Low Earth Orbit (LEO) satellite networks such as Starlink allow global and affordable internet access. This not only connects remote residents, but also enables new potentials such as connecting non-stationary agricultural systems in rural areas for high-precision spot farming. One challenge is that tractors and farming robots need to transmit large amounts of data from cameras and laser scanners in real-time, which easily overwhelms the uplink capabilities of a single LEO connection. In this paper, we examine the Starlink performance of multiple terminals within a single service cell to lay ground for multipath link bundling as a possible approach to increase link capabilities. For this purpose, we present two measurement setups consisting of four and seven Starlink dishes placed in one service cell. We conduct multiple measurements and evaluate UDP and TCP throughput, latency, and packet loss. We find that UDP throughput scales almost linearly with the number of dishes, indicating significant bundling potential up to an expected limit. In contrast, TCP BBRv1’s bundling potential is limited, due to inefficiencies in utilizing the links. Moreover, we find that some packet loss events are synchronized within a service cell. For future and independent research, we make our dataset publicly available.
Energy consumption reduction in 5G networks and beyond is a significant challenge. Discontinuous Reception (DRX) established by 3GPP is the current state of the art energy-saving strategy. With DRX, users that neither receive nor transmit for a certain duration are set in a sleep cycle. These users periodically wake up in order to consult the Physical Downlink Control Channel in case of potential transmissions. DRX timings are negotiated during the initial connection. Despite the inherent variability of user traffic patterns, DRX configuration parameters remain static, leading to energy waste. In addition, DRX does not consider delay constraints, which may incur QoS degradation due to inadequate sleep duration. This paper proposes Meta Sleep Scheduler (MSS) as an alternative to DRX. User sleep duration is dynamically adjusted to meet application time constraints while reducing the time users spend in high power states. Similarly to DRX, MSS is compatible with all radio resources allocation schedulers. MSS achieves an average reduction of up to 90% in user standby power consumption compared to DRX, while maintaining or surpassing DRX’s Quality of Experience.
Edge caching in Next-Generation Networks (NGNs) deploys cache units in Small-Cell Base Stations (SBSs), enabling User Equipment (UE) to retrieve content locally and allowing neighboring SBSs to share cached data. However, the large-scale deployment of SBSs and the diverse content preferences of UEs create significant challenges for cache strategy design by hindering accurate semantic representation during SBS collaboration, which in turn leads to redundant caching and suboptimal decisions in network environments. To address these issues, we propose a value decomposition-based reward allocation method to optimize network costs. Specifically, we design an agent model capable of explicitly exchanging information to enhance the semantic learning capabilities of SBSs. We then introduce a Neural Attention Additive Q-learning (NA2Q) model within the Advantage Actor-Critic (A2C) framework, which decomposes joint action values to capture the nonlinear relationships arising from SBS interactions. The model employs a Variational Auto-Encoder (VAE) to construct latent local semantics for each agent and integrates a self-attention mechanism to evaluate each agent’s credit. Finally, a credit-based reward allocation mechanism dynamically assesses SBS contributions, addressing traditional methods’ shortcomings in modeling inter-agent dependencies. Experimental results demonstrate that the proposed method significantly improves cache hit rates while reducing backhaul traffic, outperforming baseline methods.
