Réunion

09h00

Welcome coffee

09h30

Introduction by organizers (Eric Hardouin, Orange, and Salah El Ayoubi, L2S/CentraleSupélec)

9h40

Opening by Jean-Benoît Besset, EVP for Orange Group Environment and Energy Transition

9h55 

Invited talk by Berna Sayrac, Senior research engineer on value-based sustainable networks, Orange.

« Network Design-to-Impact : Sustainable Innovation and Digital Responsibility in Future Telecommunications ».

10h25 

Talk by Cedric Thienot, Chief Technology Officer at Firecell.

« xApps for energy consumption optimization in 5G networks ».

10h45 

Coffee break

11h15

Invited talk by Maziar Nekovee, Professor, Managing Director 6G Netzero Lab, University of Sussex.

« Intertwining Telecom and Energy Networks for a Green and Sustainable Future ».

11h45 

Talk by Asmaa Amer, ETIS Lab, CY Cergy Paris Université / ENSEA.

« Maximizing Energy Efficiency and Localization Accuracy in Pinching Antennas-Assisted sub-THz Systems ».

12h05

Talk by Mata Khalili, Robotics Research Scientist at Nokia Bell Labs

« Application-Driven Sustainability: The Example of Cloud-Connected Robots ».

12h25

Lunch

13h25 

Poster session.

Posters by Houssam Hajj Hassan (Orange), Nogbou Bénédicte Kongo (Orange), Ryad Madi (CentraleSupélec), Younes Mehloul (Orange/CentraleSupélec), Idriss Merah (Orange/CentraleSupélec), Yanting Pan (CentraleSupélec), Duy Trong (CentraleSupélec), Jiyayi Wi (Orange/CentraleSupélec)

14h10

Invited talk, Cicek Cavdar, Professor at KTH Royal Institute of Technology

« AI-Native Green Networks: From Intelligent Base Stations to Cloud-Integrated Cell-Free Architectures ».

14h40

Invited talk by Illyyne Saffar and Aurélie Boisbunon, AI Specialists at Ericsson.

« Energy-Aware AI: Designing Sustainable and Green AI Models for 5G and 6G Networks ».

15h10

Coffee break

15h40

Talk by Hamza Abbar, CentraleSupélec/Orange.

« Forecast-Driven RAN Energy Control under Telemetry Lag ».

16h00

Talk by Marc Pierre, CentraleSupélec

Optimal control and online learning for a variable service rate discrete queue.

16h20

Conclusion by organizers

Abstracts

• Berna Sayrac,« Network Design-to-Impact : Sustainable Innovation and Digital Responsibility in Future Telecommunications ».

Orange’s Network Design-to-Impact (NDI) framework proposes a value-driven, sustainability-first approach to future network design, balancing environmental, social, and economic constraints with traditional technical performance. By optimizing networks for maximum positive impact and minimal negative effects—using levers like energy efficiency, equipment lifetime extension, and circular economy principles—NDI aims to guide the telecommunications industry toward networks that deliver environmental, societal and economic value while addressing trade-offs between sustainability and technical objectives.

• Cedric Thienot, « xApps for energy consumption optimization in 5G networks ».

TBA

• Maziar Nekovee, « Intertwining Telecom and Energy Networks for a Green and Sustainable Future ».

TBA

• Asmaa Amer, « Maximizing Energy Efficiency and Localization Accuracy in Pinching Antennas-Assisted sub-THz Systems ».

The sub-THz band is a promising candidate for future 6G systems due to its potential to enable high data rates. However, the sub-THz band still suffers from severe path loss and molecular absorption, limiting both coverage and localization accuracy. Recently, pinching antenna systems (PASS) have been proposed as a promising low-complexity and energy-efficient antenna technology. We propose a pinching antenna-based sub-THz system that jointly enhances system energy efficiency (EE) and localization accuracy within an integrated sensing and communication design. We formulate this objective as an EE maximization problem under a localization accuracy constraint and employ alternating optimization to decouple it into two subproblems, antenna positioning and power allocation, solved via successive convex approximation. Numerical evaluations under different user and target locations show that the proposed system achieves significant gains in EE while maintaining sub-millimeter localization accuracy.

• Mata Khalili, « Application-Driven Sustainability: The Example of Cloud-Connected Robots ».

