The RAN is responsible for approximately 70% of the total power consumption of 5G systems. The disaggregation and virtualization of the 5G RAN specified in the O-RAN architecture introduce an unprecedented flexibility for dynamic resource management, enabling operations that enhance the energy-efficiency strategies of the overall network. In this paper, we investigate how traffic-aware scaling of the CU-UP component can be used to reduce the operational power and energy consumption of a fully disaggregated 5G O-RAN network. We design two AI-driven xApps running on the near-RT RIC that implement proactive CU-UP scaling based on machine-learning traffic forecasting. The first xApp implements a vertical scaling strategy by dynamically adjusting the CPU resources allocated to a CU-UP instance to minimize the 5G O-RAN power consumption, while the second performs latency-aware horizontal scaling by instantiating and placing multiple CU-UP instances across the infrastructure to jointly optimize energy consumption and midhaul latency. The proposed approaches are evaluated on a 5G O-RAN testbed using traffic traces from a real-world dataset. The dataset follows a real UE behavior rather than artificially prolonged low-load conditions that would act in favour of our contributions. Experimental results shed light on the applicability of scaling for the CU-UP component. Results demonstrate that vertical scaling maximizes energy efficiency, achieving up to 23% energy savings during underutilized regimes, 11% during sub-peak periods, and 8% over the full execution compared to a baseline static provisioning. In contrast, horizontal scaling delivers a complementary trade-off, achieving up to 16% savings during underutilized regimes, 7% during non-peak periods, and 4.7% for the whole execution, while significantly reducing end-to-end latency by optimally placing CU-UP instances. The results highlight that ML-driven CU-UP scaling is a practical and effective mechanism to reduce the power consumption, improve the energy efficiency, and the performance of the O-RAN 5G networks.

Energy consumption in 5 G radio access networks (RAN) remains a major operational cost and environmental concern, as base stations account for the majority of power usage. Traditional power-saving mechanisms rely on static thresholds or heuristics that cannot adapt to rapidly changing traffic and channel conditions. In this paper, we present a fully datadriven reinforcement learning framework for downlink transmitamplitude control that learns to minimize radiated power while satisfying per-user quality-of-service (QoS) constraints. Training occurs in a dataset-driven “digital twin” built on an open-source 5G stack and realistic RF channel emulation with real traffic traces. We then transfer the learned policy to live hardware via an O-RAN compliant controller and demonstrate on-air RF power reduction of 75% on average, while maintaining QoS metrics comparable to conservative threshold-based approaches that achieve only 59%. To our knowledge, this is the first end-toend demonstration of an RL-based power-control agent bridging simulation and live 5 G deployment, highlighting the promise of AI-driven RAN energy optimization.

The increasing energy footprint of mobile networks has made power consumption a critical operational metric for the B5G & 6G networks. Focusing on current O-RAN deployments, energy monitoring relies on external hardware sensors that suffer from measurement drifting and calibration errors. This paper investigates the feasibility of predicting the power consumption of O-RAN components using only 3GPP-compliant network Key Performance Indicators (KPIs). We present an AI-driven framework that performs power prediction, anomaly detection, and automatic calibration of hardware measurements. The proposed approach utilizes KPI-driven Machine Learning models, trained on data collected from a real 5G O-RAN deployment. Hybrid XGBoost and CatBoost models are used to estimate the power consumption of the O-DU and O-RU nodes. Residual analysis between predicted and measured power enables the detection of sensor anomalies and measurement drift, followed by automatic calibration. Experimental evaluation of our proposed framework on a real-world testbed demonstrates accurate power prediction with a MAPE of 1.02% for the RU and 1.22% for the DU, and a 100% detection rate for manually injected anomalies.

