OMNeT++ and the other simulators; all interactions are realized via protocol-2. XI. NDT CONSIDERATIONS Beyond its architectural and functional design, the deployment of an NDT must address several cross-cutting considerations that ensure its reliability, trustworthiness, and interoperability. These include the security and privacy mechanisms necessary to protect sensitive network data, the interfacing and data collection processes that enable seamless integration with real and virtualized network components, and the governance principles that regulate data sharing and model management. Finally, comprehensive evaluation methodologies are required to assess the NDT’s performance, accuracy, and operational impact across different lifecycle stages. The following sections examine each of these aspects in detail. A. Security and privacy considerations for NDTs An NDT is not only a visualization or analytical tool for network management but a digital copy of the real networks interacting with it bidirectionally. For this reason, NDT is planned to be positioned at the center of the real network’s decision-making cycle. Consequently, a breach in the security of an NDT or NDT’s data sources / synchronization mechanisms can directly impact the decision-making mechanisms and systemic risks. An NDT requires collection, consolidation, and processing of large-scale data from heterogeneous network environments including different devices and components. This might create new, use case-specific attack surfaces and privacy threats. The operation of an NDT relies on sensitive information such as real-time telemetry data, performance indicators, and control feedback; therefore, the confidentiality, integrity, and availability of this data must be protected throughout all stages of its life cycle. In addition to that, the integration of data governance, privacy, and compliance mechanisms into the NDT architecture should be considered from the initial design phase. 35 Table V THREAT CATEGORIES IN NDT SYSTEMS Threat category Brief description Data storage and NDT repository Unauthorized access to the NDT and associated network functions. Critical or confidential data, functional models, and ML models may be stolen or copied. Sensors or software in the data collection process An adversary may spoof or manipulate data traffic between the NDT and the physical network, leading to corrupted model states or false decisions. Physical network component An adversary can damage or interfere with physical assets Data flow between physical network assets and NDT An attacker can disrupt synchronization, listen to or alter data flow. They can also launch a denial-of-service (DoS) attack Network management models and functions An insider attacker or an attacker with unauthorized access can change configurations, security policies, or steal confidential data Application layer interfaces Application layer data can be altered and intercepted; NDT components can be affected by DoS attacks Within this framework, NDTs generate predictive and prescriptive decisions for live networks, so security-by-design must be a guiding principle for NDTs. Security and privacy analysis to be performed on NDT-based network management should identify threat vectors, attack surfaces, assess vulnerabilities and risks to define countermeasures and mitigation scenarios. The overall architecture of NDT includes physical and digital components and assets. These assets might be vulnerable to various threats. A summary of the threats for these assets is provided in the Table V [214]. All identified threats must be addressed and mitigated according to the specific use case. While there are established mitigation techniques, solutions, and security controls available to protect each asset, the most appropriate measures should be selected based on the asset’s risk profile and operational context. Additionally, the applicability and effectiveness of any chosen controls must be evaluated against real-world operational constraints to ensure they provide the intended protection with the required complexitty and cost [215]. B. NDT governance NDT Governance plays a critical role in ensuring the secure, fair, and transparent management of data, models, and interactions among multiple stakeholders. Effective governance establishes the principles, policies, and technical mechanisms required to maintain trust, accountability, and regulatory compliance across a distributed NDT ecosystem. It defines how data is accessed, shared, and monetized while safeguarding privacy and sovereignty, key aspects for the operational viability of large-scale NDT deployments. Unlike other sectors such as the European Network of Transmission System Operators for Electricity in the energy domain, or even certain wired communication infrastructures, the radio access network lacks a centralized information hub. Consequently, the governance of NDTs for 6G networks must adopt a decentralized approach. In such a framework, each MNO manages its own NDT instance while selectively sharing data with other MNOs’ NDTs when cooperation or Figure 27. NDT governance through the data space federation is required. Additionally, MNOs are responsible for providing relevant information to regulatory authorities, Virtual MNOs, and application providers. These exchanges take place within the framework of a regulated data space [ 216 ], as introduced in Section VII-A 1, which specifies the technical and legal rules governing interactions between MNOs and external stakeholders (see Figure 27). Nevertheless, providing the national network regulator with a partial but relevant view of operations is essential for ensuring compliance with regulations and legal requirements. They should take into account the different governance: • Transparency & accountability. Clear policies must define who can access, modify, or share data within the NDT, ensuring traceability and accountability for all actions. • Interoperability & standards. Adopting common data models, interfaces, and protocols enables seamless interaction between different NDTs and stakeholders, fostering collaboration and reducing fragmentation. • Data sovereignty & privacy. Governance should respect data ownership rights and enforce privacy protections, especially when sharing sensitive information