Wednesday, 3 June 2026, 14:30-15:00, room Sala de Exposiciones 1
Session Chair: tbd
P-1.1 Research on Network Traffic Prediction and Network Slicing Resource Allocation Optimization Based on Graph Neural Networks
Jen-Ai Yeh and Shan-Hsiang Shen (National Taiwan University of Science and Technology, Taiwan)
We adopt a Spatial-Temporal Graph Convolutional Network (ST-GCN) to forecast future traffic between origin-destination (OD) pairs in the Abilene backbone network. Based on the predicted load, the system dynamically adjusts virtual link capacities using a window based slicing policy.
P-1.2 Poster: Testing Network Slicing Isolation with OpenAirInterface
Paulo Marques (Allbesmart Lda, Portugal); Tiago Alves (Allbesmart, Lda, Portugal); Luis Pereira (Allbesmart Lda, Portugal); Roberto Magueta (Allbesmart, Portugal); Pedro Valente (Universidade de Aveiro, Portugal & Instituto de Telecomunicações, Portugal); Pedro Rito (University of Aveiro, Portugal & Instituto de Telecomunicações, Portugal); Duarte Raposo (Instituto de Telecomunicações, Portugal); Susana Sargento (Universidade de Aveiro, Portugal); Carlos Marques and Miguel Mesquita (Altice Labs, Portugal); Filipe Cabral Pinto (Alticelabs, Portugal)
This paper presents a lab demonstration of Beyond 5G network slicing using the open-source OpenAirInterface (OAI) stack and the OAIBOX platform, as part of the Aveiro Tech City Living Lab (ATCLL) in the scope of the 6G-PATH European project. A network slicing xApp is implemented and evaluated using commercial user equipment. Additionally, a monitoring xApp was integrated with the ATCLL data platform collecting real-time telemetry through FlexRIC. The collected data is available in real-time or historically in the 6G-PATH portal, providing it to all partners. Experimental results demonstrate effective slice isolation, with the URLLC slice (for autonomous vehicle applications) maintaining ultra-low latency even under eMBB congestion, highlighting the potential of network slicing for strict differentiated services requirements in 5G Advanced networks.
P-1.3 A Framework for Handover Management in 6G Integrated TN-NTN
Shama Noreen (RPTU, Germany); Hans D. Schotten (University of Kaiserslautern, Germany)
This work presents a unified, AI-driven handover framework for 6G integrated terrestrial and non-terrestrial networks (TN-NTN). Specifically, it presents an AI-driven mea- surement and mobility framework that dynamically tailors mea- surement activation and parameter configuration to the instan- taneous mobility scenario, UE activity, and network conditions. By reducing unnecessary signaling and UE power consumption while maintaining handover reliability, the framework supports more efficient and adaptive mobility control. More broadly, it contributes toward scalable, intelligent, and resilient mobility management for future 6G systems
P-1.4 5G Challenges in Industry and the Pathway to 6G
Parva Yazdani (Ifak, Germany); Cole Niemoeller Saunders (Institute for Automation and Communication e.V., Germany); Philipp Schulz (Technische Universität Dresden, Germany); Sarah Willmann (Institut Fur Automation Und Kommunikation, Germany); Lisa Underberg (Ifak, Germany)
This work presents a systematic investigation of the challenges encountered and lessons learned during the industrial deployment of fifth-generation cellular networks. It also sets out a set of actionable design principles for transitioning to sixth-generation networks. The study is accompanied by a structured survey that has been distributed to experts from five different stakeholder groups. It reveals that cost, configuration complexity, and the absence of critical functions in current implementations are the main impediments. The findings emphasise the need for modular, software-defined architectures, orchestration tools driven by artificial intelligence, and standards for unified, open application programming interfaces, in order to simplify deployment and operation.
