Thèse Vers des Réseaux d'Accès Radio Ran Durables de Nouvelle Génération Optimisation Énergétique Tenant Compte des Contrôles dans les Systèmes Open-Ran H/F - Doctorat.Gouv.Fr

Établissement : Université Paris-Saclay GS Informatique et sciences du numérique École doctorale : Sciences et Technologies de l'Information et de la Communication Laboratoire de recherche : Laboratoire des Signaux et Systèmes Direction de la thèse : Véronique VEQUE ORCID 0000000217526020 Début de la thèse : 2026-10-01 Date limite de candidature : 2026-05-10T23:59:59 Dans les réseaux radio d'accès de nouvelle génération, l'optimisation énergétique intégrant le plan de contrôle nécessite de modéliser à la fois la consommation du réseau et celle du plan de contrôle, ce qui complexifie les problèmes d'optimisation sous contraintes de qualité de service (QoS). Les stratégies de contrôle présentent des compromis entre efficacité et complexité de calcul qu'il est essentiel de caractériser. La nature dynamique et distribuée des systèmes Open RAN impose également des solutions adaptatives et scalables.

Cette thèse vise à développer un cadre unifié intégrant la consommation énergétique de l'infrastructure et du plan de contrôle afin de minimiser l'énergie totale du système. Elle inclut la conception de stratégies d'orchestration écoénergétiques, la modélisation conjointe réseau-contrôle de l'énergie, ainsi que l'étude de méthodes d'optimisation, en ligne ou avec apprentissage, avec un accent sur le passage à l'échelle et une évaluation réaliste Early Radio Access Network (RAN) architectures [Pana2022] followed a fully distributed model (D-RAN), where Remote Radio Heads (RRHs) and Baseband Units (BBUs) were co-located at each base station. While simple, this design led to inefficient resource utilization and energy consumption, as BBUs remain powered even under low traffic conditions, resulting in unnecessary energy expenditure alongside high capital and operational costs.

The introduction of Centralized RAN (C-RAN) improved both resource and energy efficiency by pooling BBUs in centralized cloud data centers. This enables dynamic resource allocation and load-aware processing, allowing underutilized resources to be consolidated and idle components to be powered down. Furthermore, techniques such as cell switch-off and sleep modes during low-demand periods significantly reduce overall network energy consumption. However, these gains are partially offset by the energy cost of fronthaul transport and the stringent latency and bandwidth requirements imposed by centralization.

The evolution toward Open RAN further enhances energy-saving opportunities through virtualization, disaggregation, and intelligent control. By decoupling hardware and software and enabling deployment over cloud infrastructures, Open RAN supports elastic scaling of Virtual Network Functions (VNFs), aligning energy usage with traffic demand. The integration of AI/ML-driven RAN Intelligent Controllers (RICs) enables advanced energy optimization strategies, such as traffic-aware function placement, adaptive resource provisioning, and predictive energy management. Additionally, disaggregation radio units (RUs), distributed units (DUs), and centralized units (CUs) allow for fine-grained control of energy consumption across network segments, although it introduces new trade-offs between performance, latency, and energy efficiency, particularly in the placement of VNFs and the operation of midhaul links.
Despite these advancements, achieving optimal energy efficiency in Open RAN remains challenging due to the increased system complexity, distributed cloud infrastructure, and the need for real-time control. Future developments, including real-time RIC components and data-driven optimization frameworks, are expected to play a key role in enabling holistic, end-to-end energy-efficient RAN operation in next-generation wireless networks. Introducing control-aware energy optimization raises several challenges. First, it requires modeling both RAN energy consumption and the energy cost of control mechanisms, including computation and signaling overhead. Second, the optimization problem becomes more complex, as it must minimize total system energy rather than network energy alone, while satisfying Quality of Service (QoS) constraints.

Moreover, different control strategies exhibit different tradeoffs between efficiency and complexity. Lightweight methods are less costly but less adaptive, whereas advanced approaches such as MPC or reinforcement learning provide better performance at higher computational cost [Chen2023]. Characterizing these tradeoffs is essential. Finally, the dynamic nature of wireless networks and the distributed structure of Open RAN systems require scalable and adaptive control mechanisms [Hojeij2025].
The objective of this PhD project is to develop a unified framework for the control and optimization of next-generation RANs, explicitly incorporating both infrastructure and control-plane energy consumption. The goal is to devise energy-efficient orchestration strategies that minimize the total system energy footprint while ensuring stringent QoS constraints.
This involves developing joint energy models, formulating new optimization problems, and investigating different solution approaches, including optimization-based, online, and learning-based methods, with a focus on scalability and realistic evaluation. The work will combine modeling, optimization, and algorithm design. First, existing RAN energy models will be extended to include control-plane components such as computation and signaling costs. Based on this model, optimization problems minimizing total energy will be formulated.
Different solution approaches will be explored, including exact optimization for benchmarking, online control for handling stochastic traffic, and learning-based methods for large-scale scenarios. The proposed methods will be evaluated through simulation under realistic traffic conditions, comparing their performance in terms of energy efficiency, QoS, and computational complexity.

Le candidat, titulaire d'un diplôme de master ou d'ingénieur (en télécommunications ou en informatique, par exemple), doit posséder de solides compétences en architectures réseau, en simulation/émulation, en techniques d'intelligence artificielle, en évaluation des performances et en programmation C/C++/Python. Il doit également avoir une solide formation en mathématiques appliquées, en optimisation ou en théorie du contrôle.