Periodic Reporting for period 1 - 6G-XCEL (6G Trans-Continental Edge Learning)
Berichtszeitraum: 2024-01-01 bis 2025-06-30
The 6G-XCEL project is advancing the foundations of next-generation communication networks by creating a new framework for AI-native, decentralized, and secure 6G systems. Future networks will need to manage enormous volumes of data, integrate wireless and optical technologies, and support a wide range of critical applications—from smart transport and industrial automation to immersive services and global connectivity.
To address these challenges, 6G-XCEL is developing a Decentralized Multi-party, Multi-network AI (DMMAI) framework, to coordinate controls across radio and optical networks. This framework promotes global validation, standardization, and sustainable AI-based network solutions. It also supports the creation of reference use cases, data repositories, curated training datasets, and benchmarking platforms for AI/ML solutions in 6G networks.
To achieve this, 6G-XCEL unites EU and US researchers to validate and integrate the DMMAI framework into various testbeds and programs that will accelerate its adoption and contribute to global standardization.
To address these challenges, 6G-XCEL is developing a Decentralized Multi-party, Multi-network AI (DMMAI) framework, to coordinate controls across radio and optical networks. This framework promotes global validation, standardization, and sustainable AI-based network solutions. It also supports the creation of reference use cases, data repositories, curated training datasets, and benchmarking platforms for AI/ML solutions in 6G networks.
To achieve this, 6G-XCEL unites EU and US researchers to validate and integrate the DMMAI framework into various testbeds and programs that will accelerate its adoption and contribute to global standardization.
The 6G-XCEL project has advanced the design and validation of the Decentralized Multi-Party, Multi-Network AI (DMMAI) framework, a foundation for AI-native 6G networks. At its core is an AI Controller (AIC), a modular software system that unifies radio and optical domains, supports cross-controller communication, and enables end-to-end orchestration, monitoring, and self-repair.
The framework uses a distributed AIC architecture, allowing multiple algorithms to run in parallel and remain synchronized across different vendors and networks. Tests on experimental datasets provide initial indications of the benefits of AI-driven control and its potential for achieving faster and more reliable performance compared to conventional approaches.
Progress was made in security and trust aspects through a comprehensive threat analysis that identified risks such as data poisoning and privacy breaches so that they can be addressed. New modules were developed for data integrity, model validation, and privacy-preserving governance. The project also advanced novel AI techniques for use in the DMMAI context, including transfer learning, automated tuning, and explainability methods, making AI models more adaptable and transparent.
Work on energy efficiency introduced models to evaluate trade-offs between energy use, latency, and AI accuracy. Lightweight AI algorithms and optimized communication protocols were investigated to guide the design of sustainable AI-native networks
Key achievements include:
Demonstrating that AI-driven network control improves performance by 20–30%, with faster application response times and higher throughput compared to conventional approaches.
Developing mechanisms for synchronizing applications across multiple controllers, ensuring seamless service delivery.
Identifying security and privacy safeguards for distributed AI, including protections against data poisoning, backdoor models, and privacy breaches.
Advancing explainable AI techniques to build trust and transparency into network decision-making.
Establishing end to end network energy efficiency models that allow operators to optimize performance while minimizing energy consumption.
The framework uses a distributed AIC architecture, allowing multiple algorithms to run in parallel and remain synchronized across different vendors and networks. Tests on experimental datasets provide initial indications of the benefits of AI-driven control and its potential for achieving faster and more reliable performance compared to conventional approaches.
Progress was made in security and trust aspects through a comprehensive threat analysis that identified risks such as data poisoning and privacy breaches so that they can be addressed. New modules were developed for data integrity, model validation, and privacy-preserving governance. The project also advanced novel AI techniques for use in the DMMAI context, including transfer learning, automated tuning, and explainability methods, making AI models more adaptable and transparent.
Work on energy efficiency introduced models to evaluate trade-offs between energy use, latency, and AI accuracy. Lightweight AI algorithms and optimized communication protocols were investigated to guide the design of sustainable AI-native networks
Key achievements include:
Demonstrating that AI-driven network control improves performance by 20–30%, with faster application response times and higher throughput compared to conventional approaches.
Developing mechanisms for synchronizing applications across multiple controllers, ensuring seamless service delivery.
Identifying security and privacy safeguards for distributed AI, including protections against data poisoning, backdoor models, and privacy breaches.
Advancing explainable AI techniques to build trust and transparency into network decision-making.
Establishing end to end network energy efficiency models that allow operators to optimize performance while minimizing energy consumption.
6G-XCEL has advanced the state of the art by showing how existing AI/ML techniques can be effectively embedded into the AI Controller (AIC) platform, rather than creating entirely new algorithms. The project focused on integrating widely used models such as XGBoost, Random Forest, Decision Trees, and k-Nearest Neighbour into the modified DMMAI framework and testing their impact on key performance metrics.
An important outcome is the confirmation that embedding AI/ML in AICs provides clear and measurable benefits without requiring the development of new models from scratch. This demonstrates the scalability and flexibility of the DMMAI framework, as additional AI/ML models can be integrated depending on available datasets, features, and targeted performance goals.
The next stage is to extend these findings to real-world testbeds, where O-RAN units can be managed by fully programmed AI/ML engines. This will validate the framework in operational conditions and provide stronger evidence for industry adoption. Taken together, these results represent a major step forward for AI-native 6G networks, demonstrating how distributed intelligence can significantly improve efficiency, reliability, and scalability compared to today’s approaches.
6G-XCEL marks a significant step forward in the evolution of 6G networks by delivering the first holistic framework for decentralized, multi-party AI—spanning radio and optical networks. It demonstrates how advanced AI techniques can be applied directly to network control end to end, increasing adaptability, efficiency, and trust. The project also establishes a methodology for measuring and improving energy efficiency, introduces privacy-preserving and secure AI methods, and lays the foundation for cross-domain interoperability between radio and optical networks.
An important outcome is the confirmation that embedding AI/ML in AICs provides clear and measurable benefits without requiring the development of new models from scratch. This demonstrates the scalability and flexibility of the DMMAI framework, as additional AI/ML models can be integrated depending on available datasets, features, and targeted performance goals.
The next stage is to extend these findings to real-world testbeds, where O-RAN units can be managed by fully programmed AI/ML engines. This will validate the framework in operational conditions and provide stronger evidence for industry adoption. Taken together, these results represent a major step forward for AI-native 6G networks, demonstrating how distributed intelligence can significantly improve efficiency, reliability, and scalability compared to today’s approaches.
6G-XCEL marks a significant step forward in the evolution of 6G networks by delivering the first holistic framework for decentralized, multi-party AI—spanning radio and optical networks. It demonstrates how advanced AI techniques can be applied directly to network control end to end, increasing adaptability, efficiency, and trust. The project also establishes a methodology for measuring and improving energy efficiency, introduces privacy-preserving and secure AI methods, and lays the foundation for cross-domain interoperability between radio and optical networks.