The applications of Unmanned Aerial Vehicles (UAVs) in urban environments have been expanding rapidly in recent years. This has led to an increasing number of drones in urban airspace, which could become an even more serious issue in the conceivable future, resulting in significant challenges for safety and traffic management. To address these challenges, one approach is to establish a road system for drones, similar to ground vehicle roads, to restrict their movements. In this paper, we propose a novel Drone Road System (DRS) design through an XML-based machine-readable representation. It provides high flexibility in system design and also enables drones to dynamically obtain updated system information through communication with infrastructure or other drones. We also conduct a simulation study to evaluate the feasibility of applying our DRS design in urban environments. The results indicate strong potential for the proposed DRS to reduce collisions while sustaining efficient throughput in urban drone traffic.
With the spread of Internet of Things (IoT) systems, congestion on random access channel due to the traffic from numerous devices is expected to become an important issue. For licensed wireless networks, such as 4th Generation (4 G) mobile phones Long Term Evolution (LTE) and 5th Generation $(5 \mathrm{G})$, the Third Generation Partnership Project (3 GPP) has proposed and standardized congestion control mechanisms for random access channels transmitting sensor data. In contrast, congestion control in unlicensed Low-Power Wide Area Networks (LPWANs), such as LoRa and Sigfox, has not been sufficiently investigated due to the small packet size and low traffic volume at present. This study proposes an adaptive regulatory control scheme to suppress congestion in the random access channel of unlicensed LPWANs, which is a concern with the further spread of IoT devices. The throughput characteristics of the proposed congestion control scheme are validated through theoretical analysis and simulations, demonstrating its stability even under overload conditions.
Internet of Things-enabled Wireless Sensor Networks (IoT-WSNs) are increasingly deployed in dynamic environments such as disaster response and habitat monitoring, where autonomous and adaptive deployment is critical. In largescale scenarios with multiple dynamic Region of Interests (RoIs), full coverage is often infeasible, requiring targeted deployment across sub-regions. Environmental changes driven by climate or mission priorities can alter these RoIs, necessitating inter-region sensor relocation. Efficiently reallocating nodes from surplus to deficient regions while minimizing movement and preserving donor-region coverage remains a key challenge. This paper proposes a fog-assisted, two-step decentralized selection and redeployment framework. A Destination Oriented Directed Acyclic Graph (DODAG)-based aggregation strategy transmits only a subset of node data, reducing network overhead, while Multi Criteria Decision Analysis (MCDA) guides sensor selection to balance residual energy, mobility, and local coverage. Simulations show that MCDA-based selections closely approximate the Pareto front, achieving an average Quartile Deviation (QD) of 0.0009 in objective gap versus 0.0129 for baselines, indicating substantially lower spread. Across multi-region scenarios, the proposed approach improves target-region coverage while limiting coverage loss in donor-regions with moderate relocation costs, demonstrating consistent and balanced performance in dynamic inter-region relocation.
The integration of inter-satellite links (ISLs) within Low Earth Orbit (LEO) constellations is anticipated to enhance global connectivity and reduce network latency. By enabling direct satellite-to-satellite communication, ISLs support in-orbit data relay and contribute to resilient network architectures. However, frequent path switching in dynamic topologies introduces latency variation (jitter), which degrades time-sensitive services. To address this challenge, we propose a novel delay-threshold path selection strategy that triggers a switch only when a candidate route outperforms the current path by more than a defined threshold, thereby jointly minimizing delay and jitter. Alongside this approach, we evaluate two baseline strategies: (i) dynamic shortest-path strategy that minimizes instantaneous delay, and (ii) persistent-path strategy that reduces handovers by maintaining the current path while feasible. Simulations on a Walker-Delta constellation show that the proposed strategy reduces jitter by $43 \%$ and handovers by $56 \%$, while keeping mean end-to-end delay within $5 \%$ of the shortest-path baseline. These results demonstrate that the proposed strategy substantially improves temporal stability with minimal latency penalty.