Digital research infrastructures underpin data-intensive science, reproducible experimentation, and collaborative innovation. In Africa, their development remains hindered by fragmentation, limited computing and connectivity resources, shortages of technical expertise, and sustainability constraints. This paper presents the vision, methodology, and deployment roadmap of DIGITAfrica, a Horizon Europe initiative preparing a pan-African Research Infrastructure in digital sciences. Building on a multi-country demand analysis and lessons from the ESFRI research infrastructures SLICES and SoBigData, we introduce a blueprint-based methodology for specifying interoperable, incrementally deployable, and sustainable digital infrastructures. We describe two representative blueprints—edge-interoperable AI/ML and heterogeneous networking—and the shared service architecture that operationalise them. Finally, we outline the stakeholder engagement process intended to support adoption, capacity building, and long-term governance. Overall, DIGITAfrica proposes a federated framework to strengthen digital science capabilities in Africa through shared infrastructure, reusable experimental environments, and coordinated community development.

5G and beyond networks aim to deliver higher speeds, lower latency, and denser device connectivity. Software-Defined Networking (SDN) and Virtual Network Functions (VNFs) are key enablers of these goals, offering flexible, software-driven solutions that decouple the networking functionalities from vendor-closed traditional hardware. In this context, many implementations of the User Plane Function (UPF), such as SPGW-U and VPP, rely on software for packet forwarding and traffic management. Additionally, the integration of AI and machine learning (AI/ML) offers predictive capabilities, allowing future networks to optimize resource allocation, anticipate traffic patterns, and enhance overall network performance dynamically. This work develops a proactive switching mechanism designed to dynamically switch between two different software implementations of the UPF, namely the SPGW-U and VPP-based implementation, based on throughput performance and energy consumption. SPGW-U offers limited throughput performance but consumes reasonable energy, whereas VPP delivers significantly higher throughput performance at the cost of much greater energy consumption. The proposed mechanism evaluates these trade-offs and employs an integrated LSTM model to proactively predict throughput values. Based on these predictions, the switching mechanism determines when to switch between the two implementations, optimizing network performance and energy efficiency dynamically.

This paper introduces the first-ever open-source Multicast and Broadcast Services (MBS) implementation for 5G and beyond cellular networks in OpenAirInterface5G. It outlines the challenges and essential modifications required to incorporate MBS into the 5G architecture while ensuring compliance with the 3GPP specification. The study investigates the performance of the three Hybrid Automatic Repeat Request (HARQ) feedback options for MBS, including the No-ACK, ACK-NACK, and NACK-Only HARQ-ACK options. It evaluates the user plane performance of these HARQ feedback options under realistic network conditions, providing insights into their impact on reliability, throughput, spectral efficiency, and resource utilization for MBS in 5G and beyond networks. Specifically, it examines their effectiveness in scenarios where the User Equipment (UE) devices experience uniform or diverse network conditions, analyzing their overhead and the impact of the Modulation and Coding Scheme (MCS) on their overall performance. Our experiments highlight the substantial performance benefits of MBS over Unicast transmissions and the trade-off between reliability and performance introduced by the HARQ-ACK options. This work advances the practical implementation of MBS in 5G and provides valuable guidelines on optimal deployment strategies and the HARQ feedback option selection for 5G and beyond cellular networks.

SLICES-RI is being developed as a scientific instrument, following established methodologies and best practices from the scientific community. This approach is increasingly essential to address the accelerating pace of data-driven research, which continuously generates a deluge of publications. The core mission of a scientific instrument is to provide a standardized and trusted reference, enabling direct performance comparisons of algorithms and ensuring reproducibility — thereby simplifying and strengthening the peer review process. To meet these goals, SLICES-RI adopts an intent-based design and generates tailored Blueprints for specific scientific questions. While it does not aim to be exhaustive, this approach significantly reduces the complexity involved in designing and executing experiments. In this presentation, we share the current deployment status of SLICES-RI as it entered its pre-operational phase. We illustrate its capabilities through the Post-5G and Federated Learning Blueprints. Additionally, we highlight MRS/DMI, a key feature of SLICES-RI, which offers an advanced data management framework aligned with the FAIR principles and fully integrated into the experimental workflow.

