Device-free human sensing using WiFi Channel State Information (CSI) has shown strong potential for people counting in intelligent environments, due to its ability to operate reliably without relying on lighting conditions or direct visibility. However, single-modality approaches based on either WiFi or vision often suffer from limited accuracy and robustness, especially under challenging conditions such as occlusion, low illumination, and environmental variations. Multimodal sensing that combines WiFi and vision offers a promising solution by leveraging complementary information from both modalities. Nevertheless, several critical challenges remain, including the lack of low-cost synchronized data acquisition platforms, the scarcity of public multimodal datasets under adverse conditions, and the absence of systematic studies on fusion strategies and cross-domain generalization.In this work, we propose PRoFENCH, a comprehensive WiFi–Vision multimodal sensing framework for people counting. We develop a low-cost synchronized sensing platform and introduce a public multimodal dataset collected under challenging conditions such as occlusion and low-light environments. We then propose WiVi32-Fusion, a lightweight and neural-network-based fusion model, and conduct a systematic evaluation of multiple multimodal fusion strategies under a unified framework. In addition, we analyze explainability, robustness, and cross-domain generalization. Experimental results show that the proposed framework achieves a low mean absolute error (MAE) of 0.05 and maintains stable performance across different domains. These results demonstrate the effectiveness of the proposed framework under the evaluated sensing conditions.
Unmanned aerial vehicle (UAV)-assisted mobile edge computing (MEC) can provide delay-sensitive computation services in areas with limited terrestrial infrastructure. In an open service market, however, UAVs are usually owned by self-interested sellers, and a single UAV may not satisfy the deadline, reliability, and energy requirements of a task. This paper studies a multi-task group-selling auction for collaborative offloading in multi-UAV edge networks. Multiple task requesters (TRs) submit heterogeneous task contracts, while ground-side brokers recruit UAV sellers and package them into task-specific offloading services. We formulate the broker-side procurement and allocation problem under residual-energy, deadline, reliability, and budget constraints. To support short trading epochs, we design a three-stage multi-task group-selling auction (MT-GSA), where brokers construct feasible UAV coalitions, compute critical-ask payments for selected UAVs, and compete for TR contracts through a budget-feasible reverse auction. The mechanism preserves service feasibility, provides individual rationality for selected UAVs and winning brokers, and satisfies task-level ask-monotonicity under fixed pre-task residual states and verifiable service parameters. Cross-task state coupling limits this result to task-level threshold payments and does not ensure global strategy-proofness. Simulations show that MT-GSA improves task completion ratio, requester utility, broker utility, and UAV-side participation incentives over representative baselines.
Vehicular Networks (VNs) are becoming a core component of 5G/6G intelligent transportation systems, supporting Location-Based Services (LBSs) such as navigation, traffic management, emergency response, smart parking, charging assistance, and remote driving. These services rely on continuous exchanges of location-related information among vehicles, Road-Side Units (RSUs), edge servers, cloud platforms, and LBS providers. However, location privacy in 5G/6G VNs is no longer limited to the protection of a single reported coordinate. Repeated LBS queries, RSU handovers, packet timing, physical-layer sensing traces, application-layer service semantics, and cross-platform data fusion may jointly enable trajectory reconstruction, identity-location association, semantic-place inference, and long-term mobility profiling.This survey provides a threat-driven and deployment-aware review of location privacy threats and Location Privacy-Preserving Mechanisms (LPPMs) in 5G/6G VNs. First, we consolidate location-privacy adversaries into a unified threat model and map representative tracking threats to violated privacy requirements and LPPM failure conditions. Second, we classify existing LPPMs according to their primary protection locus and dominant stage in the location-data lifecycle, including user-side, server-side, and user-server-interface mechanisms. We evaluate these mechanisms not only by their theoretical privacy properties, but also by their feasibility, latency, overhead, deployment realism, and privacy–utility trade-offs under high mobility, repeated LBS queries, dense RSU/edge deployment, and cross-layer metadata leakage. Third, we identify emerging challenges introduced by AI-driven trajectory inference, multi-source data fusion, and advanced wireless technologies such as sub-6 GHz, mmWave, THz, visible light communication, quantum communication, and reconfigurable intelligent surfaces. Based on this analysis, we outline future research directions for designing lightweight, adaptive, technology-aware, and inference-resilient LPPMs that can better balance privacy protection, service utility, and deployment feasibility in future vehicular networks.