Ever-increasing traffic demand within datacenters (DCs) necessitates intra-DC network architectures that are scalable, energy-efficient, and cost-effective. In this paper, we propose a novel, evolvable intra-datacenter network architecture that effectively utilizes wavelength selective switches (WSSs). By optimizing the network topology to minimize WSS cascading, our approach significantly reduces network cost and energy consumption. A key innovation is the evolvable network design method, which enables us to scale dynamically while maintaining operational continuity and minimizing cost overhead. Intensive performance evaluations demonstrate that our proposed method reduces total energy consumption by approximately $6 \%$ and network cost by $26 \%$ compared to the conventional models. While the evolvable design introduces a modest $12 \%$ overhead in network resource usage compared to a static, green-field design, it guarantees service continuity during expansion. This research presents a practical framework for building future intra-DC networks that are both operationally flexible and economically sustainable.
To optimise the use of drones, flocking behaviour enhances the efficiency of task execution. Inter-drone collision avoidance is a critical requirement in any flocking algorithm. Conventional approaches typically adjust individual trajectories to prevent collisions. This paper investigates the feasibility of using a start-delay scheduling approach, in which drones follow straight-line trajectories with optimised velocity profiles. Two major constraints that could result in circular dependencies are identified. The overall system relationships between drones are represented as an adjacency matrix, and the presence of cycles is detected using a topological sorting algorithm. Simulation results across fleet sizes from 25 to 5,000 drones identify the density ranges that support cycle-free scheduling, leading to the derivation of a density function that guarantees at least $95 \%$ of configurations remain cycle-free. The proposed analysis provides a foundation for collision-free flock scheduling and establishes the basis for algorithms that assign individual time delays to drones while maintaining efficient operation.
Traditional federated learning (FL) architectures are vulnerable to significant security risks. Centralized servers create single points of failure and are susceptible to adversarial attacks such as model poisoning. Additionally, they lack auditability and do not provide cryptographic privacy during model aggregation. This paper proposes a serverless FL framework that integrates blockchain technology with CKKS homomorphic encryption to address these challenges. The framework utilizes a decentralized consensus mechanism based on Practical Byzantine Fault Tolerance (PBFT), eliminating the need for a central coordinator and ensuring robustness against Byzantine faults. It incorporates CKKS encryption to enable secure aggregation through computation on encrypted gradients, preserving data privacy throughout the training process. Furthermore, it introduces an auditable trust management system using smart contracts for immutable logging and decentralized reputation tracking. Experimental results on the MNIST dataset, with $20 \%$ malicious clients, demonstrate that the framework achieves a final global accuracy of $98.68 \%$ and attains $100 \%$ precision in rejecting malicious updates without false positives. The computational overhead remains practical, with encryption and consensus times of 0.814 sec and 1.695 $ms$ per round, respectively. Compared to existing methods, the proposed approach improves accuracy by $1.68 \%$ to $10.68 \%$, demonstrating its effectiveness in enabling secure and privacypreserving FL.
The opacity of Machine Learning (ML)-based Internet of Things (IoT) Intrusion Detection Systems (IDSs) necessitates Explainable AI techniques like SHAP. However, SHAP explanations for models trained on high-dimensional, onehot encoded data are fragmented and unstable. A recent method, FusionSHAP, was introduced to aggregate SHAP values of the one-hot encoded features back into their respective semantic groups. This method improves the high-level interpretability of the explanations. We extend existing work through quantitative and qualitative evaluation of feature grouping method across multiple dimensions. We investigate its effect on faithfulness, complexity and robustness, its performance across different ML models (Random Forest vs. XGBoost) and attack types (DDoS vs. Ransomware), and its superiority over label encoding. Our evaluation of 30 random seeds demonstrates that feature-grouping preserves faithfulness, significantly improves robustness (reducing worst-case sensitivity by $\mathbf{3 0 \%}$ for Random Forest) and outperforms label encoding in worst-case robustness. The trade-off of increased complexity is consistent across all experiments and is favourable for the substantial gains in robustness, which are paramount in cybersecurity context.