Hierarchical federated learning (HFL) has attracted academic attention to improve the efficiency of federated learning (FL) in real-world applications, however, little research has been done to explore the structural advantages of HFL against Byzantine attacks and to investigate how to make HFL immune to top-level server Single Point of Failure (SPOF). To explore this field and improve the robustness of HFL, we propose a novel generalized paradigm ABD-HFL for asynchronous Byzantine-resistant decentralized hierarchical federated learning, a multi-tier structure without a central server for FL tasks with a large number of devices. Based on the layered structure, an innovative universal Byzantine resistance mechanism is designed in ABD-HFL, which enables it to apply a combination of multiple Byzantine robust techniques, making ABD-HFL more powerful than any single application of such techniques. Besides, ABD-HFL is a fully decentralized HFL, there is no central server, but rather multiple nodes at the top level agree on the global model where malicious model updates are excluded. A new concept of pipeline learning workflow is also introduced to study communication efficiency in ABD-HFL, which is based on asynchronous communication between various levels to train and propagate the global model. Our numerical evaluation validates the advantage of ABD-HFL in terms of robustness and communication efficiency.

The energy efficiency of 5G Radio Access Networks (RANs) has become a critical area of research due to the growing energy demands of the 5G disaggregated architectures and the increasing environmental concerns surrounding mobile networks. In this paper, we address the fundamental challenge of optimizing energy consumption in Open RAN (O-RAN) 5G networks by dynamically scaling Central Unit (CU) components based on traffic demands. In particular, by leveraging a real-world testbed environment, we empirically analyze the relationship between data volume, architectural configurations, and energy usage. Our study identifies key tuning parameters for energy management and optimization, showcasing the impact of dynamic resource allocation on energy consumption. Experimental results demonstrate that our implementation of a dynamic CU allocation policy can achieve energy savings of up to 60% compared to static configurations, without compromising Quality of Service (QoS).

The 5th generation and beyond of cellular networks (5G & B5G) adopted a cloud-based architecture, representing a significant evolution compared to monolithic approaches of previous generations. This transformation has even been devised for the Radio Access Network (RAN), with the Cloud-RAN (C-RAN), and its virtualized version the Virtual RAN (V-RAN) as crucial and key components. RAN cloudification has added up to the flexibility, reduced deployment and operational costs for the network operators. The high-speed demands, vast volumes of data, and the low latency response requirements of future applications (e.g. V2X, IoT, URLLC, and others) create significant stress on some V-RAN components. In the V-RAN architecture, the CU-UP is highlighted as the datapath bottleneck of the V-RAN under high traffic loads. This paper aims to enhance the reliability of 5G V-RAN by addressing the bottleneck issues of CU-UP by developing a controller that performs reactive horizontal scaling of the CU-UP component. The scaling mechanism is driven by an algorithm that estimates CPU usage based on a Linear Regression model, based on the rate of incoming packets, with the goal to increase the overall 5G network performance delivered to the end users. Using real-world traffic patterns, we tested the proposed scaling mechanism, conducting the experiments in a non-simulated real-world testbeds exposing our system to dynamic conditions. Our findings highlight the importance of adapting such a scaling mechanism to ensure the robust reliability of the 5G V-RAN.

5G networks represent a significant advancement, supporting services such as Enhanced Mobile Broadband (eMBB), ultra-reliable low-latency communications (URLLC), and massive machine-type communications (mMTC). To meet these demands, different Radio Access Technologies (RATs), such as 4G LTE, 5G NR, and non-3GPP access technologies such as WiFi, are brought together under a unified multi-RAT architecture. While combining 3GPP and non-3GPP technologies strengthens the network, existing core-level solutions like Non-3GPP Interworking Function (N3IWF) or Access Traffic Steering, Switching, and Splitting (ATSSS) do not offer coordination or shared resource management at the RAN level between 5G NR and WiFi. Core-level traffic handling mechanisms such as ATSSS operate at the flow level, but lack direct visibility into RAN conditions and cannot respond dynamically to variations in link quality and interface load. This paper proposes a unified RAN-level integration of WiFi as a Distributed Unit (DU) within the beyond-5G O-RAN architecture, enabling packet-level traffic steering. An xApp was developed, utilizing the Near-RT RIC to make traffic steering decisions based on LSTM-based UE demand forecasting. The xApp applies a non-prioritizing policy that offloads or splits traffic to the WiFi DU when the 5G capacity is exceeded. The system was implemented on a real testbed, demonstrating that the proposed O-RAN multi-RAT design effectively combines 5G and non-3GPP RATs to meet traffic demand requirements.

