In modern Open RAN architectures, the traditional gNB protocol stack is disaggregated into virtualized components, Centralized Unit (CU), Distributed Unit (DU), and Radio Unit (RU), which are flexibly deployed across the infrastructure to meet diverse QoS requirements. This paper proposes an energyaware model that jointly optimizes user association and virtual network function (VNF) placement to minimize total system energy consumption. By dynamically consolidating workloads and selectively activating radio and compute resources, the model reduces energy usage without compromising service constraints. We formulate the problem as an integer linear problem (ILP) to obtain optimal solutions and introduce a Graph Neural Network (GNN)-based heuristic that closely approximates optimal placements in real time. Simulation results demonstrate up to 75% energy savings at low loads and show the GNN reduces execution time by over $99 \%$ while maintaining near-optimal performance.

As energy concerns grow with the rise of energy-constrained devices, it becomes imperative to design an energy-efficient and adaptive multiple access (MA) scheme, supported with accurate energy efficiency (EE) evaluation. Non-orthogonal multiple access (NOMA) enhances EE, yet downlink NOMA faces challenges in terms of computational complexity and power demands of successive interference cancellation (SIC), problematic particularly for energy-limited devices. Existing studies overlook the additional SIC power consumption at NOMA receivers, thus overestimating EE, and giving misleading insights for real system design. Besides the need for more accurate EE evaluation, an adaptive MA approach based on this additional power consumption is required. This paper proposes a SIC-power-aware adaptive SDMA/cooperative NOMA system. An optimization problem is formulated by optimizing MA mode decision, BS beamforming, power allocation factors, and strong user relaying power, to maximize the system EE. We decouple the problem into SDMA/NOMA selection and power allocation sub-problems, solved via a modified semi-orthogonal user selection (SUS) algorithm, successive convex approximation (SCA), difference-of-convex (DC) programming, and semidefinite programming (SDP) approaches. Numerical evaluation confirms the efficiency of the proposed scheme, compared to the baseline schemes.

This paper summarizes the PhD research work conducted on resource orchestration in disaggregated Open RAN (O-RAN) architectures. In O-RAN, the gNB radio protocol stack is split into virtualized functions- Centralized Unit (CU), Distributed Unit (DU), and Radio Unit (RU)-that are dynamically deployed across edge and regional cloud infrastructures. This architectural transformation introduces new challenges in managing computing, radio, and transport resources while satisfying diverse Quality of Service (QoS) requirements. Our research is structured into three major contributions. First, the problem of CU/DU placement is addressed under fixed User Equipment (UE)-to-RU associations through an Integer Linear Programming (ILP) formulation and a Bi-directional Long Short-Term Memory (Bi-LSTM)-based Recurrent Neural Network (RNN) heuristic for scalable inference. The second contribution extends the model to jointly optimize CU/DU placement and UE-to-RU association, with a decomposition heuristic and an enhanced RNN-based solution. The proposed models significantly outperform conventional baselines in terms of user admittance and runtime. Finally, an energy-aware extension is introduced to minimize total system power consumption by consolidating workloads and managing radio and computing resources activation aiming to enhance the sustainability of$\mathbf{O}$RAN orchestration while maintaining QoS guarantees.