Ethereum has emerged as one of the most widely used blockchain platforms, underpinning decentralized finance, smart contracts, and distributed applications. With its growing adoption, the Ethereum peer-to-peer network is susceptible to networklayer attacks including eclipse (node-isolation) attacks. To study the threats to Ethereum and to develop effective detection and mitigation strategies, researchers require controlled, reproducible, and labeled network datasets. However, datasets are scarce due to the complexity of capturing live blockchain traffic and the difficulty of confidently labeling malicious activity on public networks. In this paper, we present the design and deployment of a private Ethereum testbed for dataset collection. Our testbed consists of five virtual machines running Geth clients interconnected via a controlled gateway: four nodes act as benign Ethereum peers and one node acts as a malicious entity that performs eclipse attacks. The testbed emulates normal blockchain operations (block propagation, transaction exchanges, and peer discovery) and adversarial scenarios focused on node isolation. Wireshark is deployed on the gateway to capture the network traffic, enabling us to record raw packet traces for benign and attack scenarios. The resulting dataset provides a comprehensive view of Ethereum network-layer behavior, with traffic labeled according to ground truth (node role and attack phase). We describe the testbed, the attack procedure for generating eclipse conditions, the capture and labeling pipeline, and potential uses of the dataset for intrusion detection and resilience analysis.
Adversarial Information Retrieval (AIR) systems are designed to retrieve, index, and detect malicious web sites. These web sites in response, employ cloaking techniques such as geolocation cloaking to thwart or minimise detection by AIR systems. Therefore for an AIR system, it is essential to mimic attributes associated with a client appropriately to be subjected to the attack. This paper uses a systematic approach, i.e., Failure Mode and Effects Analysis (FMEA) to discuss logical and physical attributes which can influence and deviate a study on geolocation cloaking from its intended purpose and result in potential failure in detection. A set of actions have also been proposed to mitigate or minimise the likelihood or the effect of the failures.
Digital twins (DTs) have emerged as indispensable assets for monitoring and control within industrial environments, particularly in the context of human-centric smart manufacturing. The effectiveness of DTs relies on how accurately the virtual model reflects its physical counterpart, a challenge exacerbated by constrained network resources, unpredictable communication channels, and rapidly changing factory conditions. In this study, synchronisation fidelity is established as the primary objective for resource scheduling in an on-premises private 5 G network. Our approach applies a context-aware resource blocks (RBs) allocation mechanism that connects the communication demands of Industry 5.0 applications and real-time operational variability to the network behaviour. This work-in-progress paper presents a system architecture that employs dynamic prioritisation for resource allocation and lightweight extrapolation strategies to address data blind spots, thereby preserving digital twin fidelity as shop floor conditions evolve.
Extended Reality (XR) is emerging as a key class of immersive applications for next-generation networks. Reflecting it’s importance, ITU-R IMT-2030 vision identifies XR as a key service class for 6 G, and 3 GPP has recognized XR as a critical use case. However, conventional schedulers such as Proportional Fair (PF), and MAX-CQI etc often fail to meet XR’s ultra-low latency, high throughput, and high reliability demands. Existing frame-level integrated transmission strategies improve XR support by prioritizing frame completion but still focus on network-level Quality of Service (QoS), overlooking user-perceived Quality of Experience (QoE). Since not all XR data equally contribute to perceptual quality, semantic communication offers a promising solution by prioritizing semantically important content. In this paper, we propose a novel semantic-aware MAC scheduling framework for XR over 6 G networks that integrates semantic relevance extracted using a 3D convolutional neural network (3D-CNN) for frame prediction with traditional network metrics to optimize resource allocation. Simulation results demonstrate significant QoE improvements and achieving higher gain compared to PF (upto $\mathbf{1 0 0 \%}$), MAX-CQI (upto 88%), and frame-level integrated schedulers (upto $87 \%$).