P-1.5 Mobile Low-Latency Content Delivery in Millimeter Wave Networks
Hong-Jhih Huang and Chin-Ya Huang (National Taiwan University of Science and Technology, Taiwan)
We adopt the concept of one-shot transmission and propose a One-shot Transmission with Dual-connectivity and Network Coding (OTDNC) framework to enhance mobile real-time low-latency content delivery (such as Extended Reality, XR) in millimeter Wave (mmWave) networks. More specifically, following the network protocol stack, the presence of packet loss caused by mmWave link fluctuations introduces packet retransmission so that the end-to-end latency requirement for content delivery would be violated. In OTDNC, network encoded packets are transmitted simultaneously through dual mmWave links so that packets can be received successfully without retransmission. The proposed OTDNC is integrated into ns-3 for performance evaluation. The current results illustrate the improvement of network performance under different scenarios.
P-1.6 AI-Driven Energy-Aware Dynamic QoS Management for 6G IoT: A Proof of Concept of the 6G-VERSUS Triplet in Port Environmental Monitoring
Rafael Gonçalves (Instituto Pedro Nunes, Portugal); Diogo Duarte Fevereiro (OneSource, Portugal); Ricardo Cardoso (JSIO, Portugal); Daniel Corujo (University of Aveiro, Portugal & Instituto de Telecomunicações, Portugal); Sérgio Figueiredo (Instituto Pedro Nunes, Portugal)
The proliferation of constrained Internet of Things (IoT) devices in 5G networks demands energy-efficient communication paradigms and flexible network control mechanisms. New Radio Reduced Capability (NR RedCap) addresses device-side constraints and is particularly suited for energy-harvesting IoT deployments where available energy is inherently variable and unpredictable – paving the path for 6G-enabled continuous IoT-based sensing. However, dynamic adaptation of network behavior to real-time and predicted device energy state remains an open challenge as the interaction between the vertical user and network Application Programming Interfaces (APIs) remains limited even with current B5G/6G evolutions. This paper proposes a conceptual architecture for energy-aware Quality of Service (QoS) management for 5G NR RedCap and energy harvesting-enabled IoT devices, leveraging Network Exposure Function (NEF) for dynamic QoS control and Common API Framework (CAPIF) as the unified API exposure and discovery framework. The proposed architecture introduces an intelligent control loop in which AI-based recommendations drive context-aware energy profile adjustments through network API invocations. A proof-of-concept is presented to demonstrate the feasibility of the end-to-end pipeline, validating API invocation flows.
P-1.7 A Full Stack Standard-Compliant QKD Network
Giuseppe De Falco, Carlo Liorni, Stefano Pepe, Gaetano Russo, Federico Grasselli and Massimiliano Proietti (Leonardo S.p.A., Italy)
In this work we present the architectural design and potential applications of the Quantum Metropolitan Area Network (QMAN) under deployment in Rome. Based on Quantum Key Distribution technology, the QMAN connects industrial facilities, enabling future-proof secure communications and new applications such as Quantum Digital Signatures.
P-1.8 Coordination of Broadband Internet Information Units via the Geographical Mapping Environment
Elmars Lipenbergs (Riga Technical University, Latvia & Regulatory Authority of Latvia, Latvia); Aleksandrs Ipatovs, Vjaceslavs Bobrovs and Inga Vagale (Riga Technical University, Latvia); Guntis Ancāns (Riga Technical University & Electronic Communications Office of Latvia, Latvia); Māris Aleksandrovs (Electronic Communications Office of Latvia, Latvia); Jurgis Porins (Riga Technical University & Faculty of Electronic and Telecommunications, Latvia); Jana Lusvere and Dainis Valdmanis (Ministry of Transport of Republic of Latvia, Latvia)
The paper emphasizes inter-institutional collaboration in Latvia aimed at ensuring efficient information exchange and outlines a framework for the development of a methodology to measure mobile network coverage. It further proposes an approach for broadband coverage validation in accordance with the guidelines of the Body of European Regulators for Electronic Communications (BEREC), including the identification of underserved (“white”) areas, and examines the extension of mobile broadband coverage through the investigation of Open RAN-based technological solutions.