across MNOs, regulators, and third parties. • Role-based access control. Access to NDT data and functionalities should be granted based on predefined roles, ensuring that each actor—whether MNO, Vritual MNO, or regulator—has appropriate permissions aligned with their responsibilities. • Regulatory alignment. The governance framework must align with national and international regulations, facilitating audits and ensuring that data exchanges meet legal and industry standards. • Dynamic adaptability. As technologies and requirements evolve, the governance model should be flexible enough to accommodate new use cases, stakeholders, and regulatory changes without disrupting existing operations. • Trust & security. Robust security measures, such as encryption and authentication, are critical to protect data integrity and build trust among all participants in the NDT ecosystem. • Federated data sharing. Mechanisms for secure, consentbased data sharing should be established, allowing stakeholders to contribute and consume data while maintaining 36 control over their own assets. Additionally, related domains, such as building information and transportation, can influence the NDT, potentially forming a federation of diverse NDTs. The key challenge lies in establishing a tailored governance framework to manage the flow of data into and out of this NDT federation, taking into account the distinct roles of both the NDTs and the actors involved. C. Data collection requirements and interfaces for NDTs Data collection is performed across the physical network, which is trending to scale up exponentially, with more and more devices connected to the internet, turning the IoT paradigm into the Internet of Everything, a term coined by Cisco [ 217 ] to describe how more people are connected through new device types, generating new types of information. This creates challenges for the NDT of the future 6G network, necessitating a new level of scalability, dealing with both the exponential increase of connected devices and the wide range of data types generated by these heterogeneous devices. All of this needs to be done in a lightweight and efficient manner, to make sure the NDT delivers its enhanced decision-making capabilities towards network and service management. As discussed in Section VI-B , the Data collection framework seeks to ensure that data is gathered, processed and stored in a way that guarantees interoperability, accuracy and efficiency across the modeling and simulation tasks of the NDT. Some recommendations have been highlighted during the development of the architecture, based on both state of the art research initiatives and by standardization body activities, such as the IETF [ 218 ]; this led to the creation of a set of requirements that must be fulfilled by the NDT in order to make sure that these challenges are addressed. Data collection performed by the NDT must be target driven and context-aware, focusing on specific operational needs rather than monitoring all available data; this can be done by operators, defining key parameters that ensure only pertinent data is collected (such as operational status and configuration of devices, events, logs, lifecycle operation data, user data and service data) rather than indiscriminately gathering all available data, making sure resource utilization of the network is optimized. Extending data collection across multiple data domains, going beyond the traditional service-based network architecture to the Cloud-Edge-IoT continuum, promotes the integration of multiple devices and enhances the functionality of the NDT. Effective naming and caching of data can also contribute to improve access speeds and reduce redundant entries within the data repository. The NDT should also be equipped with multiple tools that can handle data collection in an environment of heterogeneous data sources, integrating mechanisms that harmonize different data formats in a lightweight and efficient manner, standardizing data flows to maintain consistency and accuracy; this contributes to a more efficient data processing framework, minimizing the cost of computing, storage and communication bandwidth, as well as the likelihood of data processing and utilization errors across applications and services. Data from various sources must also be easily integrated into the data repository, allowing for the NDT to have, at all times, a comprehensive and up to date view of the network for effective analysis and decision-making processes. Interfaces between the data collection framework and the NDT should also be open and standardized, avoiding any hardware/software dependencies and ensuring interoperability between components, allowing for the management and configuration of different timescales across devices and applications, through a secure and reliable channel that exchanges information in a federated manner, and future-proofed for extensibility and backwards compatibility. This framework is further enhanced by supporting multiple communication protocols, enabling both compatibility with legacy systems and future device interfaces. Finally, beyond multi-device data collection, the NDT must also be capable of handling multi-destination delivery, as data collected from the same source could be requested by multiple instances of the NDT in the federation. Data sharing is managed by a robust set of access policies, based on the aforementioned federation paradigm, enhancing fault tolerance, access speeds and scalability, while ensuring privacy and security of any sensitive information. These requests must be fulfilled as efficiently as possible, to maintain the scalability requirements, and ensure that the systems receive the data in a timely manner, through multicast communication and intelligent routing systems, for the prerequisites of the applications and services. D. NDT evaluation considerations The evaluation process plays a crucial role in the NDT framework. To this purpose, ITU-T Y.3091 [ 37 ] Recommendation highlights three main requirements for an NDT: • Fidelity: how accurate the metrics from performance models are with the physical network; • Efficiency: made by two sub-requirements: (i) performance models should be faster than real-time physical network; (ii) models should be easy to deploy and consume rational