Experimental research in Post-5G involves complex interactions between software, hardware, and protocols. Therefore, it is crucial to develop solutions that allow researchers to conduct their experiments in a reproducible manner. To support this need, the EU SLICES Research Infrastructure (RI) provides a scientific instrument that encompasses all the needs for Post-5G experimental research. The facility is currently being built to enable experimentation with state-of-the-art resources in various fields. SLICES-RI is intent-driven and facilitates the entire lifecycle of thought experiments. This is achieved by enabling reproducible deployment of experiments over the infrastructure using blueprints and by systematically collecting and archiving all outputs through a clear and structured methodology for experimentation. For this demonstration, we focus on the Post-5G part of the facility and will showcase how the entire lifecycle of such an experiment is orchestrated using the tools and functionalities developed. We will showcase blueprints for deploying a cloud-native 5G core and a split 7.2 radio network using open-source software in a fully reproducible manner, with the results being automatically archived and published using the SLICES metadata model. The reproducibility, deployment options, experimenter control capabilities, and access to the collected results will be highlighted.

Fifth generation (5G) and beyond cellular networks bring significant advantages over their predecessors in the RAN part of the network, as they are based on a disaggregated, cloudbased architecture, marking a major shift from the monolithic designs of previous generations. This disaggregated approach has led to the development of Virtual RAN (V-RAN) and Open RAN (O-RAN) architectures, which provide greater flexibility in deployment while optimizing costs and resource utilization. Simultaneously, the increasing demands for high-speed, lowlatency connectivity to support diverse 5G applications such as IoT, V2X, and URLLC highlight the critical need for reliable and scalable network solutions. In this paper, we address the scalability challenges of the CU-UP component, identified as a bottleneck in the 5G disaggregated RAN, particularly under high traffic loads. We propose a proactive horizontal scaling mechanism for the CU-UP, leveraging machine learning to forecast traffic demands. This involves using an xApp on the Near-RT RIC that integrates a pre-trained Long Short-Term Memory (LSTM) model using a real-world dataset which predicts traffic demands, enabling proactive scaling of CUUP instances ahead of peak periods to maintain Quality of Service (QoS). The proposed mechanism was implemented and evaluated using real-world testbeds under dynamic conditions, demonstrating its effectiveness in practical environments. Our findings emphasize the importance of integrating ML techniques to forecast network loads accurately and adapting such scaling mechanisms in 5G systems to reduce packet loss and unnecessary scaling overhead while ensuring smoother operation during highdemand periods.

This paper presents a novel blockchain-based decentralized identity (DI) system, tailored for enhanced digital identity management in Internet of Things (IoT) and device-todevice (D2D) networks. The proposed system features a hierarchical structure that effectively merges a distributed ledger with a mobile D2D network, ensuring robust security while streamlining communication. Central to this design are the gateway nodes, which serve as intermediaries, facilitating decentralized identifier (DID) registration and device authentication through smart contracts and distributed storage systems. A thorough security analysis underscores the system's resilience to common cyber threats and adherence to critical principles like finality and liveness.