Low Earth orbit (LEO) satellite networks rely on dynamic wireless inter-satellite links, where rapid topology variation, link-quality fluctuations, queue accumulation, and multi-hop forwarding coupling make QoS-aware routing highly challenging. In such environments, routing decisions must adapt to time-varying connectivity and traffic conditions while jointly optimizing multiple conflicting objectives, including end-to-end delay, packet loss rate, and throughput. To address this problem, this paper formulates QoS-aware packet forwarding in LEO satellite networks as a queue-aware multi-objective cooperative routing problem and develops an Expected-Advantage Multi-Agent Reinforcement Learning (EA-MARL) framework under the centralized training and decentralized execution (CTDE) paradigm. In EA-MARL, a temperature-annealed softmax-expectation mechanism is introduced to replace the conventional hard-max advantage operator in value-decomposition MARL, thereby alleviating max-induced overestimation and improving learning stability in highly dynamic routing environments. By jointly incorporating dynamic link states, queue conditions, and end-to-end QoS objectives, the proposed method enables robust cooperative forwarding over time-varying inter-satellite networks. Simulation results on Iridium and Starlink constellations show that EA-MARL achieves faster convergence and better overall QoS performance than representative baselines in terms of end-to-end delay, packet loss rate, and throughput.
Reliable three-dimensional (3D) radio maps are a key enabler for low-altitude Unmanned Aerial Vehicle (UAV) operations in dense urban environments, since path planning, link adaptation, and coverage assessment all benefit from accurate spatial signal-quality estimates. Most existing radio-map studies, however, focus on two-dimensional estimation, rely on simulation-only benchmarks, or use Reference Signal Received Power (RSRP)-only metrics, which collectively fail to reflect actual deployment conditions. Real measurements, in contrast, are expensive to collect, sparse, and unevenly distributed across altitude layers. To bridge this gap, this paper develops a real-world 3D urban radio-map construction pipeline together with a multi-fidelity Gaussian process regression (mfGPR) estimator that fuses dense ray-tracing simulation (low fidelity) with sparse UAV measurements (high fidelity), and adopts the Third Generation Partnership Project (3GPP) cell-selection criterion (S-criterion) as the prediction target so that both received power and received quality are jointly captured. To use the limited high-fidelity budget more effectively, a two-stage sampling strategy is introduced, combining anisotropic farthest-point coverage with metric-driven densification.Experiments on a measurement campaign in an urban district of Macao show that, when averaged over sampling rates of 10%–100%, mfGPR reduces root-mean-square error (RMSE) by about 2.5decibels (dB) and mean absolute error (MAE) by about 1.9 dB and raises R2 by roughly 0.42 on the high-fidelity test set relative to the four single-fidelity baselines (inverse distance weighting (IDW), k-nearest neighbors (KNN), vanilla Gaussian process regression (GPR), and Kriging); the advantage is largest at the high-budget end, while at sparse rates the strongest classical baseline (Kriging or IDW under the proposed sampler) is within a few tenths of a dB. In the sparse-sampling regime that characterizes UAV campaigns, mfGPR also outperforms modern learning-based baselines (a deep neural network, a Conditional Neural Process, and Deep Kernel Learning), which tend to overfit the limited high-fidelity data, although these models become competitive at the full measurement budget. The trend holds across altitude layers and remains stable under simulation-bias perturbations of up to 10 dB. Ablations further show that the simulation prior and the sampling strategy contribute complementary, non-redundant gains. The proposed framework therefore offers a practical basis for 3D radio-map reconstruction in support of UAV-assisted communications.