For various wireless sensing and IoT applications, LoRa has emerged as an energy-efficient and long range solution for wireless data transfers. However, real-world deployments face multiple challenges, including packet collisions and variable link qualities, which generally lead to packet corruptions and data loss. Even though LoRa relies on forward error correction to restore corrupted packets, its usage comes with a significant energy overhead and only provides limited capabilities. In our paper, we present CEC-LoRa as an alternative to LoRa’s forward error correction feature. Its key innovation is that CEC-LoRa allows to restore corrupted packets without any error correction codes. Our approach exploits two symbiotic properties in LoRa’s encoding and chirp-based modulation scheme. On the one hand, misinterpreted chirps are often confused with neighboring symbols. On the other hand, neighboring symbol values only differ by one, keeping high similarity between misinterpreted and correct symbols. We leverage these properties by pre-computing a set of all plausible LoRa packets based on the expected payload permutations and compare them with the received packet. This allows us to identify the closest match and thus the most probable candidate packet. While the pre-computations incur a quite substantial energy overhead, CEC-LoRa shifts this from the energy-constrained end devices to the receiver side. The benefits greatly outweigh the energy demand, though: CEC-LoRa reduces the number of corrupted packets by 86.5 % on average without any additional overhead on the end device and while still being fully compliant to the LoRa specification. By including symbol information, CEC-LoRa can reduce the number of corrupt packets even further, by 99.5 % on average. This translates into SNR gains in the range of 0.79 dB to 0.93 dB.
The highly dynamic characteristics of the maritime environment present significant challenges for vessel trajectory prediction. Traditional statistical and machine learning models often struggle to adapt to changing conditions and new data streams, leading to performance degradation. To address these well known issues, we propose the use of Continual Learning that enables the system to learn incrementally from sequential data streams. Our proposal avoids catastrophic forgetting of previously acquired knowledge through a replay-based approach. This strategy ensures that the prediction model can track and adapt to shifting environmental factors and variations in vessel behavior. We test the Continual Learning-based model using high-frequency trajectory data recorded by a cruise vessel Voyage Data Recorder. Experimental results indicate that our approach achieves a lower error compared to conventional static learning models. It mitigates catastrophic forgetting, ensuring the retention of critical information from past vessel movements, and demonstrates a strong capacity to adapt to data shifts inherent in real-world maritime operations. These findings highlight the potential of Continual Learning to enhance the reliability and robustness of vessel trajectory prediction systems in an ever-changing maritime landscape.
The volume of network traffic and elusiveness of modern cyber threats challenge the ability of cyber analysts to identify anomalous behavior in networks. In this work, we propose an autonomous agent to enhance analysts’ detection capabilities. To this end, we designed an agent to analyze feature-rich network traffic by autonomously executing the Data Analytic Development Process: (1) data engineering–prepares datasets, (2) feature engineering–selects the most representative features, (3) model engineering–chooses the best algorithm-feature combination, and (4) analytic deployment–applies the optimized analytic to the dataset and identifies anomalous behavior. We refer to this agent as an Autonomous Data Scientist (ADaS), utilizing reinforcement learning as an orchestrator to determine the optimal combination of features and unsupervised clustering algorithm. ADaS operates without labeled training data, and results on popular public datasets (NB15, IoT-23, and KDD’99) demonstrate high detection and low false alarm rates, making it reliable and novel for anomaly detection.
Network Intrusion Detection Systems face increasing challenges from sophisticated evasion techniques that manipulate traffic timing patterns. This paper presents a Temporal Evasion Generation Algorithm (TEGA) for creating adversarial examples by exploiting temporal vulnerabilities, and an Adaptive Temporal Defense System (ATDS) to counter these attacks. We formalize temporal evasion mathematically and evaluate both systems using the CIC-IDS2018 dataset. TEGA achieves evasion success rates of 72.4%, significantly outperforming conventional techniques such as standard delay injection (45.6%) and burst pattern manipulation (53.2%). Conversely, ATDS demonstrates robust defense capabilities, with detection accuracy reaching 95% after adaptation and false positive rates reduced to 1.5%. Our comparative analysis reveals that sequence-based feature extraction combined with SVM classification provides optimal resilience against temporal evasion. The adaptive framework rapidly responds to new attack patterns, typically requiring only 2-3 update cycles to achieve over 90% detection accuracy. This research contributes to network security by addressing an emerging attack vector while offering promising directions for developing next-generation intrusion detection systems.