P-1.9 D-Band GCPW-SIW RDL (GS-RDL) Structure for RF Fan-Out Application in 3D Package
Zhaoying Li (Beihang University); Pengran Yang, Yanqin Jin and Lianggong Wen (Beihang University, China)
This paper presents a TGV-based GCPW-SIW transition structure for high-reliability D-band interconnects on glass interposers in 3D fan-out packages. By utilizing the TE10 mode of a substrate integrated waveguide (SIW), the through-glass vias (TGVs) serve as cavity boundaries rather than conduction paths, making the transmission inherently robust against poor via contact. The proposed GS-RDL design is simulated on a 200 µm BF33 glass substrate with 20-µm-diameter copper filled TGVs. Results show an insertion loss S21> -2.4 dB across the entire 110-170 GHz band. When extending the SIW length from 1000 µm to 3000 µm, the insertion loss remains better than -3 dB, demonstrating excellent length scalability up to 4.5 mm for practical packaging applications. The structure is a promising candidate for reliable millimeter-wave and THz 3D packaging.
P-1.10 Adaptive Security at the Edge for 6G-Enabled Healthcare IoT
Ijaz Ahmad (University of Oulu, Finland); Ijaz Ahmad (VTT Technical Research Centre of Finland & VTT Technical Research Center of Finland, Finland); Erkki Harjula (University of Oulu, Finland)
Healthcare IoT services increasingly rely on edge gateways to relay routine telemetry and deliver rare but time-critical alarms. Even short traffic bursts can inflate worst-case delay and interfere with urgent messages. We present NANOEDGEGUARD, a kernel-plane closed-loop controller that observes per-source traffic intensity at the edge and enforces an auditable, multi-tier rate policy using in-kernel traffic-control hooks. Unlike static firewall rules or user-space control loops, our design prioritizes rapid actuation and explicit recovery through hysteresis, and it records policy transitions for auditability. Using a Raspberry Pi gateway hosting an MQTT broker and two ESP32 endpoints generating vitals, alarms, and a timed burst, we show that adaptive kernel-plane rate control reduces the 99th-percentile alarm RTT by 13.3% compared to a user-space firewall baseline while maintaining no-enforcement-level RTT, and it reduces excess admitted burst traffic by 46% compared to no enforcement. These early results indicate that adaptive, auditable enforcement at the gateway can improve resilience for healthcare IoT, and it can be extended toward on-demand policy deployment in future edge intelligence.
P-1.11 W-Band Multi-Beam Frontend for High Capacity X-Haul Links
Claudio Paoloni, Rosa Letizia and Lei Wang (Lancaster University, United Kingdom (Great Britain))
The W-band (75 – 114 GHz) offers about 7.5 GHz available above 100 GHz (102 – 109.5 GHz) with relatively low attenuation suitable for high data rate transport both backhaul and fronthaul. A novel wireless W-band front end in the 102 – 109.5 GHz band with multibeam distribution is discussed. The transmission power is provided by a 30 W traveling wave tube (TWT) amplifier and flexibly distributed by lightweight high gain multi-beam antennas. The operation range is up to 2 kilometers and it supports up to 256QAM modulation with 39 Gbps data rate. The W-band frontend will enable high capacity transmission hubs for providing wireless X-haul links with tens of Gbps (e.g. small cells fronthaul) over large areas, overcoming limitations and cost of fiber deployment. The front end is in advance fabrication status and the antenna system has been already successfully tested.