Wireless sensor networks (WSNs) can use wireless energy transfer to mitigate node-level energy depletion and improve the balance of residual energy across the network. Many energy transfer coordination methods use a fixed concurrent parameter K, which specifies the number of donor nodes that may serve one receiver at the same time. A fixed value of K is simple to implement, but it may underuse available donors when the network is energy-imbalanced or create unnecessary transfer loss when concurrency is excessive.This paper proposes a Dynamic-K adaptation framework for adjusting the concurrent parameter according to the observed network state. The framework is designed as an externally attachable enhancement layer under a compatible concurrency-control interface; therefore, it can be evaluated with existing scheduling methods without changing their internal scheduling logic. A multi-objective reward function is used to capture energy-balance improvement, effective energy delivery, and transfer loss. Candidate K values are then assessed through a short-horizon prospective simulation mechanism, and the value with the highest simulated reward is selected for the next coordination interval.The framework is evaluated in simulation on a 25-node WSN with heterogeneous solar-energy harvesting capability and mobility. Five representative scheduling methods are considered, including average-based, heuristic, Lyapunov-based, prediction-based, and power-control schedulers, under both static-K and Dynamic-K configurations.Within the evaluated simulation settings, Dynamic-K substantially improves energy-distribution uniformity while keeping average transfer efficiency broadly stable. Across the five representative methods, the average energy standard deviation is reduced by about 64.4%, from 4219.3 to 1503.6 J, while the mean transfer efficiency changes from 52.9% to 52.5% (−0.4 percentage points). These results indicate that adaptive concurrency control can serve as a complementary enhancement layer for compatible energy transfer coordination methods in WSNs, particularly when energy uniformity is a primary objective.
This study develops a centralized cloud radio access network (C-RAN) architecture for vehicular communications, investigating joint resource allocation in a heterogeneous vehicular network consisting of UAV-mounted base stations (UAV-BSs), ground Active Antenna Units (AAUs), and Roadside Units (RSUs), where caches, wireless channel bands, bitrates and transmission time slots are jointly optimized to support multimedia transmission in highway environments. The proposed framework addresses challenges such as network load levels, traffic jams, video playback latency, and dynamic UAV-BS deployment through joint optimization of caches, wireless channel bands, bitrates and transmission time slots (resource). To enhance user quality of experience, we analyze cooperative transmission scheduling and the mobility characteristics of UAV-BSs, and identify key factors affecting multimedia playback performance. Based on these insights, a cluster-based scheduling algorithm is proposed for resource allocation, together with a dynamic UAV-BS deployment strategy that enables timely and location-aware service adaptation. Simulation results under network load levels scenarios characterized by high vehicle density demonstrate that the proposed approach maintains stable performance and improves system throughput according to the vehicular mobility and video playback progress. Furthermore, experimental results demonstrate that the proposed algorithms significantly reduce progressive video playback lag compared with existing methods. These findings provide practical guidelines for transmission scheduling and UAV-BS deployment in intelligent transportation systems, contributing to sustained multimedia service quality even under congested network conditions.
Intelligent connected vehicles (ICVs) integrate dozens of electronic control units, advanced driver-assistance systems, and vehicle-to-everything (V2X) communication, creating an expanded attack surface that directly threatens functional safety. Automotive cybersecurity incidents numbered 422 CVEs in 2024 alone, with 60% affecting thousands to millions of assets simultaneously. Despite this growing risk, existing intrusion detection approaches share three structural gaps: reliance on single-source monitoring that misses coordinated multi-vector attacks; absence of formal coupling between cybersecurity threat severity and functional safety constraints; and dependence on reactive, externally commanded responses rather than endogenous self-organizing mechanisms.An Endogenous Threat Recognition and Hazard Mitigation (ETRHM) framework is proposed to address these gaps. A Polygene Threat Recognition Network (PTR-Net) fuses CAN bus traffic, V2X messages, and application-layer telemetry through hierarchical cross-source attention, enabling detection of coordinated attacks invisible to single-source methods. An Endogenous Adversarial Mechanism (EAM) couples detected threat severity to functional safety constraints through a Lyapunov-stable feedback loop, adapting defense strategies autonomously in real time. A Hazard Mitigation Maneuver (HMM) then executes a four-phase resilience protocol—prevention, resistance, recovery, and adaptation—to drive the vehicle to a Minimal Risk Condition (MRC). Formal proofs establish asymptotic stability of MRC and bound the recovery time.Experiments conducted on an NXP S32G399A ASIL-D automotive controller using real vehicle network traffic demonstrate an AUC of 0.953 and a 26.5% reduction in time-to-MRC (4.9s to 3.6s) relative to the strongest baseline, with an end-to-end detection-to-response latency of 8.1ms—within the 10ms automotive control cycle constraint.
