Prioritized EDCA (P-EDCA) is a proposed IEEE 802.11bn mechanism that provides a short protected contention period for low-latency (LL) traffic. We investigate an authorization policy that invokes one protected opportunity when the preceding channel-unavailability interval exceeds a threshold. The analysis couples the trigger-conditioned distribution of active LL stations with the unique-minimum outcome of one-shot discrete backoff. Because collision probability is nonlinear in the number of contenders, the protected window is selected from the full distribution rather than its mean. We then develop a distribution-aware controller that adjusts the protected window and trigger threshold using a collision constraint and head-of-line urgency. A discrete-event simulator with persistent EDCA backoff, retransmissions, bursty arrivals, and multiple unavailability-duration models evaluates the mechanism under matched normal-EDCA parameters. For 5–20 LL stations in stable mixed traffic, adaptive P-EDCA reduces 99th-percentile LL delay by 5.2%–15.9% and the 10-ms deadline-miss ratio by 10.0%–12.3% relative to the same tuned-EDCA configuration without the protected opportunity. Paired 95% confidence intervals show no statistically significant change in non-LL throughput. The gains persist across the tested traffic and unavailability models, whereas one protected opportunity is insufficient under persistent overload.
Network intrusion detection is critically constrained by data imbalance, where minority attack classes suffer from insufficient training samples. Existing data augmentation methods typically focus on single-modal representations, failing to leverage complementary information in dual-modal network traffic data. Moreover, real-world class imbalance prevents advanced multi-modal fusion architectures from achieving their full potential. To address these challenges, this paper proposes Dual-Modal Representation and Hierarchical Diffusion Model (DMR-HDM), a framework that systematically generates synchronized dual-modal traffic data. In this framework, we establish a dual-modal representation where Network Flow Features and Network Traffic Images are extracted from each network flow with strict one-to-one correspondence. The hierarchical diffusion architecture decomposes generation into two specialized layers: Network Flow Features DDPM (NFF-DDPM) generates network flow features conditioned on traffic categories to establish global semantic constraints, while Improved Feature-Conditional DDPM (IFC-DDPM) generates network traffic images conditioned on first-layer features to preserve local structural details. To enhance generation quality, we implement Adaptive Feature Fusion that dynamically adjusts conditioning strength across diffusion timesteps, and Progressive Condition Injection that provides layer-specific conditioning across network hierarchical levels. Extensive experiments on CIC-IDS-2017 and TFC-2016 datasets demonstrate superior performance, with DMR-HDM achieving 0.9982 accuracy on TFC-2016 and 0.9876 on CIC-IDS-2017. Compared with state-of-the-art methods, DMR-HDM improves F1-score by 4.32%–11.15% over recent CSV-based diffusion approaches and up to 3.07% over single-modal image methods, with improvements up to 26.10% for particularly challenging minority classes.
Mobile Edge Computing (MEC) enables resource-constrained mobile users to offload computation-intensive and delay-sensitive tasks to nearby edge servers, but the size and frequency of offloaded tasks may reveal users’ locations and usage patterns, raising serious privacy concerns. Although differential privacy (DP) has been widely adopted to protect sensitive data in MEC, existing privacy-preserving task offloading methods still face several limitations, including single-server assumptions, limited adaptability of traditional Laplace-based perturbation mechanisms, and inefficient heuristic optimization in highly dynamic environments To address these challenges, we propose FH-Offload, a novel Flipped Huber (FH) distribution based online privacy-preserving task Offloading framework for multi-server MEC systems. To the best of our knowledge, this is the first work to introduce the FH distribution into MEC task offloading for adaptive bounded-range privacy perturbation Specifically, FH-Offload first leverages the FH distribution to dynamically perturb offloading ratios within a bounded range, providing fine-grained noise control that strengthens privacy guarantees. We then rigorously prove that FH-Offload satisfies (ϵ,δ)-differential privacy under the FH noise model, providing the first formal privacy guarantee specifically for MEC task offloading scenarios. Furthermore, we design a Soft Actor–Critic (SAC) based privacy-preserving task offloading algorithm to adaptively learn both offloading and noise perturbation strategies, enabling efficient offloading, fast convergence, and strong privacy protection in dynamic multi-server environments. Extensive experiments on a real-world MEC dataset show that FH-Offload consistently provides stronger privacy protection with comparable or lower computation cost, demonstrating its effectiveness for multi-server MEC deployments.