Data distillation is an efficient machine learning technique that generates a compact yet representative subset of a large dataset, reducing computational overhead while preserving essential features. This work proposes a novel state-of-the-art data distillation approach that enables model training on significantly fewer data without compromising performance. Pneumonia and Brain Tumor datasets were used to evaluate the performance of distilled data using three machine learning models, i.e., ResNet50, MobileNetV2, and VGG16. The model is trained on the full dataset and the distilled subset, with a comparative performance analysis. Experimental results show that the distilled dataset achieves comparable accuracy to that of the full dataset. This demonstrates the efficiency of the proposed method in maintaining high predictive performance while significantly reducing memory usage, computational cost, and training time. Additionally, by minimizing data exposure, the method enhances data privacy and supports privacy-aware learning, which is crucial in sensitive domains like healthcare. The lightweight nature of the distilled data also makes the approach well-suited for deployment on edge devices, enabling on-device model training and inference in real-time, even in low-resource environments.
With the rapid growth of maritime activity, maintaining reliable ship-to-shore connectivity faces growing challenges from shadowing caused by passing tall vessels, offshore structures, and other maritime obstructions. An analysis of data for one month from the Norwegian Coastal Administration Automatic Identification System (AIS) shows that shadowing occurs in more than $60 \%$ of transmissions in Norwegian coastal waters. These frequent Non-Line-of-Sight (NLoS) conditions cause severe signal attenuation and undermine communication reliability. To address this, we investigate the use of Simultaneously Transmitting and Reflecting Reconfigurable Intelligent Surfaces aided Non-Orthogonal Multiple Access (STAR-RISNOMA) to enhance both coverage in 360° NLoS regions and system capacity. By simultaneously transmitting and reflecting incident signals, STAR-RIS can effectively reduce signal blockage and combat fading, while the integration of NOMA further enhances spectral efficiency. Simulation results show that STARRIS aided NOMA consistently delivers gains over conventional OMA schemes, improving near-user capacity by up to $5 \mathrm{bps} / \mathrm{Hz}$ in shadowed maritime scenarios. These results highlight STAR-RIS-NOMA as a scalable and practical solution for high-capacity, shadowing-resilient ship-to-shore communications.
The transition toward 6G is driven by the need for ubiquitous global coverage which necessitates the seamless integration of Terrestrial Networks (TN) and Non-Terrestrial Networks (NTN). Historically, TN and NTN have advanced as separate networks with distinct standards, infrastructural requirements, cost structures, and pricing models - forcing users who need both services to maintain multiple subscriptions and billing arrangements. Given NTN’s high capital costs and global scope, partnerships with local Mobile Network Operators (MNOs) are essential to leverage their existing subscriber bases motivating the need for a unified pricing framework that supports multi-operator interactions in Integrated TN and NTN (ITNTN). This paper proposes a Stackelberg game-theoretic pricing model for an ITNTN that accommodates multiple LEO operators and an MNO with heterogeneous user requirements under quality-of-service (QoS) and budget constraints. The MNO acts as the leader, setting end-user prices while procuring wholesale NTN resources from competing LEO operators. The model jointly optimizes radio resource acquisition, budget-constrained and heterogeneous end-user pricing, and cross-domain reallocation to maximize MNO profit while ensuring affordability through price discrimination. Results from simulations show that the proposed approach enhances MNO profitability and enables seamless multi-domain connectivity under a unified subscription model.
The growing scale of cyber attacks demands automated behavioural analysis to enable timely mitigation. Current approaches often rely on Cyber Threat Intelligence (CTI) reports and Common Vulnerabilities and Exposures (CVE) descriptions, but these depend heavily on expert input, introducing subjectivity and limiting scalability. We propose an automated method that analyses honeypot-captured attacks by mapping Shell commands to Linux man page descriptions and aligning them with the MITRE ATT&CK framework using the Semantic Mapping of Exposures to Techniques (SMET) tool. To provide richer context, command descriptions are also chained into reports. Evaluation on a public dataset achieved up to $82 \%$ similarity with manual annotations.