Intrusion detection systems play major role in security of Internet of Things (IoT) networks against various types of cyber threats. However, traditional Machine Learning (ML) and Deep Learning (DL) models still struggle with challenges such as critical class imbalance, reliance on manual tuning of hyperparameters, and the expensive cost of obtaining labeled data. In this study, we addressed these limitations using the TON_IoT dataset. To address data imbalance, the proximity weighted random affine shadow sampling technique is utilized. Further, Bayesian optimization is applied on LeNet, resulting in LeBayesNet, which provides the optimal configuration for high-accuracy threat detection. Additionally, EntroLeNet integrates entropy-based uncertainty into the learning process for improved robustness. Next, MargiLeNet leverages marginal-based active learning, annotating the most uncertain samples. Experimental results obtained show that LeBayesNet, MargiLeNet, and EntroLeNet improve performance over existing ML and DL models by 6.90%, 5.80%, and 4.27% in accuracy and 6.19%, 6.43%, 8.05%, and 7.52% in receiver operating characteristic-area under the curve, respectively. The LeBayesNet, MargiLeNet, and EntroLeNet models significantly reduce Hamming loss by 72.84%, 60.49%, and 43.21%, respectively. For robustness and generalizability assurance, 10-fold cross-validation is used during evaluation and validated the statistical significance of performance improvements using the Mann–Whitney U test. Furthermore, the models’ explainability and interpretability is enhanced through Shapley additive explanations and local interpretable model-agnostic explanations, providing insights into feature importance and decision transparency.
Multipath Transmission Control Protocol (MPTCP) is relevant to Industrial Internet of Things (IIoT) scenarios with heterogeneous access links, but dynamic variations in bandwidth, delay, and packet loss make multipath congestion control challenging. To improve adaptability while preserving control stability, this paper proposes Attention-enhanced Multi-Agent Congestion Control (AT-MACC), a hybrid congestion control method that combines the BALIA heuristic with a multi-agent proximal policy optimization (MAPPO) framework. BALIA provides a stable baseline congestion window, while each subflow agent generates a bounded correction factor based on local observations. An actor network with LSTM and attention is used to capture temporal link dynamics, and a local–global reward is designed to balance transmission efficiency and internal fairness among subflows. Simulation results under Mininet-based heterogeneous link scenarios show that AT-MACC achieves higher measured throughput and lower RTT than LIA, OLIA, and BALIA under the evaluated settings. These results provide controlled simulation evidence for the feasibility of learning-assisted congestion-window correction, while more comprehensive validation in real wireless IIoT deployments remains future work.
As a vital component of 6G networks, low Earth orbit (LEO) satellite communications face frequent beam handover caused by high-speed movement while achieving wide-area coverage. To address this, a velocity-adaptive beam handover algorithm is proposed. For mobile terminals (MTs) in different scenarios, the algorithm dynamically adapts to their velocities by selecting either an improved double deep Q-Network (DDQN) or the technique for order preference by similarity to an ideal solution (TOPSIS) to achieve global optimal decision-making. This approach ensures real-time performance while comprehensively evaluating beam quality. Furthermore, it adaptively adjusts the hysteresis parameter (HP) to optimize handover decisions. In addition, the algorithm constructs utility functions and weight allocation mechanisms to meet the diverse Quality of Service (QoS) requirements of different services. Simulation results demonstrate that the proposed algorithm achieves superior performance. Compared with existing algorithms, this algorithm effectively suppresses ping-pong handovers, thereby reducing the number of handovers and handover failures while improving user throughput.
Underwater networks are becoming a vital component of 6G networks. To integrate them with both terrestrial and non-terrestrial networks, it is necessary to address inherent challenges in stability, scalability, and efficiency. We propose a spectrum- and energy-aware software-defined underwater network (SEA-SDUN) framework to simultaneously overcome spectrum scarcity (specifically in acoustic networks), path instability, and energy constraints. To balance the load, a dual-layer SDN architecture is employed, where local controllers (LCs) manage local topologies, including updates for spectrum, available paths, and residual energy, providing intelligence and flexibility to the network, while a single surface-based main controller (MC) serves as a backup, maintains global topology, and manages the battery status of the nodes by serving as a charging station. SEA-SDUN addresses these issues by jointly optimizing link delay and energy consumption while also providing optimized trajectories for AUVs. Consequently, SEA-SDUN establishes the most optimal paths available between source and destination pairs, significantly improving the packet delivery ratio while reducing delay, overhead, and energy consumption compared to two reference schemes.