Replication is crucial for maintaining the credibility and integrity of scientific research and is one of Europe's key enablers for Open Science. Several challenges must be addressed to facilitate replication, including applying robust methodologies, holistic data sharing, and detailed data lineage and provenance to allow researchers to leverage insights and findings from prior investigations and introduce novel perspectives or solutions in their field. Research Infrastructures are an important catalyst towards addressing the replication “crisis” by enforcing data sharing by design/default, with appropriate protocols and procedures that provide visibility and transparency to the whole data journey, and compliance with national and international regulations. SLICES Research Infrastructure will construct one of Europe's most advanced scientific platforms in the field of digital sciences, promoting scientific research replication through sophisticated policies and services. SLICES-RI transcends traditional data sharing of common digital objects, such as datasets, services and tools, by introducing replication of complex digital objects, such as experimental workflows, which orchestrate advanced tools and services to perform sophisticated experiments in smart networks and systems. This paper presents the preliminary design of the SLICES-RI replication framework, providing insight into its internal structures and mechanisms, and demonstrating how experimental workflows can be replicated. The proposed SLICES Trustworthy Experimental Replication Framework (STEF) provides visibility into the experiment workflow and the underlying data journey, and demonstrates the potential integration of sophisticated indicators, such as explainability, interpretability, and safety, contributing to trustworthy experimentation.

Traditional machine learning systems were designed in a centralized manner. In such designs, the central entity maintains both the machine learning model and the data used to adjust the model's parameters. As data centralization yields privacy issues, Federated Learning was introduced to reduce data sharing and have a central server coordinate the learning of multiple devices. While Federated Learning is more decentralized, it still relies on a central entity that may fail or be subject to attacks, provoking the failure of the whole system. Then, Decentralized Federated Learning removes the need for a central server entirely, letting participating processes handle the coordination of the model construction. This distributed control urges studying the possibility of malicious attacks by the participants themselves. While poisoning attacks on Federated Learning have been extensively studied, their effects in Decentralized Federated Learning did not get the same level of attention. Our work is the first to propose a methodology to assess poisoning attacks in Decentralized Federated Learning in both churn free and churn prone scenarios. Furthermore, in order to evaluate our methodology on a case study representative for gossip learning we extended the gossipy simulator with an attack injector module.

We propose Challenger a peer-to-peer blockchain-based middleware architecture for narrative games, and discuss its resilience to cheating attacks. Our architecture orchestrates nine services in a fully decentralized manner where nodes are not aware of the entire composition of the system nor its size. All these components are orchestrated together to obtain (strong) resilience to cheaters. The main contribution of the paper is to provide, for the first time, an architecture for narrative games agnostic of a particular blockchain that brings together several distinct research areas, namely distributed ledgers, peer-to-peer networks, multiplayer-online games and resilience to attacks.

5G has evolved into a cloud-native disaggregated infrastructure, enabling the concept of modularization in its design and supporting a service-based architecture. As such the concept is not new, except that it has not been applied in the telecommunication world, which has resulted in vendor lockin, limited innovation, and high costs. With modular design at all levels, 5G allowed the telco world to meet the IT world and outstanding innovations followed with the so-called post-5G propositions. The EU SLICES Research Infrastructure (RI) is developing an open, reproducible, distributed post-5G architecture built on top of blueprints aimed to be replicated by researchers, companies, and operators and to evolve in a collaborative manner. In this demo, we will explore the blueprints that allow building a cloud-native 5G core and a split 7.2 radio network, based on open-source software while being fully reproducible. The objective of the demo is not only to show a deployment but to motivate the research community to participate in the collaborative SLICES-RI project and to adopt a reproducible methodology supporting the full research life cycle.

The countless needs of the modern world for connectivity have led to the rapid evolution of wireless networks and the creation of the latest and most current generations of 5G cellular networks and WiFi wireless networks that promise unprecedented levels of data speeds, low latency, and seamless connectivity. The integration of WiFi in the architecture of the 5th-generation cellular networks brings about an increase in the coverage, the exploitation of unlicensed spectrums, and the load-balancing of the network. Still, it also paves the way for creating a more robust and versatile next-generation wireless ecosystem. This paper presents and implements a 5G hybrid Distributed Unit (DU) architecture, integrating a WiFi DU into real-world 5G networks using the OpenAirInterface (OAI) RAN and 5GCN software. This work also tries to align the WiFi technology with the dynamic and flexible capabilities of 5G such as slicing. The WiFi functionality is extended, in order to support a User-Based Slicing scheme using Software-Defined Networking (SDN) combined with proper exploitation of the Type of Service (ToS) field of IP packets to prioritize traffic and manage network resources.