Open Radio Access Network (O-RAN) has recently emerged as a new trend for mobile network architecture. It is based on four founding principles: disaggregation, intelligence, virtualization, and open interfaces. In particular, RAN disaggregation involves dividing base station virtualized networking functions (VNFs) into three distinct components - the Open-Central Unit (O-CU), the Open-Distributed Unit (O-DU), and the Open-Radio Unit (O-RU) - enabling each component to be implemented independently. Such disaggregation improves system performance and allows rapid and open innovation in many components while ensuring multi-vendor operability. As the disaggregation of network architecture becomes a key enabler of O-RAN, the deployment scenarios of VNFs on O-RAN clouds become critical. In this context, we propose an optimal and dynamic placement scheme of the O-CU and O-DU functionalities on the edge or in regional O-clouds. The objective is to maximize users’ admittance ratio by considering mid-haul delay and server capacity requirements. We develop an Integer Linear Programming (ILP) model for O-CU and O-DU placement in O-RAN architecture. Additionally, we introduce a Recurrent Neural Network (RNN) heuristic model that can effectively emulate the behavior of the ILP model. The results are promising in terms of improving users’ admittance ratio by up to 10% when compared to baselines from state-of-the-art. Moreover, our proposed model minimizes the deployment costs and increases the overall throughput. Furthermore, we assess the optimal model’s performance across diverse network conditions, including variable functional split options, link capacity bottlenecks, and channel bandwidth limitations. Our analysis delves into placement decisions, evaluating admittance ratio, radio and link resource utilization, and quantifying the impact on different service types.

O-RAN architecture introduces a new level of flexibility in managing Radio Access Networks (RANs), facilitating the development of different applications. One of these applications is spectrum sharing, in which cellular traffic can share the unlicensed band with WLAN technologies, such as Wi-Fi. A key component of this application is a spectrum classification unit that identifies the communication technology used in the medium to support decision making in the RAN. This classification can be performed using Deep Neural Networks (DNNs) that receive I/Q samples and infer which communication technology is generating the traffic. Despite the high accuracy of DNNs in this task, the inference must be performed quickly to allow timely action to avoid interference. One promising approach to enhancing the performance of DNNs is to use early-exit DNNs (EE-DNNs), which are designed to reduce computations by allowing the inference process to terminate at intermediate layers when a certain confidence level is achieved. In this paper, we explore the application of EE-DNNs for spectrum classification by applying early exits to the Convolutional Neural Network (CNN) used by the ChARM (Channel-Aware Reacting Mechanism) framework. Using the ChARM dataset, we show that an EE-DNN can accelerate inference by 10% and even achieve higher accuracy than a conventional CNN by approximately 2%.

Non-orthogonal multiple access technique (NOMA) has appeared at the forefront as a viable solution capable of improving spectral and energy efficiency in fifth-generation (5G) and beyond-5G networks. This study aims at evaluating the benefits of adopting network slicing in cooperative NOMA-based systems with underlay Device-to-Device (D2D) communications. We formulate an optimization problem that maximizes the overall system's throughput while guaranteeing slices' technical requirements. We decouple the problem into two sub-problems: first, assigning the cellular users to resource blocks allocated to each NOMA group and then assigning D2D pairs to NOMA groups. A two-stage resource allocation solution by swapping-based matching theory is implemented. Numerical results show that the proposed scenario outperforms other ones in terms of the overall system's throughput and the number of admitted D2D pairs.

Open Radio Access Network (O-RAN) is a novel architecture that enables the disaggregation and the virtualization of network components. This would provide new ways to mix and match network components by “opening up” the interfaces between them. O-RAN enables driving down the costs of network deployments and allows the entry of new players into the RAN market. It enables network operators to maximize resource utilization and deliver new network edge services at a lower cost, resulting in higher profits for operators. In this context, we consider a computing resource allocation problem for maximizing the operator’s profit. Given that an operator receives subscribers’ payments and pays the infrastructure provider's costs, we model the problem using Mixed Integer Linear Programming (MILP). Then, we propose to solve the problem using Reinforcement Learning (RL). Our simulation results demonstrate the ability of the RL agent to increase the operator's profit while reducing the algorithmic complexity of the MILP solver.