A central dimension of sustainability in 6G lies on the application side, which relies on enhanced connectivity. Applications that intelligently exploit communication capabilities can influence how networks are dimensioned and operated by activating connectivity only when they create real functional value, thereby contributing to sustainability. To address this challenge, we introduce event-triggered control and localization mechanisms in robotics applications that request control commands and positioning services from a remote server only when stability, safety, or estimation accuracy may be degraded. Between communication events, internal and computationally low-cost local dynamic models estimate the robot state while maintaining performance guarantees. Accurate positioning services, including RAN-based localization in 5G and future 6G systems when used, are therefore activated on demand rather than continuously. Using this robotics example, we show in real-world scenarios that substantial reductions in communication overhead are achievable without compromising safety or operational performance. Beyond robotics, this work advocates a shift in research toward sustainability, where network providers and application designers share responsibility and collectively contribute to reducing energy consumption and lowering CO₂ emissions.

• Cicek Cavdar, « AI-Native Green Networks: From Intelligent Base Stations to Cloud-Integrated Cell-Free Architectures ».

How can we significantly reduce the energy consumption of future mobile networks without sacrificing performance? In this talk, I present two complementary approaches toward AI-native green 6G systems. First, I focus on base stations, introducing a traffic-aware deep reinforcement learning framework that dynamically controls sleep modes to save energy, enhanced by a digital twin model that proactively estimates the risk of performance degradation and enables safe decision-making. Second, I extend the perspective to the full network, exploring how cell-free massive MIMO deployed over virtualized O-RAN architectures can achieve substantial energy savings through end-to-end orchestration of radio, fronthaul, and cloud resources. Together, these works demonstrate that intelligent, data-driven, and system-level optimization is key to building sustainable, high-performance future networks.

• Illyyne Saffar and Aurélie Boisbunon, « Energy-Aware AI: Designing Sustainable and Green AI Models for 5G and 6G Networks ».

As AI becomes increasingly embedded and present across wireless networks from 5G deployments to the emerging 6G use cases, its environmental footprint grows, making sustainability design a priority. In this presentation, we tackle the sustainable AI paradigm by outlining an overview of Green AI that covers the different steps of the AI/ML lifecycle: data collection and preprocessing, model design and training, efficient deployment and inference, and, finally, long-term adaptability.

• Hamza Abbar, « Forecast-Driven RAN Energy Control under Telemetry Lag ».

Most carrier sleep-mode studies target on/off latency at time scales of seconds to a few minutes. In live networks, the binding constraint is minute-scale telemetry lag: controllers act on aged key performance indicators (KPIs), and estimating the current state becomes a forecasting task. We model carrier shutdown as forecast-driven control under lag and develop two implementations: (i) a cell-level policy with short-horizon load predictions and threshold-based hysteresis; and (ii) a sector-level controller that predicts physical resource block (PRB) demand to keep the minimal active set. Both rely on delay-aligned features and practical safeguards (anchor carrier, minimum active capacity). Using operator-grade traces from a production multi-band network, the cell approach yields about 22--27% energy savings under a conservative calibration, while the sector policy reaches 19--22% depending on the hysteresis setting, with quality of service (QoS) preserved. These results highlight that explicitly accounting for minute-scale lag is key to achieving safe and effective radio access network (RAN) energy control.

• Marc Pierre, "Optimal control and online learning for a variable service rate discrete queue".

We consider the optimal control of the service rate of a queue with a general i.i.d. distribution for the input, subject to a convex cost (energy, spectrum). This problem is a structured Markov Decision Process (MDP) in which the state and action spaces are unbounded, and the cost is unbounded. Hence, classical techniques such as value iteration and Q-learning do not apply. We first derive strong structural results on the optimal policy so that one can restrict the search to a finite set of policies. Remarkably, those structural results are very general and hold for any input distribution (i.e., they are universal). Leveraging structure, we propose a learning algorithm that learns the optimal policy when the statistics of the input process are unknown, with small computational complexity. This same algorithm achieves sublinear regret in online learning. We finally derive an information-theoretic lower bound on the sample complexity of any learning algorithm, which almost matches that of our algorithm. This shows that our proposed learning algorithm is near-optimal.