P-1.12 Intelligent Traffic Steering Demonstration for Hybrid Terrestrial and Non-Terrestrial Networks
Md Munjure Mowla (IS-Wireless, Poland); Peretz Shekalim (Pente Networks, Israel); Robert Gdowski and Adam Flizikowski (IS-Wireless, Poland)
The integration of terrestrial networks (TN) and non-terrestrial networks (NTN) is a key enabler for meeting emerging beyond 5G and 6G requirements, including service continuity, and global coverage at any time at an affordable cost. This poster paper presents the design and demonstration of an intelligent traffic steering mechanism implemented as a Traffic Steering (TS) xApp for Open RAN-based TN and NTN, developed within the EU CELTIC-NEXT MECON project. The demonstrated system dynamically determines whether a user equipment is served by the TN and/or the NTN by jointly evaluating real-time radio and network indicators together with service-specific weighting preferences. The demonstration integrates near-real-time KPI collection from an O-RAN-based terrestrial network with simulated non-terrestrial datasets, enabling unified decision-making within a Near-RT control loop. A use-case-driven, weight-based scoring mechanism, augmented with hysteresis and a short decision hold time, is employed to ensure stable steering behavior under fluctuating network conditions. The results validate the operation of the xApp, from KPI ingestion to steering decision execution, and underline the critical role of KPI availability and reporting periodicity in achieving effective real-time traffic steering. These findings provide an initial validation of the proposed approach and inform future large-scale evaluations towards 6G hybrid TN-NTN networks.
P-1.13 An Agentic AI Approach to Data Space Access: A CyclOps Case Study
Marla Grunewald (Technische Univeristät Braunschweig, Germany); Fin Gentzen, Iulisloi Zacarias and Admela Jukan (Technische Universität Braunschweig, Germany)
The European Union (EU) is taking a clear stance on data security and data sovereignty. Various regulations and laws, such as the General Data Protection Regulation (GDPR) and the EU Data Act, strive to enable secure data access and fair data use, as well as the protection of personal data. In line with this, the EU has been strongly committed to the concept of common European data spaces since the early 2020s. The core concept of data spaces is the secure exchange of data within the EU in order to harness the potential of big data. Currently, there are several data spaces for different topics, but in the future, these shall be merged to create a single European data market. Funded by the EU, the CyclOps project develops interoperable and secure system to automate the management and governance of large-scale, distributed data. By streamlining the entire data life cycle, the project facilitates reliable data sharing and exchange across diverse data spaces. Similarly, in this paper, we present a concept whereby a user can submit a request in natural language and, using agentic AI, the correct data space is selected and a connector is used to establish contact, complete negotiations and transfer the data.
P-1.14 Characterization and Scalable Compact Modeling of III-V DHBTs Up to 220 GHz
Valentin Thary (University of Bordeaux, France)
The validation of a scalable HiCuM compact model of InGaAs/InP Double Heterojunction Bipolar Transistor over a wide range of bias, temperature and frequency is presented. The simulation of the transit frequency is investigated and compared to measurement results for frequencies up to 220 GHz. The scalable model displays a very strong agreement between simulation and measurements. It could, therefore, be used to reasonably predict the performances of different InGaAs/InP transistor geometries.
P-1.15 Hardware-Measured Energy and Latency Profiling of AI Inference for Sustainable Edge Intelligence
Mislav Has, Dora Kreković, Mario Kusek and Ivana Podnar Zarko (University of Zagreb, Croatia)
Edge AI services increasingly execute inference across heterogeneous platforms, where deployment decisions must consider both latency and energy efficiency. While latency is commonly reported, energy per inference is rarely characterized across different classes of edge hardware under comparable workloads. This work presents a hardware-measured profiling study of end-to-end inference latency and energy consumption across representative edge platforms, including microcontroller-class devices, single-board computers, high-performance embedded systems, and portable WebAssembly execution. Two lightweight time-series models (CNN and RNN) are evaluated under burst and continuous workloads, with rail- and AC-level measurements enabling comparable per-inference energy estimation across platforms. Results show that both the model architecture and execution mode strongly influence efficiency, with convolutional models outperforming recurrent models in both latency and energy. These cross-tier profiles provide quantitative indicators for energy-aware inference placement in heterogeneous edge systems.
P-1.16 Design and Validation of an Open-Source Standards-Compliant Network Slice Controller
Javier Velázquez Martínez (Telefonica Innovacion Digital, Spain); Luis Contreras (Telefonica, Spain)
This work presents the development of a prototype for the IETF Network Slice Controller (NSC) [1], an open-source software-defined network controller with support for standard interfaces (e.g. 3GPP) to orchestrate the request, realization and lifecycle control of IETF Network Slices. The paper summarizes the IETF specifications and describes the prototype developed based on them. This work also fosters integration with open-source solutions such as ETSI TeraflowSDN, of which the NSC is a constituent component, promoting alignment with community-driven initiatives. In addition, it presents several use cases supported by the NSC that demonstrate its capabilities and practical applicability.