Open Radio Access Network (O-RAN) has recently emerged as a new trend for mobile network architecture. It is based on four founding principles: disaggregation, intelligence, virtualization, and open interfaces. In particular, RAN disaggregation involves dividing base station virtualized networking functions (VNFs) into three distinct components -the Open-Central Unit (O-CU), the Open-Distributed Unit (O-DU), and the Open-Radio Unit (O-RU) -enabling each component to be implemented independently. Such disaggregation aims to improve system performance and allow rapid and open innovation in many components while ensuring multi-vendor operability. As the disaggregation of network architecture becomes a key enabler of O-RAN, the deployment scenarios of VNFs over ORAN clouds become critical. In this context, we propose an optimal and dynamic placement scheme of the O-CU and O-DU functionalities either on the edge or in regional O-clouds. The objective is to maximize users’ admittance ratio by considering mid-haul delay and server capacity requirements. We develop an Integer Linear Programming (ILP) model for VNF placement in O-RAN architecture. Additionally, we introduce a Recurrent Neural Network (RNN) heuristic model that can effectively replicate the behavior of the ILP model. We get promising results in terms of improving users’ admittance ratio by up to 10% when compared to baselines from state-of-the-art. Moreover, our proposed model minimizes the deployment costs and increases the overall throughput.

Cloud Radio Access Network is a promising mobile network architecture based on centralizing the baseband processing of many cellular base stations in a BBU (BaseBand Unit) pool. Such architecture has many advantages. However, computing resources are shared among the base stations connected to the BBU pool. It is challenging to schedule the processing of users’ data, especially on overloaded BBU pools, while respecting the time constraints imposed by the Hybrid Automatic Repeat Request (HARQ) mechanism. Given that the processing time of users’ data and the computing requirement depends on the radio parameters such as the Modulation and Coding Scheme (MCS), we propose to enable the coordination between radio and computing resources schedulers; such coordination makes the selection of MCS dependent on the availability of radio and computing resources and on the ability to process data while respecting the HARQ-deadline. In this context, we propose and evaluate three Integer Linear Programming (ILP)-based schemes and three low-complexity heuristics, demonstrating their ability to reduce the wasted transmission power. Moreover, we evaluate the performance of the coordination under a multi-services scenario consisting of two services having heterogeneous requirements, enhanced Mobile Broadband (eMBB) and Ultra-Reliable Low-Latency Communication (URLLC).

Cloud-RAN is a key 5G-enabler; it consists in centralizing the baseband processing of base stations by executing the baseband functions in a centralized, virtualized, and shared entity known as the Base Band Unit (BBU)-Pool. Cloud-RAN paves the way for joint management of the resources of multiple base stations. This paper aims to analyze the potential reduction in power consumption brought by the joint allocation of the radio and computing resources. We formulate a Mixed Integer Linear Programming (MILP) problem, considering the objective of power consumption minimization. For comparison, we consider the objective of throughput maximization. When the goal is power minimization, the joint allocation can minimize the total power consumption by up to 21.2%, with respect to the case where radio and computing resources in the BBU pool are allocated sequentially.

Cloud Radio Access Network (Cloud-RAN) is a novel architecture that aims at centralizing the baseband processing of base stations. This architecture opens paths for joint, flexible, and optimal management of radio and computing resources. To increase the benefit from this architecture, efficient resource management algorithms need to be devised. In this paper, we consider a coordinated allocation of radio and computing resources to mobile users. Optimal resource allocation that respects the Hybrid-Automatic-Repeat-Request deadline may require formulating high-complexity and resource-heavy algorithms. We consider two Integer Linear Programming problems (ILP) that implement a coordinated allocation of radio and computing resources with the objectives of maximizing throughput and maximizing users' satisfaction, respectively. Since solving these highly-complex problems requires a high execution time, we investigate low-complexity alternatives based on machine learning models; more precisely on Recurrent Neural Networks (RNN). These RNN models aim to depict the performance of the ILP problems with a much lower execution time. Our simulation results demonstrate the great ability of RNN models to perform very closely to the ILP problems while being able to reduce the execution time by up to 99.65%.