Incorporating Device-to-Device (D2D) communication and Unmanned Aerial Vehicles (UAVs) into next-generation cellular networks is crucial for addressing the growing demand for high-data-rate applications. However, increasing communication distances in both cellular and D2D environments raises transmission power requirements, leading to higher energy consumption and reduced network efficiency. Additionally, in the conventional technique, UAVs communicate directly with each UE individually, further exacerbating the energy demands of UAVs. To overcome these challenges, this paper presents an energy-efficient clustering technique designed to minimize power consumption for both UEs and UAVs. Hypergraph-based clustering enables adaptive grouping by leveraging received signal strength, while the Whale Optimization Algorithm (WOA) selects central users based on UE distance, residual energy, and connectivity metrics. This structured clustering strategy enhances system throughput and optimizes energy efficiency. Extensive simulations validate the proposed model, demonstrating significant improvements in energy efficiency, reduced computational complexity, and superior network performance compared to existing methods.
Recent advances in Byzantine Fault-Tolerant State Machine Replication have led to practical protocols for partially synchronous or asynchronous networks by combining classical methods with modern cryptographic tools like verifiable random functions. Despite improved performance in throughput and latency, these protocols remain limited by the FLP impossibility and the Dwork-Lynch-Stockmeyer bound, tolerating at most ⌊(n − 1) /3⌋ adversaries. In contrast, Proof-of-Work (PoW) achieves up to ⌊(n − 1) /2⌋ fault tolerance under asynchrony, albeit with high energy costs and probabilistic finality. Protocols like Avalanche similarly exceed the ⌊(n − 1) /3⌋ bound by relaxing deterministic guarantees. In this paper, we propose Halpha, a leaderless consensus protocol that tolerates nearly half Byzantine faults with overwhelming probability. Unlike PoW and Avalanche, which provide probabilistic finality post-termination, Halpha relaxes liveness during proposal. It leverages a probabilistic multi-valued validated Byzantine agreement (P-MVBA) to propose transaction sets, ensuring each referenced transaction is seen by at least one honest node. While P-MVBA may halt, it progresses under the intermittent existence of a bounded-delay interval, and ensures consistency with overwhelming probability without synchrony. Halpha then runs multiple asynchronous binary agreement instances, using verifiable random functions to support randomized liveness under Byzantine conditions and achieves finality when synchrony arrives. Experiments show Halpha achieves safety with high probability and low-latency deterministic finality in normal cases.
Nowadays, data centers contribute to a large portion of global carbon emissions due to the growing demand for cloud services. Traditional green data center technologies have utilized all kinds of methods to reduce energy consumption, e.g., optimizing HVAC (Heating, Ventilation, and Air Conditioning), shutting down idle servers or switches and using renewable energy, but they typically target a single data center and have pushed the optimization space to its limits. However, carbon emissions vary wildly across data centers due to differences in local energy sources, and delay-tolerant tasks can be flexibly allocated to data centers with lower emissions. This opens new opportunities for carbon-aware task scheduling across geographical locations. However, carbon-aware scheduling faces key challenges, including the fluctuation of carbon intensity due to weather and power dynamics, and the dynamic nature of task arrivals with diverse QoS (Quality of Service) requirements. To address this, we propose CEDTS-RL (Carbon-Emission Driven Task Scheduling based on Reinforcement Learning), a real-time cross-geographical scheduling framework based on reinforcement learning, which unlocks new potential for reducing carbon emissions by relaxing the location restriction during scheduling. The evaluation is conducted using real-world datasets from Google, Alibaba, and Microsoft, along with solar and meteorological data from NASA. Results demonstrate that CEDTS-RL significantly outperforms both traditional and recent scheduling algorithms.
The collaboration between end devices and edge servers has been extensively investigated to enable adaptive service provisioning, particularly in scenarios requiring trade-offs between differential model accuracies and heterogeneous resource consumption costs. In this paper, we propose freshness-aware hierarchical inference service provisioning in a Mobile Edge Computing (MEC) network, where inference models are trained and maintained in edge servers to address resource limitations on devices. Devices dynamically download updated models to maintain local inference fidelity, mitigating performance degradation caused by model staleness. We formulate a freshness and cost minimization problem that maximizes overall inference fidelity while minimizing total costs, subject to long-term average energy budgets on devices and computing capacities on cloudlets. We design an online algorithm with provable competitive ratio, by leveraging Lyapunov optimization and randomized rounding techniques. We conduct simulations to evaluate the performance of the proposed online algorithm. Simulation results demonstrate that the proposed algorithm is promising.