resources; • Flexibility: means that the NDT should be flexible to provide service on-demand according to various network applications by selecting a variety of cross domain resources on demand, flexibly collecting and storing data, combining different data models and interacting with other DTs. In addition to the requirements, the ITU-T Y.3091 recommendation provides a framework for assessing the maturity and functionality of an NDT in a hierarchical and recursive fashion. More specifically, five NDT capability levels are defined: • L1 Representation level: NDT can realize one-way mapping from the physical network to the virtual twin. • L2 Interaction level: Based on the representation level (L1), the control channel is added from the virtual twin to the physical network. • L3 Prediction level: Based on the interaction level (L2), the virtual twin can analyze the characteristics and trends 37 Table VI NDT EVALUATION INDICATORS BY DIMENSION AND LEVEL Dimension Evaluation Indicators NDT L1 NDT L2 NDT L3 NDT L4 NDT L5 Data service Data richness ≥L1 ≥L2 ≥L3 ≥L4 =L4 Update frequency ≥L1 ≥L2 ≥L3 ≥L4 =L5 Compatibility ≥L1 ≥L2 ≥L2 ≥L3 =L4 Data quality ≥L1 ≥L2 ≥L2 ≥L3 =L3 Data service interface ≥L1 ≥L1 ≥L2 ≥L3 =L3 Efficiency ≥L1 ≥L1 ≥L2 ≥L3 =L3 Digital twin modelling Basic model integrity ≥L1 ≥L2 ≥L3 ≥L3 =L4 Functional model integrity ≥L1 ≥L2 ≥L3 ≥L4 =L5 Standardization ≥L1 ≥L2 ≥L3 ≥L4 =L4 Interfaces ≥L1 ≥L2 ≥L3 ≥L4 =L4 Update frequency ≥L1 ≥L2 ≥L3 ≥L4 =L5 Flexibility ≥L1 ≥L2 ≥L3 ≥L3 =L4 Efficiency ≥L1 ≥L2 ≥L3 ≥L4 =L4 Interactive mapping Mapping mode ≥L1 ≥L1 ≥L2 ≥L2 =L3 Real to virtual mapping ≥L1 ≥L2 ≥L3 ≥L4 =L4 Virtual to real mapping ≥L1 ≥L2 ≥L3 ≥L4 =L5 Interface richness ≥L1 ≥L2 ≥L3 ≥L4 =L5 Interaction quality ≥L1 ≥L2 ≥L3 ≥L4 =L5 Intelligence Orchestration ≥L1 ≥L1 ≥L2 ≥L3 =L3 Analysis ≥L1 ≥L2 ≥L3 ≥L3 =L3 Decision-making ≥L1 ≥L1 ≥L2 ≥L2 =L3 Instruction execution ≥L1 ≥L2 ≥L2 ≥L3 =L3 AI/ML model training and inference ≥L1 ≥L1 ≥L1 ≥L2 =L3 Model quality ≥L1 ≥L2 ≥L2 ≥L3 =L4 AI/ML model explainability ≥L1 ≥L2 ≥L3 ≥L3 =L4 User experience Visualization scope ≥L1 ≥L2 ≥L3 ≥L4 =L5 Data visualization mode ≥L1 ≥L2 ≥L3 ≥L3 =L3 Entity visualization mode ≥L1 ≥L2 ≥L3 ≥L3 =L4 Interaction ≥L1 ≥L2 ≥L3 ≥L3 =L4 Trustworthiness Security ≥L1 ≥L2 ≥L3 ≥L3 =L4 Privacy ≥L1 ≥L2 ≥L3 ≥L3 =L4 Reliability ≥L1 ≥L2 ≥L3 ≥L3 =L4 Resilience ≥L1 ≥L2 ≥L3 ≥L3 =L4 of the data collected from the physical network and can use strategies and algorithms to infer indicators. • L4 Optimization level: Based on the prediction level (L3), the virtual twin cannot analyze and predict the performance and future trends of the physical network but can also use AI algorithms, expert knowledge, big data analysis, and other intelligent technologies. • L5 Autonomy level: As the ideal goal of NDT, the virtual twin and the physical network live in symbiosis with each other. Based on the above levels, the maturity of the NDT can be assessed for the following along the following set of dimensions: •Data service •Digital Twin modeling •Interactive mapping •Intelligence user experience •Trustworthiness Each dimension is associated with a set of evaluation indicators for each of which a specific level, as detailed in [ 37 ], must be assessed. The indicators are designed following the SMART principles, i.e. Specific, Measurable, Assignable, Realistic and Time-related, as detailed in [ 219 ]. Moreover, evaluations at different levels are strongly interrelated and the synthetic capability level of the NDT is calculated in a recursive fashion. In Table VI, Dimensions and Evaluation indicators are shown together with the minimum requirements to achieve a desired capability at NDT level. In order to assess the level of each evaluation indicator, a set of KPIs should be established, and each KPI needs to be linked to one or more indicators. Such KPIs can be both qualitative and quantitative. Qualitative KPIs are related to aspects that cannot be measured, such as the ability to replicate a real network, the coverage of the basic and functional models as opposed to a specific use case, the flexibility of the NDT in the integration with other DTs, the conformance of methods adopted to ensure security and privacy for data in transit and at rest and the contribution to the standardization efforts. On the contrary, quantitative KPIs are mostly related to network performance metrics that can be numerically determined during specific simulation campaigns. In Table VII, a sample set of those KPIs is provided in which the main performance metrics of a typical network evaluation scenario are considered. In order to compute the KPIs, a data collection must be performed during the simulation campaign. To this purpose, consolidated methodologies, such as FESTA-V [ 220 ] can be used. More specifically, FESTA-V provides a clear framework made of three steps: (i) PREPARE, in which the simulation scenarios and the data to be collected are defined, (ii) USAGE, 38 Table VII KEY NETWORK PERFORMANCE INDICATORS (KPIS) KPI category Brief description Data Rate/Capacity The amount of data transmitted over a network in a given amount of time. Latency Delay introduced by the network and applications, due to transmission, processing, queuing, or propagation. Service Reliability & Availability Measures the stability of performance (overall QoS) and the operability of the network against expected performance. Packet Loss The percentage of data packets transmitted across the network that are lost or fail to reach their destination. Edge Node Utilization The amount of processing or resource capacity of an edge node used over time. Energy Efficiency of network and device power consumption, including energy use and management within the infrastructure. Channel/Spectral Efficiency Utilization of spectral resources for wireless communication, including spectral efficiency, SINR, and channel reliability. in which the simulation is performed and the data are collected and (iii) EVALUATION and IMPACT, in which the data are processed and the KPIs are computed. In this context, each KPI should be linked to its corresponding Evaluation Indicator. Finally, the individual capability level at each relevant Evaluation Indicator and Dimension level and, then, the overall maturity of the NDT, can be computed by using the recursive formula detailed in [37]. XII. CONCLUSION AND FUTURE WORK This paper proposes an NDT framework integrating AI, simulation and adaptive management for future 6G systems, treating the twin as a lifecycle-managed capability rather than a monolithic artefact. It brings together a harmonized data pipeline, reusable models, and an orchestration layer that instantiates scoped NDT instances on demand, enabling closedloop control from cloud to far edge. In a field where NDT work across the literature and industry is fragmented, this interoperable, AI-native architecture offers a coherent way forward by harmonizing data and unifying model management. In doing so, it supports seamless transitions between datadriven and simulation-driven reasoning, enables closed-loop analysis and control from cloud to far-edge, and keeps scalability, interoperability, and governance at the forefront. Early prototyping across telemetry ingestion, model training and re-training, network and mobility simulation, suggests that the approach is feasible and that it offers a practical path for safe “what-if” experimentation and repeatable actuation. The framework is designed to be transferable across domains: its modular decomposition, emphasis on model catalogs and reusable pipelines, and alignment with contemporary orchestration practices make it suitable for scenarios where fidelity, responsiveness, reliability, and energy efficiency must be balanced. By elevating the NDT concept to a managed, ondemand capability, the proposal provides a concrete blueprint that can be replicated and extended, accelerating the move from concept to operation in real systems. Future work will concentrate on end-to-end validation through concrete and tangible use-cases that exercise the full lifecycle: instrumenting real networks, harmonizing data flows, instantiating targeted NDT instances, training and adapting basic and functional models with auditable KPIs. As part of 6G-TWIN, this will be pursued through two use-cases, which will serve as proving grounds to showcase repeatable results. Beyond demonstrating performance and feasibility, these demonstrators are intended to deliver reusable assets—data schemas, model and experiment catalogs, and MLOps and co-simulation recipes—that facilitate uptake beyond the project and help translate the framework into deployable practice. FUNDING SOURCES The 6G-TWIN project has received funding from the Smart Networks and Services Joint Undertaking (SNS JU) under the European Union’s Horizon Europe research and innovation program under Grant Agreement No 101136314. REFERENCES [1] Ericsson, “Mobile data traffic forecast – Ericsson Mobility Report,” https://www.ericsson.com/en/reports-and-papers/mobility-report/ dataforecasts/mobile-traffic-forecast, 2025, accessed: 2025-10-24. [2] D.-H. Tran et al., “Network Digital Twin for 6G and Beyond: An End-to-End View Across Multi-Domain Network Ecosystems,” IEEE Open Journal of the Communications Society, vol. 6, pp. 6866–6911, 2025. [3] A. Zaki-Hindi et al., “Network Digital Twins for 6G: Defining Data Taxonomy and Data Models,” in IEEE Conference on Standards for Communications and Networking (CSCN 2024). Belgrade, Serbia: IEEE, Nov. 2024, pp. 124–128. [4] S. Faye et al., “A Functional Framework for Network Digital Twins,” in Joint European Conference on Networks and Communication & 6G Summit (2025 EuCNC & 6G Summit). Poznan, Poland: IEEE, Jun. 2025. [5] G.-T. Consortium, “D1.1 – 6g-twin architecture and technical foundations,” 6G-TWIN Project (SNS-JU Grant), Public Deliverable D1.1, Dec. 2025, to be published. [6] ——, “D2.1 – data governance, privacy and harmonization,” 6G-TWIN Project (SNS-JU Grant), Public Deliverable D2.1, Dec. 2025, to be published. [7] ——, “D1.2 – secured, scalable and distributed data exposure and collection framework,” 6G-TWIN Project (SNS-JU Grant), Public Deliverable D1.2, Dec. 2025, to be published. [8] ——, “D2.2 – basic models,” 6G-TWIN Project (SNS-JU Grant), Public Deliverable D2.2, Dec. 2025, to be published. [9] ——, “D5.1 – evaluation methodology, planning and coordination,” 6G-TWIN Project (SNS-JU Grant), Public Deliverable D5.1, Dec. 2025, to be published. [10] ——, “D2.3 – functional models,” 6G-TWIN Project (SNS-JU Grant), Public Deliverable D2.3, Dec. 2025, to be published. [11] ——, “D1.3 – frameworks for zero-touch service and network management and the orchestration of its ai-based nf and ns,” 6G-TWIN Project (SNS-JU Grant), Public Deliverable D1.3, Dec. 2025, to be published. [12] ITU-T, “Digital Twin Network - Requirements and Architecture,” ITU, Recommendation Y.3090, Feb. 2022. [Online]. Available: https://www.itu.int/rec/T-REC-Y.3090/en [13] 3GPP, “Study on Management Aspects of Network Digital Twin (Release 19),” 3GPP, TR 28.915 V1.0.1 (2024-09), 2024. [14] O-RAN next Generation Research Group (nGRG), “Research Report on Digital Twin RAN Use Cases,” https://mediastorage.o-ran.org/ngrg-rr/ nGRG%20RS01%20Digital%20Twin%20Use%20Case%20Research% 20Report%20v1.0.pdf, accessed: 2025-10-21. [15] ETSI, “Zero-touch network and Service Management (ZSM); Network Digital Twin,” ETSI, Group Report ZSM 015 V1.1.1, Feb. 2024. [16] C. Zhou et al., “Network Digital Twin: Concepts and Reference Architecture,” Tech. Rep., 2022. [17] Spirent, “Simplifying 5G with a Network Digital Twin,” Spirent, White Paper 11/22 Rev B, 2022. 39 [18] ZTE, “Wireless Network Digital Twin as A Service,” ZTE, Technology White Paper 2023-10-11, 2023. [19] Ericsson. 