Human attention is a dynamic and limited resource, essential for navigating complex tasks and environments. In the digital age, this resource is increasingly strained by frequent task switching, driven by constant notifications and fragmented workflows. While the cognitive costs of task switching are well-documented, its implications for digital privacy remain underexplored. This study investigates a novel privacy threat: the inference of task switching from network traffic by network service providers. We demonstrate that behavioral signatures, such as task durations and switching patterns, can be extracted from temporal metadata, potentially revealing sensitive information about users’ cognitive states and habits. We introduce a machine learning-based approach to infer these patterns and empirically validate its effectiveness on network traffic traces. To counter this threat, we propose a mitigation strategy that obfuscates task-switching signals without substantially impacting network performance. Our findings bridge cognitive science and network security, highlighting a new dimension of privacy risk in modern digital infrastructures and offering practical pathways toward more privacy-preserving communication systems.
The proliferation of Internet of Things (IoT) applications imposes stringent end-to-end Quality of Service (QoS) requirements on inter-domain communication, rendering efficient and stable routing across multiple Autonomous Systems (ASes) a critical and unresolved challenge. Conventional inter-domain routing protocols primarily prioritize reachability between adjacent ASes, neglecting global optimization of composite QoS metrics such as delay and packet loss. Furthermore, existing solutions frequently necessitate the explicit exchange of sensitive routing information, raising significant privacy concerns in multi-AS environments. To address these limitations, this paper presents Q-FRLR, a QoS-aware Federated Reinforcement Learning Routing algorithm. We construct a unified comprehensive delay metric that maps packet loss rates to expected retransmission delays, thereby transforming the multi-objective QoS optimization problem into a tractable single-objective optimization model. Privacy-preserving cross-AS collaboration is enabled via secure exchange of encrypted 5-tuple routing intent information without disclosing detailed routing data. We then develop an online federated Deep Q-Network (DQN) framework in which QoS-aware routing policies are computed collaboratively and deployed by Software-Defined Wide Area Network (SDN-WAN) controllers. We conduct extensive experiments on three Multi-AS Mininet WAN topologies comprising 22, 70, and 105 routers, with iperf3 data-plane validation. Q-FRLR achieves the lowest mean comprehensive delay and online inference delay on all three topologies, reducing mean comprehensive delay by 9.01% relative to the best classical baseline and retaining a 2.7% advantage over PPO-R. It also attains the lowest packet loss of 1.9% and path-switch count of 2.1 among all six baselines.
Federated Learning (FL) represents a distributed training paradigm designed to enable collaborative model training without sharing raw data. However, sophisticated attackers can infer or reconstruct sensitive client data from shared model updates, thereby weakening privacy protection for clients. Moreover, FL also struggles with the data heterogeneity (non-IID distribution) among participants and with high communication overhead due to the transmission of large model weights over multiple rounds. This article presents an integrated framework of three novel algorithms to address the aforementioned issues in FL: (i) FedSketch leverages Differential Privacy (DP) with probabilistic count sketches to compress high-dimensional model updates. This compression approach reduces communication overhead while enhancing privacy. (ii) The CKKSFED algorithm, which utilizes Fully Homomorphic Encryption (FHE) to perform client clustering, mitigating the negative effects of non-IID data distribution in a privacy-preserving manner. (iii) The MetricBasedSelection algorithm selects clients in each round based on metrics to improve convergence and reduce communication. We evaluate the effectiveness of our solutions in an emulation environment using image classification and real-world time series datasets. FedSketch significantly reduced communication overhead, achieving up to a 73-fold reduction in model size. Clustering clients with CKKSFED achieved convergence in approximately half the number of rounds compared to non-clustered methods in non-IID scenarios. Finally, FedSketch, with the metric-based selection algorithm, selected fewer clients for training in each round, thereby contributing to the reduction of communication costs. Jointly, these algorithms reduced communication overhead while ensuring strong privacy guarantees.
Quantum information technology has progressed rapidly, advancing from theoretical foundations to experimental demonstrations and practical applications. The growing potential of this field has drawn significant attention from both academia and industry. A major step toward realizing large-scale quantum communication lies in the interconnection of quantum nodes through quantum networks, which enable reliable transmission of quantum information over long distances. However, routing in such networks is governed by the unique principles of quantum mechanics, superposition, entanglement, and the no-cloning theorem, which distinguish it from classical communication systems. This paper presents an overview of routing protocols for quantum communication networks, emphasizing how different approaches manage challenges such as qubit fragility, fidelity degradation, and probabilistic entanglement generation. Existing protocols are categorized into major types, including static routing, entanglement-based routing, qubit-specific routing, and fault-tolerant routing strategies. The study analyzes how key performance metrics, such as link fidelity, purification efficiency, qubit cost, path selection, and entanglement generation rate, influence routing performance and scalability. Furthermore, the paper outlines the architectural principles, operational models, and development stages of quantum networks while highlighting their differences from classical systems. Finally, it identifies open research challenges, including network architecture design and entanglement routing, that must be addressed to realize future wide-area quantum networks.