Although LoRa is predominantly employed with the single-hop LoRaWAN protocol, recent advancements have extended its application to multi-hop mesh topologies. Designing efficient routing for LoRa mesh networks remains challenging due to LoRa’s low data rate and ALOHA-based MAC. Prior work often adapts conventional protocols for low-traffic, aboveground networks with strict duty cycle constraints or uses flooding-based methods in subterranean environments. However, these approaches inefficiently utilize the limited available network bandwidth in these low-data-rate networks due to excessive control overhead, acknowledgments, and redundant retransmissions. In this paper, we introduce a novel position- and energy-aware routing strategy tailored for subterranean LoRa mesh networks aimed at enhancing maximum throughput and power efficiency while also maintaining high packet delivery ratios. Our mechanism begins with a lightweight position learning phase, during which LoRa repeaters ascertain their relative positions and gather routing information. Afterwards, the network becomes fully operational with adaptive routing, leveraging standby LoRa repeaters for recovery from packet collisions and losses, and energy-aware route switching to balance battery depletion across repeaters. The simulation results on a representative subterranean network demonstrate a 185% increase in maximum throughput and a 75% reduction in energy consumption compared to a previously optimized flooding-based approach for high traffic.
Mission-critical systems (MCSs) have evolving latency and reliability requirements, even under challenging conditions such as node and link failures and cyberattacks. To fulfill these requirements, emerging networking technologies like the IEEE 802.1 Time-Sensitive Networking standards provide several protocols for deterministic communication on top of off-the-shelf Ethernet equipment. While Ethernet-based networks offer better configurability than legacy fieldbus systems, they still require the design of adequate topologies for MCSs that fulfill various design objectives such as optimal quality of service and increased resilience against challenges. In this paper, we propose MITHRIL, a multi-objective topology synthesis model with reinforcement learning. It leverages deep reinforcement learning to optimize Ethernet-based topologies in terms of resilience and effectiveness, while adhering to realistic MCS constraints. Our evaluation indicates that MITHRIL enhances the failure and attack tolerance of network topologies while reducing the associated costs, compared to well-connected topologies and other heuristics from the literature.
Unmanned Aerial Vehicles enable a wide range of applications, including search and rescue, environmental monitoring, and disaster response. These aerial platforms can form dynamic flying ad hoc networks to support both sensing and data communication. In particular, UAVs can establish a resilient communication backbone that adapts to varying propagation conditions and fluctuating traffic demands. However, maintaining reliable performance in such highly dynamic and unpredictable environments remains a critical challenge. This work investigates online link quality estimation by autonomous agents, with a focus on real-time detection of link model changes (e.g., due to mobility or interference) through model drift. Building on this, we propose and evaluate adaptive formation control strategies that adjust UAV placement to optimize the network’s Packet Delivery Ratio. Simulation results demonstrate that the proposed optimal placement strategy significantly outperforms baseline approaches for line-based UAV networks.
Phishing attacks continue to be a prevalent cyber-security threat that mimics trusted websites to steal sensitive user information. This results in considerable personal and organisational damage. Traditional countermeasures like URL blacklisting and recent machine learning and deep learning models are limited in terms of generalisability. While more recent large language models (LLMs) have been investigated for phishing detection, few have evaluated LLM-based URL embeddings for phishing detection. To address this issue, in this paper, we evaluate three different LLMs using three URL datasets, for phishing detection based on URL embeddings extracted from fine-tuned models. Our evaluations showed Gemma-2B to be the best-performing model with an average F1 score of 0.91 across all test sets. Using the best-performing model we further demonstrate two model implementations: a webpage implementation for users to determine if a URL is phishing or not, and a Gmail API integration to classify which emails are spam.