P-1.17 Smart 6G Drones: Resilient Multi-RAT Connectivity for Airborne Use Cases
Sergiy Melnyk (German Research Center for Artificial Intelligence, Germany); Jason Rambach (German Research Center for Artificial Intelligence GmbH (DFKI), Germany); Hans Dieter Schotten (German Research Center for Artificial Intelligence, Germany); Hans D. Schotten (RPTU Kaiserslautern-Landau, Germany); Philipp Köhn (BREUER GmbH, Germany); Arndt Kritzner (Logic Way GmbH, Germany); Wolfgang Rüther-Kindel, Fabian Quaeck, Nick Stuckert and Robert Vilter (Technical University of Applied Sciences Wildau, Germany)
Combining 6G communications and UAVs enables demanding use cases with coordinated multi-drone setups, drone-borne networks and autonomous precision surveying and positioning tasks. Further enhancement is provided by 3D Digital Twin with open vocabulary capabilities. Nevertheless, reliable wireless coverage remains one of the main challenges. Multi-RAT connectivity techniques might be used to increase the resilience of the wireless connection. In this paper, we provide an overview on challenges and required technologies to make automated airborne use cases reality.
P-1.18 Reinforcement Learning for Energy Saving in Realistic O-RAN Scenarios
Ali Diab and Andreas Mitschele-Thiel (Ilmenau University of Technology, Germany); Zubair Shaik (AiVader, Germany)
Energy Saving (ES) is a core design goal in O-RAN. Traditional heuristic or rule-based Radio Units (RUs) ON/OFF control lacks scalability and adaptability under dynamic conditions, while Deep Reinforcement Learning (DRL), particularly Deep Q-Network (DQN)-based, offers a data-driven alternative. This work evaluates a DQN-based ES xApp on the ns-O-RAN platform and compares it with a baseline ES xApp under realistic as well as static traffic/mobility scenarios. Simulation results show that the DQN-based ES xApp outperforms the baseline in static scenarios. However, it exhibits unstable behavior in highly dynamic environments, violating QoS for 59% of UEs. These findings highlight the need for self-organized, cell-level RL-based ES solutions that rely on local rather than global mobility and load information.
P-1.19 A Contractive Lightweight GRU for Causal Doppler-Driven Wireless Channel Prediction
Themistoklis Charalambous (University of Cyprus, Cyprus & Aalto University, Finland); Kyriakos M Deliparaschos (Cyprus University of Technology, Cyprus); Risto Wichman (Aalto University, Finland)
Wireless channels exhibit temporal correlation induced by user mobility, multipath propagation, and Doppler effects, making short-horizon channel prediction an important enabler for adaptive physical-layer operation. In practical low-latency wireless systems, however, prediction models must satisfy strict computational and memory constraints while maintaining robustness under streaming operation. Standard recurrent neural architectures such as Long Short-Term Memory (LSTM) networks and conventional Gated Recurrent Units (GRUs) provide strong temporal modelling capability, but their hidden-state dynamics are not explicitly constrained, which may lead to sensitivity under prolonged causal deployment. This work investigates a lightweight recurrent predictor designed for stable temporal channel tracking under controlled fading dynamics. Specifically, we consider a contractive variant of the lightweight GRU, in which the recurrent candidate-state operator is spectrally constrained to enforce non-expansive hidden-state evolution.The proposed architecture is evaluated in a controlled wireless prediction setting using 3GPP TR 38.901 CDL-A channel traces. The current study intentionally focuses on a simplified scenario to isolate Doppler-driven temporal fading dynamics. Sliding windows of temporal observations are used to construct supervised learning sequences for causal recurrent prediction. To ensure efficient model configuration under limited computational budgets, Bayesian hyperparameter optimisation is employed using Optuna with Tree-Structured Parzen Estimators (TPE). The optimisation jointly tunes hidden-state dimension, dropout probability, sequence length, and batch size.