The rapid growth of distributed machine learning, particularly Split Learning, has intensified the challenge of communication overhead, e.g., in networking environments with limited bandwidth and fluctuating traffic conditions. Existing approaches typically apply a fixed compression level throughout training, failing to account for network traffic variability and the trade-off between communication efficiency and model accuracy.In this paper, we propose a dynamic compression technique for Split Learning that adapts in real-time to network traffic conditions by combining a network-aware drop rate with quantization. The drop rate determines which intermediate features are transmitted based on current bandwidth and latency. On the other hand, quantization reduces the precision of the transmitted data, jointly minimizing communication without significantly impacting model performance.Our approach also provides an implicit regularization effect, improving generalization by preventing the model from memorizing irrelevant patterns. Extensive experiments on a real-world testbed with Non-Independent and Identically Distributed data across heterogeneous devices demonstrate compatibility with multiple Split Learning architectures. They also show that key network performance indicators (Round Trip Time, Transmission Control Protocol retransmissions, and bandwidth utilization) remain stable under heavy congestion.Overall, our method can reduce data transfer by up to 82% during training and inference, while controlling accuracy degradation, making it highly suitable for communication-constrained environments such as 5G or 6G networks.
Metaverse-enabled multi-access edge computing (MEC) integrates the metaverse with MEC by deploying user task requests for metaverse resources and scenes at the network edge, thereby enhancing users' immersive experience. When accessing the metaverse via virtual reality (VR) devices, users require high real-time performance and diverse computing resources. Additionally, wireless charging is adopted in metaverse-enabled MEC to improve device mobility and avoid experience interruptions caused by battery depletion. Meanwhile, distributed computing, which leverages the spatial distribution of computing, storage, and communication resources across multiple edge nodes, addresses the demands of large-scale interactive scenes and heterogeneous infrastructures in the metaverse, boosting system scalability and reducing processing latency through task allocation and resource sharing. However, inefficient task offloading strategies and suboptimal charging decisions severely degrade network efficiency and user experience such as resource waste, user discomfort and dizziness, highlighting key challenges in task offloading and wireless charging for metaverse-enabled MEC. To address these issues, we formulate a human-centric optimization problem for VR task offloading and wireless charging in wireless-powered heterogeneous metaverse-enabled MEC networks, prioritizing the enhancement of user physical comfort and access point (AP) charging efficiency. To tackle this problem, we propose a framework of the metaverse-enabled MEC Deep Reinforcement Learning (DRL)-based Hybrid Proximal Policy Optimization (MetaMEC-HPPO) with K-Means algorithm preconditioning the network. Furthermore, the experimental results show that the proposed algorithm can achieve superior convergence performance and, meanwhile, improve users' physical comfort and charging profit of AP devices efficiently compared to other baseline approaches. In particular, the average reward of our proposed MetaMEC-HPPO is 11.4% higher than that of Asynchronous Advantage Actor-Critic (A3C) method.
Open Radio Access Networks redefine traditional RAN by introducing functional disaggregation and open interfaces that foster interoperability and innovation, reduce vendor lock-in, and enable flexible 5G and beyond deployments. However, this architectural openness also enlarges the attack surface, creating new vectors of compromise that existing 3GPP protection models do not fully address. Although O-RAN Alliance Working Group 11 (WG11) prescribes standardized security protocols (such as TLS, IPsec, MACsec, and OAuth), their practical implications across heterogeneous interfaces and their underlying cryptographic efficiency remain poorly understood. This paper employs a systematic study methodology that thoroughly evaluates and analyzes the security challenges and research findings related to O-RAN, synthesizing the state of knowledge across architectural layers, protocols, and cryptographic primitives. Our comprehensive and homogeneous evaluation exposes research gaps, biases, and underexplored aspects in O-RAN security, particularly with regard to cryptographic performance and hardware acceleration feasibility. Finally, we present a detailed discussion of open issues and propose targeted future directions to strengthen O-RAN’s security posture and cryptographic resilience in evolving deployments.