L-band Digital Aeronautical Communications System (LDACS) is the selected Air-to-Ground (A2G) technology for future aeronautical communications and a proposed candidate for Air-to-Air (A2A) links. Geographic greedy routing in sparse LDACS A2A networks, typical during gradual system deployment, frequently encounters local minima, necessitating backup mechanisms that are inefficient. Previous research has primarily focused on refining backup mechanisms, neglecting the root cause of geographic greedy routing failures. This paper investigates why geographic greedy routing performance deteriorates in sparse network scenarios, where failures occur more frequently than in dense deployments. We introduce a novel metric to quantify the quality of hop-by-hop forwarding decisions in geographic greedy routing. Furthermore, we develop a second-order absorbing Markov chain model to predict the success ratio and hop stretch factor. The model is validated through Monte-Carlo simulations over the French airspace with varying LDACS equipage fractions, achieving an average difference from simulation results of less than 3.4% for the success ratio and 1.5% for the hop stretch factor. The proposed model demonstrates high accuracy and can be generalized to evaluate other geographic routing protocols. Consequently, the outcomes provide valuable insights toward designing optimized geographic greedy routing protocols.
The growing reliance on digital technologies onboard vessels has significantly increased their attack surface. As a result, both IT and OT systems are now vulnerable to a range of cyberattacks. However, existing methods used to assess vulnerability and threats often rely on outdated threat or vulnerability information, limiting their effectiveness. Consequently, a more proactive approach to assessing the security of vessel systems is needed. Threat hunting offers a proactive way of gathering the latest threat and vulnerability data from operational maritime vessels, which can be used for comprehensive security assessments. However, there is a lack of systems specifically designed to perform both threat-hunting and security assessment operations. In this paper, we propose a threat-hunting and security assessment framework that collects and processes real-time data from vessels and conducts security analysis using a graphical security model designed for vessel systems. Our approach demonstrates how the collected information can be used to evaluate a ship’s security posture by simulating potential attack scenarios and understanding how an adversary might attempt to compromise the vessel’s network. It also provides a foundation for more informed, data-driven cybersecurity strategies for the unique systems found onboard maritime vessels.
Modern data centre networks demand ultra-low latency and high throughput, making Remote Direct Memory Access (RDMA) increasingly important. However, RDMA’s performance degrades significantly under congestion due to limited built-in congestion control. Existing approaches, based on Explicit Congestion Notification (ECN) and Round-Trip Time (RTT) suffer from limitations including delayed feedback and noisy signals when used alone. We propose Dynamic Hybrid ECN-RTT Congestion Control (DHERC), a novel algorithm that integrates RTT-based early detection with ECN-based confirmation to achieve balanced congestion control. Through extensive simulations, we demonstrate that DHERC outperforms state-of-the-art schemes across key performance metrics. DHERC achieves better flow completion time performance, improved throughput fairness, and good goodput efficiency while maintaining competitive tail performance. The results establish multi-signal congestion control as a promising approach for RDMA networks requiring predictable performance, efficiency, and fairness.
This paper presents the implementation of various quantised neural network-based Network Intrusion Detection Systems on a SmartNIC using the P4 programming language. SmartNICs impose significant memory and computational constraints, making deep learning integration particularly challenging. Prior work has been restricted to shallow networks and 1-bit quantisation due to these limitations. In contrast, we deploy a binary neural network with a 64–16–8–1 architecture, along with several other quantised models not previously implemented on SmartNICs. To overcome the limited support for quantisation-aware training, which is restricted to 8-bit, we introduce a variable quantiser scheme for neural networks that allows adaptable quantisation levels during training. We further deploy and evaluate 2-bit and 4-bit models on the SmartNIC, made feasible by optimising resource utilisation. The model achieves a 91.65% accuracy and a 91.46% F1 score while achieving throughputs above 6Mpps.