6G - Follow the Journey to the next Generation Networks. [Online]. Available: https://www.ericsson.com/en/6g [20] A. Geelen, “Deutsche Telekom, Google Cloud, and Ericsson Demonstrate Network Transformation Milestone with 5G Cloud-Native Network Pilot,” https://www.telekom.com/en/media/media-information/ archive/5g-cloud-native-pilot-shows-efficiency-1026992, accessed: 2025-10-21. [21] Nokia, “Toward a 6G AI-Native Air Interface,” https://www.nokia.com/ asset/210299/, accessed: 2025-10-21. [22] ——, “Nokia, DOCOMO and NTT make two key 6G advances,” https://www.nokia.com/newsroom/ nokia-docomo-and-ntt-make-two-key-6g-advances-mwc23/, accessed: 2025-10-21. [23] ——, “Dynamic Digital Twin,” https://www.nokia.com/bell-labs/ research/air-lab/modelling-optimization/dynamic-digital-twin/, accessed: 2025-10-21. [24] Y. Chen et al., “NetGPT: An AI-Native Network Architecture for Provisioning Beyond Personalized Generative Services,” IEEE Network, vol. 38, no. 6, pp. 404–413, Nov. 2024. [25] Huawei, “Huawei Releases the Intelligent World 2030 Report to Explore Trends in the Next Decade,” https://www.huawei.com/en/news/2021/9/ huawei-releases-intelligent-world-report-2030, accessed: 2025-10-21. [26] X. Wang et al., “Artificial Intelligence-Based Techniques for Emerging Heterogeneous Network: State of the Arts, Opportunities, and Challenges,” IEEE Access, vol. 3, pp. 1379–1391, 2015. [27] R. Dong et al., “Deep Learning for Hybrid 5G Services in Mobile Edge Computing Systems: Learn From a Digital Twin,” IEEE Transactions on Wireless Communications, vol. 18, no. 10, pp. 4692–4707, Oct. 2019. [28] Y. Zhou et al., “A Hierarchical Digital Twin Network for Satellite Communication Networks,” IEEE Communications Magazine, vol. 61, no. 11, pp. 104–110, Nov. 2023. [29] I. Vilà et al., “On the Design of a Network Digital Twin for the Radio Access Network in 5G and Beyond,” Sensors, vol. 23, no. 3, p. 1197, Jan. 2023. [30] N. P. Kuruvatti et al., “Empowering 6G Communication Systems With Digital Twin Technology: A Comprehensive Survey,” IEEE Access, vol. 10, pp. 112 158–112 186, 2022. [31] M. Polese et al., “Colosseum: The Open RAN Digital Twin,” IEEE Open Journal of the Communications Society, vol. 5, pp. 5452–5466, 2024. [32] H. Gao et al., “Digital Twin Enabled 6G Radio Testing: Concepts, Challenges and Solutions,” IEEE Communications Magazine, vol. 61, no. 11, pp. 88–94, Nov. 2023. [33] P. Soto et al., “Designing the Network Intelligence Stratum for 6G networks,” Computer Networks, vol. 254, p. 110780, Dec. 2024. [34] Y. Wang et al., “End to End AI Architecture for Next Generation Network,” IEEE Wireless Communications, vol. 31, no. 1, pp. 86–92, Feb. 2024. [35] S. Zeb et al., “Toward AI-Enabled NextG Networks with Edge Intelligence-Assisted Microservice Orchestration,” IEEE Wireless Communications, vol. 30, no. 3, pp. 148–156, Jun. 2023. [36] S. Faye et al., “Integrating Network Digital Twinning into Future AIbased 6G Systems: The 6G-TWIN Vision,” in 2024 Joint European Conference on Networks and Communications & 6G Summit (EuCNC/6G Summit). IEEE, Jun. 2024, pp. 883–888. [37] ITU-T, “Digital Twin Network - Capability levels and evaluation methods,” ITU, Recommendation Y.3091, Dec. 2023. [Online]. Available: https://www.itu.int/rec/T-REC-Y.3091/en [38] J. Lucek, “LLM Connector: a Conversational AI Assistant for Your Network,” https://community.juniper.net/blogs/julian-lucek/2025/06/27/ llm-connector-a-conversational-ai-assistant, 2025. [39] Next Generation Mobile Networks Alliance (NGMN), “Network Architecture Evolution Towards 6G,” NGMN Alliance, Tech. Rep., 2025. [Online]. Available: https://www.ngmn.org/wp-content/uploads/ 250218_Network_Architecture_Evolution_towards_6G_V1.0.pdf [40] Y.-B. Lin et al., “Effects of Transport Network Slicing on 5G Applications,” Future Internet, vol. 13, no. 3, p. 69, Mar. 2021. [41] B. A. A. Nunes et al., “A Survey of Software-Defined Networking: Past, Present, and Future of Programmable Networks,” IEEE Communications Surveys & Tutorials, vol. 16, no. 3, pp. 1617–1634, 2014. [42] M. Liyanage et al., “A survey on Zero touch network and Service Management (ZSM) for 5G and beyond networks,” Journal of Network and Computer Applications, vol. 203, p. 103362, Jul. 2022. [43] ITU-T, “Autonomous networks - Architecture framework,” ITU, Recommendation Y.3061, Dec. 2023. [Online]. Available: https: //www.itu.int/rec/T-REC-Y.3061/en [44] C. Zhou et al., “Data Collection Requirements and Technologies for Digital Twin Network,” Internet Engineering Task Force, InternetDraft draft-zcz-nmrg-digitaltwin-data-collection-03, Jul. 2023, work in Progress. [45] J. Case et al., “A Simple Network Management Protocol (SNMP),” 1990. [Online]. Available: https://datatracker.ietf.org/doc/html/rfc1157 [46] R. Enns et al., “Network Configuration Protocol (NETCONF),” 2011. [Online]. Available: https://datatracker.ietf.org/doc/html/rfc6241 [47] sFlow, “sFlow Overview,” 2003. [Online]. Available: https://sflow.org/ sFlowOverview.pdf [48] S. Panchen et al., “InMon Corporation’s sFlow: A Method for Monitoring Traffic in Switched and Routed Networks,” RFC 3176, Sep. 2001. [49] J. Scudder et al., “BGP Monitoring Protocol (BMP),” RFC 7854, Jun. 2016. [Online]. Available: https://www.rfc-editor.org/info/rfc7854 [50] G. Fioccola et al., “Alternate-Marking Method,” RFC 9341, Dec. 2022. [Online]. Available: https://www.rfc-editor.org/info/rfc9341 [51] F. Brockners et al., “Data Fields for In Situ Operations, Administration, and Maintenance (IOAM),” 2022. [Online]. Available: https: //www.rfc-editor.org/rfc/inline-errata/rfc9197.html [52] B. Claise et al., “Packet Sampling (PSAMP) Protocol Specification,” 2009. [Online]. Available: https://www.rfc-editor.org/rfc/rfc5476.txt [53] “O-RAN Architecture Description v15.00,” O-RAN Alliance, Tech. Rep. O-RAN.WG1.TS.OAD-R004-v15.00, Oct. 2025, o-RAN Working Group 1, Technical Specification R004. [Online]. Available: https: //orandownloadsweb.azurewebsites.net/specifications [54] “Network Functions Virtualisation (NFV); Report on Network Slicing Support with ETSI NFV Architecture Framework,” ETSI, Tech. Rep. ETSI GR NFV-EVE 012 V3.1.1, Dec. 2017. [Online]. Available: https://www.etsi.org/deliver/etsi_gr/NFV-EVE/001_099/012/ 03.01.01_60/gr_nfv-eve012v030101p.pdf [55] “Network Functions Virtualisation (NFV); Management and Orchestration,” ETSI, Tech. Rep. ETSI GS NFV-MAN 001 V1.1.1, 2014. [Online]. Available: https://www.etsi.org/deliver/etsi_gs/ NFV-MAN/001_099/001/01.01.01_60/gs_nfv-man001v010101p.pdf [56] R. Mijumbi et al., “Network Function Virtualization: State-of-the-Art and Research Challenges,” IEEE Communications Surveys & Tutorials, vol. 18, no. 1, pp. 236–262, 2016. [57] P. Soto et al., “Network Intelligence for NFV Scaling in Closed-Loop Architectures,” IEEE Communications Magazine, vol. 61, no. 6, pp. 66–72, Jun. 2023. [58] T. Usländer et al., “Symbiotic Evolution of Digital Twin Systems and Dataspaces,” Automation, vol. 3, no. 3, pp. 378–399, Aug. 2022. [59] “IDS Reference Architecture Model 4.0,” International Data Spaces Association (IDSA), Tech. Rep., 2022. [Online]. Available: https://internationaldataspaces.org/ [60] 3GPP, “Management and orchestration; 5G performance measurements,” 3GPP, TS 28.552 V18.0.0 (2022-09), 2022. [61] 3GPP, “5G;NG-RAN; Architecture description,” 3GPP, Technical Specification (TS) 38.401, 2020. [Online]. Available: https://www.etsi.org/deliver/etsi_ts/138400_138499/138401/16. 