Cloud RAN (C-RAN) is a very promising architecture for future mobile network deployment, where the cloud-centric approach is useful in improving total processing load. In this context, radio and baseband network functions processing pose interesting problems that we try to expose and solve in this paper. A novel architecture for C-RAN and a first modeling of the system are proposed. Furthermore, we study the impact of many radio parameters on the processing time. Moreover, a mathematical and a deep learning model are proposed and evaluated for processing time prediction. Results show the feasibility of the proposed approaches.

In networking and computing, resource allocation is typically addressed using classical resource allocation protocols as the proportional rule, the max-min fair allocation, or solutions inspired by cooperative game theory. In this paper, we argue that, under awareness about the available resource and other users demands, a cooperative setting has to be considered in order to revisit and adapt the concept of fairness. Such a complete information sharing setting is expected to happen in 5G environments, where resource sharing among tenants (slices) need to be made acceptable by users and applications, which therefore need to be better informed about the system status via ad-hoc (northbound) interfaces than in legacy environments. We identify in the individual satisfaction rates the key aspect of the challenge of defining a new notion of fairness in systems with complete information sharing, consequently, a more appropriate resource allocation algorithm. We generalize the concept of user satisfaction considering the set of admissible solutions for bankruptcy games and we adapt to it the fairness indices. Accordingly, we propose a new allocation rule we call mood value: for each user, it equalizes our novel game-theoretic definition of user satisfaction with respect to a distribution of the resource. We test the mood value and a new fairness index through extensive simulations about the cellular frequency scheduling use-case, showing how they better support the fairness analysis. We complete the paper with further analysis on the behavior of the mood value in the presence of multiple competing providers and with cheating users.

In this paper, we leverage two large-scale real-world datasets to provide the first results on the limits of predictability of cellular data traffic demands generated by individual users over time and space. Using information theory tools, we measure the maximum predictability that any algorithm has potential to achieve. We first focus on the predictability of mobile traffic consumption patterns in isolation. Our results show that it is theoretically possible to anticipate the individual demand with a typical accuracy of 85% and reveal that this percentage is consistent across all user types. Then, we analyze the joint predictability of the traffic demands and mobility patterns. We find that the two dimensions are correlated, which improves the predictability upper bound to 90% on average.

Transmit power allocation techniques are very important to manage interference in small-cell networks. While available power allocation algorithms in the literature rely on a predefined routing protocol, we propose in this paper a power-efficient two-step algorithm that allows power allocation and routing to be performed jointly in femto-relay networks. First, we propose an interference-based partitioning method to cluster the femto-relays, then we adopt an iterative and distributed algorithm, inspired from game theory, for efficient transmit power allocation. We show that the corresponding power allocation game possesses a pure Nash equilibrium which is reached by the proposed algorithm within a number of iterations per femto-relay which can be as small as 1. Moreover, we show that our approach grants significant improvements in terms of power consumption, and permits the total consumed power to be divided by about 6 and 3 when respectively compared to the direct transmission and shortest path techniques.

A present issue in the evolution of mobile cellular networks is determining whether, how and where to deploy adaptive content and cloud distribution solutions at the base station and backhauling network level. Intuitively, an adaptive placement of content and computing resources in the most crowded regions can grant important traffic offloading, improve network efficiency and user quality of experience. In this paper we document the content consumption in the Orange cellular network for the Paris metropolitan area, from spatial and application-level extensive analysis of real data from a few million users, reporting the experimental distributions. In this scope, we propose a hotspot cell estimator computed over user's mobility metrics and based on linear regression. Evaluating our estimator on real data, it appears as an excellent hotspot detection solution of cellular and backhauling network management. We show that its error strictly decreases with the cell load, and it is negligible for reasonable hotspot cell load upper thresholds. We also show that our hotspot estimator is quite scalable against mobility data volume and against time variations.