Numerical results indicate that the proposed architecture achieves predictive performance comparable to standard lightweight GRUs while preserving identical inference complexity. The contraction constraint introduces only negligible computational overhead, since spectral normalisation is applied once after training. Additional experiments comparing single-layer and deeper recurrent variants show that increasing network depth significantly raises optimisation and training cost without proportionate gains in predictive performance.
These findings suggest that shallow contractive recurrent predictors provide an attractive operating point for real-time wireless prediction under strict hardware constraints. Although the present study isolates temporal fading under ideal observations, the proposed stability-aware design naturally extends toward more realistic OFDM settings with noisy pilots, multi-subcarrier channel grids, and MIMO channel evolution.
Future work will incorporate pilot-based observations, additive noise, and multi-subcarrier prediction in order to assess robustness under practical PHY-layer conditions and compare against classical model-based predictors such as Kalman filtering.
P-1.20 Planar Low-Cost RF Sensor for Volatile Organic Compound Detection
Zabdiel Brito-Brito and Jesús Salvador Velázquez-González (Centre Tecnològic de Telecomunicacions de Catalunya (CTTC), Spain); Fermin Mira (Centre Tecnologic de Telecomunicacions de Catalunya (CTTC), Spain); Yi Wang (University of Birmingham, United Kingdom (Great Britain)); Ignacio Llamas-Garro (Centre Tecnològic de Telecomunicacions de Catalunya (CTTC), Spain)
In this manuscript, a Polydimethylsiloxane (PDMS)-coated microwave ring resonator (MRR) operating at a resonant frequency of 3.05 GHz was investigated to detect acetone vapor, a common Volatile Organic Compound (VOC), within a closed chamber at ambient temperature. Integrating a solid PDMS layer with an MRR provides an effective platform for vapor detection by leveraging absorption-induced changes in the PDMS layer’s dielectric properties. These changes shift the resonant frequency, which, in this case, occurs at 260 MHz in the presence of acetone vapor. The measured results indicate that the proof-of-concept sensor has strong potential for detecting VOCs in air at ambient temperature and for being incorporated into future 6G integrated sensing and communication systems.
P-1.21 Real-Time Crowd and Environmental Monitoring System with Automated Alert Generation
Ankit Kumar Gupta, Lokare Devesh Chandrakant and Prashant Sharma (Indian Institute of Technology Indore, India); Nayim Ahamed (IIT Indore, India); Swaminathan R (Indian Institute of Technology Indore, India)
Ensuring safety in large-scale public gatherings requires intelligent monitoring systems capable of real-time sensing, artificial intelligence (AI)-driven analytics, and ultra-low-latency communication. This work presents a fifth generation (5G)-enabled intelligent monitoring framework aligned with emerging 6G paradigms by integrating you only look once (YOLO)-based crowd detection with internet-of-things (IoT)-assisted environmental sensing. The proposed architecture employs a private 5G URLLC network and message queuing telemetry transport (MQTT)-based multi-modal data fusion to support sub-second alert dissemination, high-throughput video streaming, and significantly reduced communication latency compared to conventional wireless monitoring approaches. In addition to vision-based crowd analysis, continuous monitoring of environmental parameters, including temperature, humidity, light intensity, total dissolved solids (TDS), and soil nutrient levels, improves situational awareness and enables timely anomaly detection. Experimental evaluation demonstrates the scalability and low-latency performance of the proposed framework, highlighting its suitability for next-generation public safety surveillance and intelligent infrastructure applications.