6G wireless networks are envisioned to deliver pervasive, intelligent, and resilient connectivity across terrestrial, aerial, and space domains, supporting ultra-high data rates, sub-millisecond latency, massive device density, and tightly integrated communication, computation, and sensing. Realising this vision requires adaptive network intelligence capable of operating under uncertainty in highly dynamic environments, motivating the adoption of computational intelligence (CI), which brings together machine learning and optimisation for continual adaptation, autonomous decision-making, and robust control. Although a growing body of survey literature examines AI in 6G, most existing works focus on isolated technologies, application domains, or specific model classes, providing limited guidance on which CI methods are structurally suited to particular 6G tasks. This distinction matters because deep reinforcement learning, evolutionary algorithms, federated learning, graph neural networks, and fuzzy systems exhibit different inductive biases whose effectiveness depends on the underlying problem structure. This survey addresses this gap through a method-centric synthesis of CI across nine representative 6G enabling technologies: THz communications, reconfigurable intelligent surfaces, integrated sensing and communication, AI-native communications, extreme edge computing, zero-energy communications, space-air-ground integrated networks, semantic communications, and digital twins. For each technology, canonical tasks are identified, representative CI families are mapped to those tasks, and methodological trade-offs are analysed. The survey introduces the method-structure alignment principle as a unifying lens for CI method selection, identifies recurring conditions under which methods succeed or fail, and presents a consolidated research agenda for robust, scalable, cross-domain, and responsible CI in 6G.
Accurate traffic forecasting is crucial for intelligent transportation systems but remains challenging due to four fundamental problems: (1) multi-scale spatial dependencies where traffic exhibits hierarchical correlations from local intersections to city-wide flows, (2) non-stationary temporal patterns influenced by external factors and temporal context, (3) distribution shift and robustness issues caused by sensor failures, topological changes, and traffic regime shifts, and (4) adaptive multi-modal fusion requiring context-aware integration of spatial, temporal, and external features. To address these challenges, we propose MAST-Net, a Manifold-constrained Adaptive Spatio-Temporal Network that integrates spatial, temporal, and external features through three core innovations: (1) a manifold-preserving learnable augmentation module that adapts regularisation intensity to traffic regimes, providing provable bounds on distribution shift under explicit smoothness assumptions; (2) a parallel multi-scale graph convolution that achieves computational efficiency while maintaining representational power; and (3) a temporal-consistency-constrained meta-knowledge integrator with Lipschitz guarantees ensuring smooth prediction evolution. Under the assumptions of manifold smoothness and bounded data distribution, our theoretical analysis establishes generalisation bounds and robustness guarantees under real-world perturbations. Extensive experiments on four benchmark datasets demonstrate that MAST-Net consistently outperforms state-of-the-art methods across all prediction horizons, with ablation studies confirming the contribution of each architectural innovation. This work provides a robust and adaptable solution for real-world traffic forecasting, with theoretical results that hold under specified conditions empirically validated on the datasets considered. The implementation code is publicly available at https://github.com/afofanah/MAST-Net to ensure reproducibility.
This paper introduces DB2, a risk-aware behavioural identity framework that derives device identity from CPU-RTC timing deviation and Performance Monitoring Unit (PMU) microarchitectural events, without relying on GPUs, radios, sensors, or dedicated hardware. The method captures oscillator-coupled timing variation and execution behaviour through a structured signal-processing pipeline, producing device-specific behavioural signatures that remain distinguishable across reboots, temperature variation, and core transitions. DB2 structures identity assurance into three layers: closed-set identification, calibrated open-set rejection, and stability-aware risk scoring. Evaluation under a strict three-way split with reboot separation for training, calibration, and unseen testing yields a macro-F1 of 0.957 on unseen reboots. The open-set layer rejects previously unseen devices with a mean true-positive rate of 0.990 at a calibrated event-level false-reject rate of approximately 0.08 under strict leave-one-device-out validation, with operating-point selection performed exclusively on the calibration split. A Dynamic-Aware Identification and Risk (DAIR) mechanism decomposes behavioural stability across temperature, reboot, and core factors to provide interpretable posture monitoring for enrolled devices. Under identity-claim manipulation via spoofing, Sybil, and relabelling scenarios involving cloning, targeted identities exhibit reduced identification consistency and elevated risk, while non-targeted devices remain stable under identical calibration settings. These results show that behavioural fingerprints can be derived from standard CPU, RTC, and PMU-accessible resources on edge devices, enabling device-identity and behavioural-assurance monitoring in IoT and edge environments without specialised hardware.