Nowadays, the huge worldwide mobile-phone penetration is increasingly turning the mobile network into a gigantic ubiquitous sensing platform, enabling large-scale analysis and applications. In recent years, mobile data-based research reaches important conclusions about various aspects of human mobility patterns and trajectories. But how accurately do these conclusions reflect the reality? In order to evaluate the difference between the reality and the approximation methods, we study in this paper the error between real human trajectory and the one obtained through mobile phone data using different interpolation methods (linear, cubic, nearest and spline interpolations) while taking into account some mobility parameters. From extensive evaluations based on real cellular network activity data of the Boston metropolitan area, we show that the linear interpolation offers the best estimation for sedentary people and the cubic one for commuters. Moreover, the nearest interpolation appears as the best one for “ordinary people” doing regular stops and standard displacements. Another important experimental finding described in this paper is that trajectory estimation methods show different error regimes whether used within or outside the “territory” of the user defined by the radius of gyration.

Wireless mesh networks (WMNs) are emerging as a key solution to provide broadband and mobile wireless connectivity in a flexible and cost-effective way. In suburban areas, a common deployment model relies on orthogonal frequency division multiple access (OFDMA) communications between mesh routers (MRs), with one MR installed at each user premises. In this paper, we investigate a possible user cooperation path to implement strategic resource allocation in OFDMA WMNs, under the assumption that users want to control their interconnections. In this case, a novel strategic situation appears: How much an MR can demand, how much it can obtain, and how this shall depend on the interference with its neighbors. Strategic interference management and resource allocation mechanisms are needed to avoid performance degradation during congestion cases between MRs. In this paper, we model the problem as a bankruptcy game taking into account the interference between MRs. We identify possible solutions from cooperative game theory, namely the Shapley value and the nucleolus, and show through extensive simulations of realistic scenarios that they outperform two state-of-the-art OFDMA allocation schemes, namely, centralized-dynamic frequency planning, and frequency-ALOHA. In particular, the nucleolus solution offers best performance overall in terms of throughput and fairness, at a lower time complexity.

Femtocells have recently appeared as a viable solution to enable broadband connectivity in mobile cellular networks. Instead of redimensioning macrocells at the base station level, the modular installation of short-range access points can grant multiple benefits, provided that interference is efficiently managed. In the case where femtocells use different frequency bands than macrocells (i.e., split-spectrum approach), interference between femtocells is the major issue. In particular, congestion cases in which femtocell demands exceed the available bandwidth pose an important challenge. If, as expected, the femtocell service is going to be separately billed by legacy wire-line Internet Service Providers, strategic interference management and resource allocation mechanisms are needed to avoid performance degradation during congestion cases. In this paper, we model the resource allocation in cooperative femtocell networks as a bankruptcy game. We identify possible solutions from cooperative game theory, namely the Shapley value and the Nucleolus, and show through extensive simulations of realistic scenarios that they outperform two state-of-the-art schemes, namely Centralized-Dynamic Frequency Planning, C-DFP, and Frequency-ALOHA, F-ALOHA. In particular, the Nucleolus solution offers best performance overall in terms of throughput and fairness, at a lower time complexity.

Wireless Mesh Networks (WMNs) are emerging as a key solution to provide broadband and mobile wireless connectivity in a flexible and cost effective way. In suburban areas, a common deployment model relies on OFDMA communications between mesh routers (MRs), with one MR installed at each user premises. In this paper, we investigate a possible user cooperation path to implement strategic resource allocation in OFDMA WMNs, under the assumption that users want to control their interconnection. In this case, a novel strategic situation appears: how much a MR can demand, how much it can obtain and how this shall depend on the interference with its neighbors. Strategic interference management and resource allocation mechanisms are needed to avoid performance degradation during congestion cases between MRs. In this paper, we model the problem as a bankruptcy game taking into account the interference between MRs. We identify possible solutions from cooperative game theory, namely the Shapley value and the Nucleolus, and show that they outperform two state-of-the-art schemes, namely Centralized-Dynamic Frequency Planning, C-DFP, and Frequency-ALOHA, F-ALOHA. In particular, the Nucleolus solution offers best performance overall in terms of throughput and fairness.