P-1.22 5G and O-RAN Testbed for Validation of Optimization Algorithms for Unmanned Aerial Vehicle-Assisted Networks
Álvaro Durán Martínez, Félix Vidarte Vidarte, Salvador Luna-Ramírez, Marta Solera-Delgado and Miao Han (University of Malaga, Spain)
This work presents a flexible 5G Standalone (5G SA) O-RAN-compliant experimental testbed designed to support and validate optimization algorithms for 5G and O-RAN mobile networks. Experimental scenarios involving multiple UEs, realistic traffic, and mobility have been tested. A log-processing framework has been developed in Python and Bash to process and aggregate network parameters from real-time logs. The aggregated performance metrics are used to feed optimization algorithms for Unmanned Aerial Vehicle (UAV)-assisted 5G networks. The proposed platform enables reproducible experimentation and provides a practical environment for validating optimization strategies for 5G and O-RAN networks.
P-1.23 QoS-Aware Narrowband Internet of Things Uplink Scheduling in Low Earth Orbit Networks
Shih-Han Lin and Chin-Ya Huang (National Taiwan University of Science and Technology, Taiwan)
The NarrowBand Internet of Things (NB-IoT) is applied to deploy Low Earth Orbit (LEO) networks to provide wireless coverage around the world. Considering the Doppler effects and the Quality of Service (QoS) requirement of each IoT, we propose a group based Delay-Weighted Max Profit Scheduling Algorithm (DWMPSA) to transmit IoT packets in the NB-IoT over LEO networks. To schedule packets, DWMPSA first evaluates the Delay-Weighted Profit (DWP) of each UE and then allocates the resources based on profits. The simulation results demonstrate that DWMPSA effectively prioritizes high-priority packets. In the future, we will further consider the influence of handover and other scenarios to consolidate the design of DWMPSA.
P-1.24 Distributed Network-and-Compute Backbone for Emerging 6G Applications
Mandy Galkow-Schneider (Deutsche Telekom AG, Germany); Ingo Friese (Deutsche Telekom Laboratories, Germany); Sergiy Melnyk (German Research Center for Artificial Intelligence, Germany); Qiuheng Zhou (German Research Center for Artificial Intelligence (DFKI GmbH), Germany); Hans Dieter Schotten (German Research Center for Artificial Intelligence, Germany); Hans D. Schotten (RPTU Kaiserslautern-Landau, Germany); Alexander Zoubarev and Louay Bassbouss (Fraunhofer FOKUS, Germany)
Emerging applications such as immersive media, cyber-physical systems, and autonomous vehicles require communication infrastructures that tightly integrate networking and distributed computing capabilities. In this context, communication networks evolve from passive data transport systems into platforms that can dynamically support computation-intensive and latency-sensitive services. This paper presents insights and experimental results from a three-year research project investigating such integrated network-compute infrastructures. The proof of concept was validated and evaluated using two demanding application scenarios: device-independent immersive 3D communication and a collision avoidance system for mixed air traffic. Both use cases require the coordinated use of networking and computing resources under strict real-time constraints.
P-1.25 Building Network Digital Twins for Edge Computing Infrastructures: A Practical Deployment Approach
Julian Jimenez Agudelo (University of Antwerp, Belgium); Paola Soto (IMEC, Belgium & University of Antwerp, Belgium); Nina Slamnik-Krijestorac (University of Antwerp-IMEC, Belgium); Johann Marquez-Barja (University of Antwerpen & IMEC, Belgium); Miguel Camelo Botero (University of Antwerp – imec, Belgium)
This paper presents a Network Digital Twin (NDT) for edge computing environments, designed as a sandbox to evaluate the trade-off between energy consumption and latency in MEC applications. The proposed system replicates a real kubernetes (k8s) cluster instrumented with monitoring, telemetry streaming, and energy measurement components. Infrastructure and application data are collected under controlled workload patterns and used to build basic infrastructure models and a functional machine learning model. In particular, XGBoost is used to predict system behavior under different load and scaling conditions due to its good prediction accuracy and low training cost. The resulting NDT is deployed as a containerized service and used to estimate the number of replicas required to serve a given workload while analyzing the associated energy consumption and latency. This enables the evaluation of alternative configurations before applying changes to the real infrastructure. The proposed approach provides a practical step toward more efficient and sustainable management of edge computing systems.
