The target outcomes include:

• A curated collection of high-quality real-world datasets, that have enough quality to train AI models that can run in operations, capturing realistic deployment scenarios—including user density, mobility patterns, network conditions, and realistic data traffic patterns from various applications, including vertical applications (non-exhaustive list). The collection could be obtained by any means or tools indicatively including but not limited to operational networks (real data) and/or network digital twins or advanced (high TRL) experimental platforms (emulators) or trials, gen AI (augmenting field measurements), and be significant in scale and broad representativeness (in terms of technologies, use-cases and verticals) designed to ensure scalability, providing a robust foundation for the training and validation of AI models for 6G networks and for AIaaS.

• An open-source simulator to create synthetic data sets (correlated between different network layers, and across different network points) of high quality capturing realistic deployment scenarios—including user density, mobility patterns, network conditions, anomalies and network attacks and realistic data traffic patterns from various applications, including vertical applications, etc.

• A modular framework and related methodology for generating and processing (e.g., cleaning, preprocessing) high quality realistic synthetic datasets to train AI models usable by 6G systems.

• A data space of appropriate scale to manage the datasets foreseen in the first expected outcome, covering the full data lifecycle for managing datasets, produced by SNS JU projects, for different reference use-cases at different scales. The data space should follow a framework that supports data/metadata sharing and governance within the SNS JU ecosystem, facilitating collaboration and interoperability, ensuring data sovereignty, privacy, security and compliance with EU regulations. Existing solutions for managing SNS produced data sets including repositories should be considered to avoid duplication with existing initiatives.

• A framework to audit any synthetic data sets that the project will create so as to ensure the validity/credibility of produced datasets.

• Activities that will encourage widespread use of the dataset (e.g., the integration with CAMARA / Network APIs / agents as a means to expose / enrich / monetise datasets, or the design of new control plane functions) and the overall framework by the SNS community, other EC initiatives and eventually by standardization bodies.

The produced datasets are targeted to be used by the SNS community to train AI models that will improve the performance of 6G networks or serve to develop AI solutions for 6G services and applications (AIaaS) for SNS JU verticals.

Objective:

Please refer to the "Specific Challenges and Objectives" section for Stream B-01 in the Work Programme, available under ‘Topic Conditions and Documents - Additional Documents’.

Scope:

The focus is on:

• Collecting, and making available in an operationally effective way, high-quality real-world datasets from advanced (e.g., 5G Advanced, 6G) operational networks (real data) or network digital twins and/or advanced (high TRL), gen AI tools, experimental platforms (emulators) or trials, capturing realistic deployment scenarios—including user density, mobility patterns, network conditions, and realistic data traffic patterns from various applications, including verticals. Proponents should not consider datasets that have limitations in terms of, for example, coverage, capacity, number of devices, or where data and metadata do not have enough quality to train AI models. The datasets should originate from actors that have significant experience from operational networks and network components as well as service providers (including verticals), so that there is strong level of confidence that the datasets are useful for training AI models to be used by 6G networks. For the real-world datasets from advanced (e.g., beyond 5G, 6G) operational networks and/or advanced (high TRL) experimental platforms or trials and datasets produced by SNS JU projects this project should ensure the provision the presence of Metadata definition (to have a common descriptor for the data), methods to verify that the data is valuable for the training of realistic AI models, and ensure data reusability.

• Development of full protocol stack, end-to-end implementations enabling high-fidelity, system-level simulations of 6G networks building upon and enhancing existing open-source simulators for producing correlated reference datasets at different network layers, and across different network points. These simulations will span from the physical layer to the application layer and will include:

- Support for multi-radio access technologies, including e.g., cellular, Wi-Fi, and non-terrestrial systems.

- Disaggregated RAN architectures, enabling flexible and scalable deployment models.

- Multi-band operation with accurate propagation and channel modelling across various frequency ranges (e.g., FR3, mmWave, cmWave, sub-THz), incorporating ray tracing-based channel models for enhanced realism.

- Realistic traffic patterns that reflect anticipated data flows in future network scenarios.

- Anomalies and network attacks that can be eventually used to test the resilience of AI solutions in 6G networks.

• Design of an open-source framework and toolset for generating high quality realistic synthetic data, tailored to diverse environmental scenarios (e.g., urban, suburban, rural, indoor, industrial), user densities, security threats, mobility patterns, and node behaviour - including memory, CPU, storage, and energy consumption - as well as traffic profiles from a variety of vertical applications. These datasets should be produced following existing calibration directives from standardization bodies (e.g., 3GPP) and expected traffic patterns from European and international organizations (e.g., 5GAA, 5G-ACIA, etc.). The datasets should be validated from proponents that have significant experience from operational networks and network components as well as service providers (including verticals), so that there is strong level of confidence that the datasets are useful for training AI models to be used by 6G networks and for AIaaS. Optionally the tool may consider the use of reliable LLM solutions to enable a user-friendly interface for users and/or to calibrate the simulator and/or create the desired datasets.

• Creation of large-scale, open-source high quality synthetic datasets, following well established reference use-cases, containing measurements, channel and network indicators, and performance metrics across multiple protocol layers (RF, physical, MAC, network, transport, and application). These datasets will cover a broad range of network scenarios, architectures, technologies, and system configurations related to smart networks and services.

• Validation, quality assessment of the existing SNS JU project datasets, verifying the data’s accuracy, consistency, and completeness ensuring their alignment with the specific use case and performance requirements of the 6G network.

• For the real-world datasets from advanced (e.g., 5G Advanced, 6G) operational networks and/or advanced (high TRL) experimental platforms or trials and datasets produced by SNS JU project, this project should ensure the provision the presence of: Metadata definition (to have a common descriptor for the data), methods to verify that the data is valuable for the training of realistic AI models and ensure data reusability.

• Engagement with standardization bodies and relevant open-source communities to promote the adoption of the framework, the associated simulation tools, and open datasets enabling industry-wide collaboration on shared software platforms and data resource.

• Creation of tutorials and implementation of dissemination activities to encourage widespread use of the framework, its synthetic datasets, and the underlying simulation tools including the development of APIs, new intelligence control plane function, etc.

This Topic expects proposals with strong industrial participation with demonstrated AI and operational expertise to ensure credibility, usability, and engagement with standardisation bodies. Academic institutions and RTOs will complement consortia where their expertise adds clear value.