Covert channels exploiting application-layer protocols represent a persistent threat in enterprise environments, particularly when targeting endpoints that remain outside the scope of endpoint detection and response solutions, such as embedded Voice over Internet Protocol (VoIP) phones. Existing SIP-based covert channels either embed low-bandwidth steganographic data within call-lifecycle signaling messages (Mazurczyk and Szczypiorski, 2008; Mehić et al., 2014; Tsiatsikas et al., 2015), or achieve higher capacity by requiring active Real-time Transport Protocol (RTP) media streams (Schmidt et al., 2018; Saenger et al., 2020). However, to our knowledge, no prior work implements continuous Layer 2/Layer 3 tunneling over standalone SIP signaling outside of call contexts, nor validates such a channel on real embedded hardware against multiple network intrusion detection systems. We present a covert channel that encapsulates arbitrary Ethernet frames and IP packets within SIP OPTIONS messages, which are lightweight requests typically used for NAT keep-alive and capability probing, without establishing any voice call. A portable proof-of-concept with TUN/TAP virtual interfaces and ChaCha20 symmetric encryption has been implemented and deployed on two ARM-based IP phones from different vendors, showing that even resource-constrained embedded devices can serve as covert network egress nodes. Empirical evasion testing against Suricata, Snort 3, Zeek, and a FortiGate firewall with full Unified Threat Management (UTM) services reveals that the channel achieves a secure operating bandwidth of 128 kbit/s without triggering alerts on any of the tested platforms. Analysis of the tested rulesets and parsers reveals the absence of entropy-based or header-length anomaly rules for Session Initiation Protocol (SIP), accounting for the detection gap. To address this, we implement a lightweight statistical detection mechanism that analyzes non-standard SIP header entropy, showing that the covert channel can be identified through targeted analysis of header content. The gap is not inherent to SIP: current rulesets inspect syntax but not header content, length, or entropy. Adding any of these features is sufficient to detect the channel, as shown by the Zeek script and Suricata 8 rule provided as supplementary material. A single entropy threshold can itself be evaded through low-entropy padding. We therefore show that multi-feature and anomaly-based detectors recover identification of the channel where the scalar test fails. We also contribute a synthetic benign SIP corpus, generated with SIPp to emulate multiple devices and signaling scenarios, released as supplementary material for evaluating false positives beyond the single legacy dataset previously available.
As Internet of Things (IoT) technology rapidly evolves, Wireless Sensor Networks (WSNs) have become indispensable in applications such as environmental monitoring and precision agriculture. However, their deployment in unattended environments and dynamic communication conditions make them vulnerable to sophisticated attacks, including selective forwarding and masquerading behaviors. Current trust management methods have not well considered the evaluation of dynamic and complex trust evidence under resource constraint. To tackle these issues, in this paper we propose an adaptive trust management mechanism for malicious node detection. The proposed mechanism integrates communication, behavioral, and energy evidence, which captures behavioral patterns via temporal decay and spatial deviation analysis. To address time-varying evidence reliability, a dynamic weight adjustment mechanism based on fuzzy clustering is further developed, which monitors link quality metrics and categorizes trust evidence by environmental sensitivity. This enables the dynamic mechanism to automatically reconfigure evidence weights according to environmental adaptability under dynamic conditions. Extensive experiments demonstrate that our proposed mechanism achieves accuracy above 98% and false positive rate (FPR) below 1% in dynamic environments, outperforming baseline methods by up to 16% in accuracy and 19% in FPR reduction.
Due to the diversity of quality of service (QoS) requirements and the complexity of link conditions in space–air–ground–sea integrated network (SAGSIN), traditional congestion control (CC) methods often respond slowly and lack the capability to adaptively balance multiple performance metrics according to multi-type QoS requirements, making them insufficient for meeting heterogeneous application demands. To address these limitations, we propose QHICC, a QoS-aware hierarchical intelligent CC framework composed of an experience layer, a learning layer, and a reliability layer. These layers collaborate to enable rapid policy reuse, adaptive optimization, and stable online operation under the complex and heterogeneous link conditions across space, air, ground, and sea segments. The experience layer infers QoS requirement profiles from observed network metrics, enabling the rapid reuse of the corresponding pre-trained policy from the repository. The learning layer employs multi-objective deep reinforcement learning (MO-DRL) to dynamically balance various performance metrics according to QoS requirements, while incorporating training strategies to improve learning efficiency and convergence speed. The reliability layer applies an action constraint mechanism to mitigate the impact of severe observation noise and prevent overly aggressive policy adjustments, thereby enhancing decision stability in fluctuating environments. It further utilizes model migration to ensure decision continuity in the event of frequent handovers across heterogeneous segments and intermittent connectivity across cross-domain links. Our extensive experimental evaluations show that QHICC consistently outperforms existing CC methods across different QoS requirements in terms of throughput, delay, packet loss, and load balancing, and demonstrates strong robustness and adaptability in highly dynamic SAGSIN environments.