02.00_60/ts_138401v160200p.pdf [62] ITU-T, “Machine learning marketplace integration in future networks including IMT-2020,” ITU, Recommendation Y.3176, Feb. 2020. [Online]. Available: https://www.itu.int/rec/T-REC-Y.3176-202009-I/en [63] “3GPP TS 32.401 version 17.0.0 Release 17; Telecommunication management; Performance Management (PM); Concept and requirements,” Tech. Rep. [64] 3GPP TS 28.541, “5G Network Resource Model (NRM); Stage 2 and stage 3 (Release 16),” Tech. Rep. V16.7.0, Dec. 2020. [65] 3GPP TS 28.622, “Generic Network Resource Model (NRM) Integration Reference Point (IRP) (Release 18),” Tech. Rep., Mar. 2024. [66] T. ISO, “184/SC 4; Automation Systems and Integration-Digital Twin Framework for Manufacturing—Part 1: Overview and General Principles,” Tech. Rep., 2021. [67] F. Tao et al., “Digital twin modeling,” Journal of Manufacturing Systems, vol. 64, pp. 372–389, Jul. 2022. [68] J. Wen et al., “Toward Digital Twin Oriented Modeling of Complex Networked Systems and Their Dynamics: A Comprehensive Survey,” IEEE Access, vol. 10, pp. 66 886–66 923, 2022. [69] J. Akroyd et al., “Universal Digital Twin - A Dynamic Knowledge Graph,” Data-Centric Engineering, vol. 2, p. e14, 2021. 40 [70] R. Verdecchia et al., “Designing a Future-Proof Reference Architecture for Network Digital Twins,” in European Conference on Software Architecture, Springer. Springer, 2024, pp. 287–306. [71] H. Hong et al., “NetGraph: An Intelligent Operated Digital Twin Platform for Data Center Networks,” in Proceedings of the ACM SIGCOMM 2021 workshop on network-application integration. ACM, Aug. 2021, pp. 26–32. [72] S. Gil et al., “Survey on open-source digital twin frameworks–A case study approach,” Software: Practice and Experience, vol. 54, no. 6, pp. 929–960, Jan. 2024. [73] A. Meijers, Hands-On Azure Digital Twins: A practical guide to building distributed IoT solutions. Packt Publishing Ltd, 2022. [74] K. Schweichhart, “Reference architectural model industrie 4.0 (rami 4.0) – An Introduction,” talk, 2016. [75] J. Robles et al., “OpenTwins: An open-source framework for the development of next-gen compositional digital twins,” Computers in Industry, vol. 152, p. 104007, Nov. 2023. [76] U. Ahle and J. J. Hierro, “FIWARE for Data Spaces,” in Designing Data Spaces. Springer, 2022, pp. 395–417. [77] T. Hartmann et al., “GreyCat: Efficient what-if analytics for data in motion at scale,” Information Systems, vol. 83, pp. 101–117, Jul. 2019. [78] F. Abbasi et al., “Understanding Semantic Drift in Model Driven Digital Twins,” in Proceedings of the ACM/IEEE 27th International Conference on Model Driven Engineering Languages and Systems. ACM, Sep. 2024, pp. 419–430. [79] I. Alfonso et al., “Building BESSER: An Open-Source Low-Code Platform,” in Enterprise, Business-Process and Information Systems Modeling. Springer, 2024, pp. 203–212. [80] X. Lin et al., “6G Digital Twin Networks: From Theory to Practice,” IEEE Communications Magazine, vol. 61, no. 11, pp. 72–78, Nov. 2023. [81] P. Almasan et al., “Network Digital Twin: Context, Enabling Technologies, and Opportunities,” IEEE Communications Magazine, vol. 60, no. 11, pp. 22–27, Nov. 2022. [82] L. Hui et al., “Digital Twin for Networking: A Data-Driven Performance Modeling Perspective,” IEEE Network, vol. 37, no. 3, pp. 202–209, May 2023. [83] N. Chukhno et al., “Models, methods, and solutions for multicasting in 5G/6G mmWave and sub-THz systems,” IEEE Communications Surveys & Tutorials, vol. 26, no. 1, pp. 119–159, 2024. [84] A. Boudi et al., “AI-Based Resource Management in Beyond 5G Cloud Native Environment,” IEEE Network, vol. 35, no. 2, pp. 128–135, Mar. 2021. [85] A. Micu et al., “Modelling Probabilistic Digital Twins of Complex Inland Waterway Transportation Systems Using Bayesian Networks,” in Transport Research Arena Conference. Springer, Oct. 2025, pp. 874–879. [86] T. Tripura et al., “Probabilistic machine learning based predictive and interpretable digital twin for dynamical systems,” Computers & Structures, vol. 281, p. 107008, Jun. 2023. [87] M. Segovia and J. Garcia-Alfaro, “Design, Modeling and Implementation of Digital Twins,” Sensors, vol. 22, no. 14, p. 5396, Jul. 2022. [88] R. F. Nielsen, “Hybrid Modelling Strategies for Development of Digital Twins,” Ph.D. dissertation, 2021. [89] F. Bari et al., “Orchestrating Virtualized Network Functions,” IEEE Transactions on Network and Service Management, vol. 13, no. 4, pp. 725–739, Dec. 2016. [90] D. De Vleeschauwer et al., “A Method to Compare Scaling Algorithms for Cloud-Based Services,” IEEE Transactions on Cloud Computing, vol. 13, no. 1, pp. 34–45, Jan. 2025. [91] J. Jimenez Agudelo et al., “Resource Allocation of Multi-user Workloads in Cloud and Edge Data-Centers Using Reinforcement Learning,” 2023. [Online]. Available: https://zenodo.org/doi/10.5281/zenodo.10007880 [92] P. Soto et al., “Towards Autonomous VNF Auto-scaling using Deep Reinforcement Learning,” in 2021 Eighth International Conference on Software Defined Systems (SDS). IEEE, Dec. 2021, pp. 01–08. [93] N. Slamnik-Kriještorac et al., “AI-Empowered Management and Orchestration of Vehicular Systems in the Beyond 5G Era,” IEEE Network, vol. 37, no. 4, pp. 305–313, Jul. 2023. [94] L. Gu et al., “Intelligent VNF Orchestration and Flow Scheduling via Model-Assisted Deep Reinforcement Learning,” IEEE Journal on Selected Areas in Communications, vol. 38, no. 2, pp. 279–291, Feb. 2020. [95] E. Amiri et al., “Edge-AI Empowered Dynamic VNF Splitting in ORAN Slicing: A Federated DRL Approach,” IEEE Communications Letters, vol. 28, no. 2, pp. 318–322, Feb. 2024. [96] J. C. Cisneros et al., “Coordination algorithm for migration of shared vnfs in federated environments,” in 2020 6th IEEE Conference on Network Softwarization (NetSoft). IEEE, Jun. 2020, pp. 252–256. [97] T.-M. Pham and H.