This paper deals with scalability and security for large-scale deployments of Federated Learning (FL) and proposes a two-tier secure FL architecture that incorporates edge servers as intermediate aggregators between clients and the cloud. This hierarchical framework enhances scalability and minimizes cloud communication by utilizing edge-level aggregation. Security is maintained with lightweight cryptographic protocols, including X25519 for key exchange, HKDF with BLAKE2b for key derivation, and ChaCha20-Poly1305 for authenticated encryption, ensuring end-to-end confidentiality and integrity of model updates. An experimental evaluation on the MNIST dataset across three deployment scenarios (ranging from 9 to 100 clients) reveals improvements in model accuracy from 93.90% to 96.04%. Communication overhead to the cloud was reduced by up to 90%, and cryptographic overhead remained under 5.3ms per operation. Additionally, the architecture achieved 100% resistance to model poisoning, 97.8% prevention of gradient leakage, and 100% confidentiality preservation. The proposed approach outperformed existing state-of-the-art methods by a margin of 1.27% to 7.64% and demonstrated a strong balance between performance, security, and communication efficiency in edge-cloud FL environments.
Access to high-quality, realistic datasets is critical for cybersecurity research, however real-world data is often constrained by privacy, classification, and legal restrictions. These limitations contribute to the outdated, imbalanced, or limited realism of widely available datasets. This paper reviews the emerging use of LLMs as a mechanism for generating synthetic cybersecurity data, with a particular emphasis on augmenting existing datasets. The review identifies phishing emails and system logs as the most suitable data types for LLM-driven augmentation due to their structured, textual nature. Outside the specific cybersecurity domain, LLMs have demonstrated aptitude for generating realistic, label-rich artefacts, addressing challenges such as class imbalance, labelling inconsistency, and limited data diversity. These capabilities could be readily applied to cybersecurity datasets. This review consolidates key studies in the field and identifies future directions for integrating LLMs into cybersecurity dataset pipelines, particularly through automated labelling, metadata enrichment, and hybrid generative architectures.
With the growing reliance on global navigation satellite systems, the need for alternative localization methods has become critical. Doppler based localization is attracting increasing interest; however, its performance is constrained by a limited frequency resolution, which introduces noise and reduces positioning accuracy. This paper presents a novel Doppler extraction framework that integrates matched filtering, frequency agile filter, and a phase locked loop to mitigate these limitations. The estimated Doppler profile is used to localize a terrestrial network device based on the unique fingerprint captured by multiple satellites. We show that matched filtering and frequency agile filter enhances the frequency resolution, while the phase locked loop refines extracted frequencies by tracking phase deviation over time. The proposed approach is evaluated in a simulated satellite based localization scenario, demonstrating a significant improvement in localization accuracy compared to the Fast Fourier Transform based method. Monte Carlo simulations results show that employing four localization iterations and six satellites can achieve a reliable localization of network devices.
The scalability of Quantum Key Distribution (QKD) networks remains a primary challenge for the realization of the quantum internet. Although full mesh topologies offer superior performance, their quadratic scaling of quantum links renders them impractical for large-scale deployment. This paper proposes a novel Hierarchical Quantum Backbone (HQB) topology that addresses this challenge. It is being done by employing a hierarchical architecture composed of a core ring of quantum repeaters interconnected with distribution hubs and trusted nodes serving end-users. A statistical simulation framework incorporating realistic models for channel loss, latency, and quantum bit error rate evaluates the design’s eficiency. The performance evaluation demonstrate that HQB reduces the number of quantum links compared to full-mesh networks. The topology supports near-instantaneous, high-rate intra-hub key distribution as well as metropolitan-scale inter-hub rates. This comprehensive analysis confirms HQB as a practical, efficient, and fault-tolerant architecture suitable for scalable quantum network deployment.