-N. Chu, “Multi-Provider and Multi-Domain Resource Orchestration in Network Functions Virtualization,” IEEE Access, vol. 7, pp. 86 920–86 931, 2019. [98] Z. Wei et al., “Integrated Sensing and Communication Driven Digital Twin for Intelligent Machine Network,” IEEE Internet of Things Magazine, vol. 7, no. 4, pp. 60–67, Jul. 2024. [99] S. Alikhani and A. Alkhateeb, “Digital Twin Aided RIS Communication: Robust Beamforming and Interference Management,” in 2024 IEEE 100th Vehicular Technology Conference (VTC2024-Fall). IEEE, Oct. 2024. [100] B. Salehi et al., “Multiverse at the Edge: Interacting Real World and Digital Twins for Wireless Beamforming,” IEEE/ACM Transactions on Networking, vol. 32, no. 4, pp. 3092–3110, Aug. 2024. [101] S. Jiang and A. Alkhateeb, “Digital Twin Based Beam Prediction: Can We Train in the Digital World and Deploy in Reality?” in 2023 IEEE International Conference on Communications Workshops (ICC Workshops). IEEE, May 2023. [102] N. Zeulin et al., “ML-Assisted Beam Selection via Digital Twins for Time-Sensitive Industrial IoT,” IEEE Internet of Things Magazine, vol. 5, no. 1, pp. 36–40, Mar. 2022. [103] M. Masoudi et al., “Digital Twin Assisted Risk-Aware Sleep Mode Management Using Deep Q-Networks,” IEEE Transactions on Vehicular Technology, vol. 72, no. 1, pp. 1224–1239, Jan. 2023. [104] M. Dryjanski, “Quality of Service-based Resource Allocator xApp (QRA-xApp),” Rimedo Labs, Technical Specification, 2022. [105] B. Németh et al., “Delay and reliability-constrained VNF placement on mobile and volatile 5G infrastructure,” IEEE Transactions on Mobile Computing, vol. 21, no. 9, pp. 3150–3162, Sep. 2022. [106] A. De Domenico et al., “Optimal Virtual Network Function Deployment for 5G Network Slicing in a Hybrid Cloud Infrastructure,” IEEE Transactions on Wireless Communications, vol. 19, no. 12, pp. 7942– 7956, Dec. 2020. [107] A. M. Hafez et al., “Modified Ant Colony Placement Algorithm for Containers,” in Proceedings of the 15th International Conference on Computer Engineering and Systems (ICCES). Cairo, Egypt: IEEE, Dec. 2020, pp. 1–6. [108] R. Zhang et al., “A Genetic Algorithm-Based Energy-Efficient Container Placement Strategy in CaaS,” IEEE Access, vol. 7, pp. 121 360–121 373, 2019. [109] R. Solozabal et al., “Virtual Network Function Placement Optimization With Deep Reinforcement Learning,” IEEE Journal on Selected Areas in Communications, vol. 38, no. 2, pp. 292–303, Feb. 2020. [110] J. Kim et al., “Deep Reinforcement Learning based Cloud-native Network Function Placement in Private 5G Networks,” in Proceedings of the 2020 IEEE Globecom Workshops (GC Wkshps). IEEE, Dec. 2020. [111] M. Goudarzi et al., “A Distributed Deep Reinforcement Learning Technique for Application Placement in Edge and Fog Computing Environments,” IEEE Transactions on Mobile Computing, vol. 22, no. 5, pp. 2491–2505, May 2023. [112] Y. Liu and J. Zhang, “Service Function Chain Embedding Meets Machine Learning: Deep Reinforcement Learning Approach,” IEEE Transactions on Network and Service Management, vol. 21, no. 3, pp. 3465–3481, Jun. 2024. [113] A. F. Ocampo et al., “Optimal Service Function Chain Composition in Network Functions Virtualization,” in Security of Networks and Services in an All-Connected World: 11th IFIP WG 6.6 International Conference on Autonomous Infrastructure, Management, and Security (AIMS 2017). Zurich, Switzerland: Springer, Jul. 2017. [114] X. Wang et al., “A Graph Neural Network and Pointer Network-Based Approach for QoS-Aware Service Composition,” IEEE Transactions on Services Computing, vol. 16, no. 3, pp. 1589–1603, May 2023. [115] X. Ren et al., “DeepQSC: a GNN and Attention Mechanism-based Framework for QoS-aware Service Composition,” in Proceedings of the 2021 International Conference on Service Science (ICSS). Xi’an, China: IEEE, May 2021, pp. 76–83. [116] R. K. Gupta and R. Misra, “Machine Learning-based Slice allocation Algorithms in 5G Networks,” in 2019 International Conference on Advances in Computing, Communication and Control (ICAC3). IEEE, Dec. 2019. [117] M. H. Abidi et al., “Optimal 5G network slicing using machine learning and deep learning concepts,” Computer Standards & Interfaces, vol. 76, p. 103518, Jun. 2021. 41 [118] A. Thantharate et al., “Secure5G: A Deep Learning Framework Towards a Secure Network Slicing in 5G and Beyond,” in 2020 10th Annual Computing and Communication Workshop and Conference (CCWC). IEEE, Jan. 2020. [119] A. Prado et al., “Enhanced Mobility Management with SD-RAN in 5G Networks,” in Proceedings of the IFIP Networking Conference (IFIP Networking). IEEE, Jun. 2024, pp. 159–167. [120] N. Saha et al., “MonArch: Network Slice Monitoring Architecture for Cloud Native 5G Deployments,” in Proceedings of IEEE/IFIP NOMS 2023. IEEE, May 2023. [121] M. Mekki et al., “A Scalable Monitoring Framework for Network Slicing in 5G and Beyond Mobile Networks,” IEEE Transactions on Network and Service Management, vol. 19, no. 1, pp. 413–423, Mar. 2022. [122] G. K. Pavani and B. Veeramallu, “Hybrid Machine Learning Framework for Anomaly Detection in 5G Networks,” Journal of Information Systems Engineering and Management, vol. 32, 2025. [123] S. Chennoufi et al., “SoK: Federated Learning based Network Intrusion Detection in 5G: Context, State of the Art and Challenges,” in Proceedings of ARES ’24: 19th International Conference on Availability, Reliability and Security. ACM, Jul. 2024. [124] B. Farzaneh et al., “DTL-IDS: Deep Transfer Learning-based Intrusion Detection System in 5G Networks,” in Proceedings of the 19th International Conference on Network and Service Management (CNSM). IEEE, Oct. 2023, pp. 1–5. [125] A. Dhar Dwivedi et al., “Blockchain and artificial intelligence for 5Genabled Internet of Things: Challenges, opportunities, and solutions,” Transactions on Emerging Telecommunications Technologies, vol. 35, no. 4, Jul. 2021. [126] B. R. Chirra, “Dynamic Cryptographic Solutions for Enhancing Security in 5G Networks,” International Journal of Advanced Engineering Technologies and Innovations, vol. 1, no. 3, 2022. [127] C. Sommer et al., “Bidirectionally Coupled Network and Road Traffic Simulation for Improved IVC Analysis,” IEEE Transactions on Mobile Computing, vol. 10, no. 1, pp. 3–15, Jan. 2011. [128] D. S. Buse et al., “Bridging Worlds: Integrating Hardware-in-the-Loop Testing with Large-Scale VANET Simulation,” in 2018 14th Annual Conference on Wireless On-demand Network Systems and Services (WONS). IEEE, Feb. 2018, pp. 33–36. [129] ——, “Demo Abstract: Integrating a Driving Simulator with City-Scale VANET Simulation for the Development of next Generation ADAS Systems,” in IEEE INFOCOM 2018 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS). IEEE, Apr. 2018, pp. 1–2. [130] D. S. Buse and F. Dressler, “Towards Real-Time Interactive V2X Simulation,” in 2019 IEEE Vehicular Networking Conference (VNC). IEEE, Dec. 2019, pp. 1–8. [131] D. S. Buse et al., “Accelerating the Simulation of Wireless Communication Protocols Using Asynchronous Parallelism,” in Proceedings of the 24th International ACM Conference on Modeling, Analysis and Simulation of Wireless and Mobile Systems. ACM, Nov. 2021, pp. 57–66. [132] M. M. G. Farag et al., “INTEGRATION Large-Scale Modeling Framework of Direct Cellular Vehicle-to-All (C-V2X) Applications,” Sensors, vol. 21, no. 6, p. 2127, Mar. 2021. [133] T. Queck et al., “Realistic Simulation of V2X Communication Scenarios,” in 2008 IEEE Asia-Pacific Services Computing Conference. IEEE, Dec. 2008, pp. 1623–1627. [134] R. Protzmann et al., “Extending the V2X Simulation Environment VSimRTI with Advanced Communication Models,” in 2012 12th International Conference on ITS Telecommunications. IEEE, Nov. 2012, pp. 683–688. [135] N. Goebel et al., “Coupled Simulation of Mobile Cellular Networks, Road Traffic and V2X Applications Using Traces,” in 2016 IEEE International Conference on Communications (ICC). IEEE, 2016, pp. 1–7. [136] D. Cui et al., “Extensible Co-Simulation Framework for Supporting Cooperative Driving Automation Research,” Transportation Research Record: Journal of the Transportation Research Board, vol. 2677, no. 3, pp. 1067–1079, Sep. 2022. [137] K. Schrab et al., “A Large-Scale Traffic Scenario of Berlin for Evaluating Smart Mobility Applications,” in Smart Energy for Smart Transport, E. G. Nathanail et al., Eds. Springer Nature Switzerland, 2023, pp. 276–287. [138] T. H. Pritom et al., “Towards a Generic Middleware-Based Interface for Mobile Communication Simulation,” in 29th International Symposium on Distributed Simulation and Real Time Applications (DS-RT 2025). Prague, Czechia: IEEE, Sep. 2025. [139] T. Hardes et al., “Towards an Open Source Fully Modular Multi Unmanned Aerial Vehicle Simulation Framework,” in 2022 IEEE 42nd International Conference on Distributed Computing Systems Workshops (ICDCSW). IEEE, Jul. 2022, pp. 284–289. [140] M. Yin et al., “Demo: SkyRoute, a Fast and Realistic UAV Cellular Simulation Framework,” in 2021 IEEE 29th International Conference on Network Protocols (ICNP). IEEE, Nov. 2021, pp. 1–2. [141] A. Al-Mousa et al., “UTSim: A Framework and Simulator for UAV Air Traffic Integration, Control, and Communication,” International Journal of Advanced Robotic Systems, vol. 16, no. 5, Sep. 2019. [142] G. Castellanos et al., “Performance Evaluation of Direct-Link Backhaul for UAV-Aided Emergency Networks,” Sensors, vol. 19, no. 15, p. 3342, Jul. 2019. [143] ——, “Evaluation of Flying Caching Servers in UAV-BS Based Realistic Environment,” Vehicular Communications, vol. 32, p. 100390, Dec. 2021. [144] ——, “Evaluation of Beamsteering Performance in MultiuserMIMO Unmanned Aerial Base Stations Networks,” IEEE Access, vol. 10, pp. 62 565–62 580, 2022. [145] M. Kudelski et al., “RoboNetSim: An Integrated Framework for MultiRobot and Network Simulation,” Robotics and Autonomous Systems, vol. 61, no. 5, pp. 483–496, May 2013. [146] J. L. F. Pereira and R. J. F. Rossetti, “An Integrated Architecture for Autonomous Vehicles Simulation,” in Proceedings of the 27th Annual ACM Symposium on Applied Computing. ACM, Mar. 2012, pp. 286– 292. [147] F. A. Schiegg et al., “A Novel Simulation Framework for the Design and Testing of Advanced Driver Assistance Systems,” in 2019 IEEE 90th Vehicular Technology Conference (VTC2019-Fall). IEEE, Sep. 2019, pp. 1–6. [148] A. Ibrahim et al., “Co-simulation Framework for Control, Communication and Traffic for Vehicle Platoons,” in 2018 21st Euromicro Conference on Digital System Design (DSD). IEEE, Aug. 2018, pp. 352–356. [149] A. Choudhury et al., “An Integrated Simulation Environment for Testing V2X Protocols and Applications,” Procedia Computer Science, vol. 80, pp. 2042–2052, 2016. [150] D. Krajzewicz et al., “Simulation of V2X Applications with the iTETRIS System,” Procedia - Social and Behavioral Sciences, vol. 48, pp. 1482– 1492, 2012. [151] M. R. Cantas and L. Guvenc, “Customized Co-Simulation Environment for Autonomous Driving Algorithm Development and Evaluation.” [152] D. Nalic et al., “Development of a Co-Simulation Framework for Systematic Generation of Scenarios for Testing and Validation of Automated Driving Systems,” in 2019 IEEE Intelligent Transportation Systems Conference (ITSC). IEEE, Oct. 2019, pp. 1895–1901. [153] ——, “Design and Implementation of a Co-Simulation Framework for Testing of Automated Driving Systems,” Sustainability, vol. 12, no. 24, p. 10476, Dec. 2020. [154] S. Joerer et al., “Enabling Situation Awareness at Intersections for IVC Congestion Control Mechanisms,” IEEE Transactions on Mobile Computing, vol. 15, no. 7, pp. 1674–1685, Jul. 2016. [155] F. Malandrino et al., “Content Downloading in Vehicular Networks: Bringing Parked Cars Into the Picture,” in 23rd IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 2012). Sydney, Australia: IEEE, Sep. 2012, pp. 1534–1539. [156] F. Hagenauer et al., “A Simulator for Heterogeneous Vehicular Networks,” in 6th IEEE Vehicular Networking Conference (VNC 2014), Poster Session. Paderborn, Germany: IEEE, Dec. 2014, pp. 185–186. [157] A. Hegde and A. Festag, “Artery-C: An OMNeT++ Based Discrete Event Simulation Framework for Cellular V2X,” in Proceedings of the 23rd International ACM Conference on Modeling, Analysis and Simulation of Wireless and Mobile Systems. ACM, Nov. 2020, pp. 47–51. [158] B. McCarthy and A. O’Driscoll, “OpenCV2X Mode 4: A Simulation Extension for Cellular Vehicular Communication Networks,” in 2019 IEEE 24th International Workshop on Computer Aided Modeling and Design of Communication Links and Networks (CAMAD). IEEE, Sep. 2019, pp. 1–6. [159] M. K. Müller et al., “Flexible Multi-Node Simulation of Cellular Mobile Communications: The Vienna 5G System Level Simulator,” EURASIP Journal on Wireless Communications and Networking, vol. 2018, no. 1, Sep. 2018. [160] M. Franke and C. Sommer, “Toward Space-Air-Ground Integrated Network Simulation with 4D Topologies,” in 19th IEEE/IFIP Conference