Low earth orbit (LEO) satellite networks, as integrated service systems, are typically divided into domains based on application functions, such as observation and communication. However, independent domain resources and the high dynamics of satellites make cross-domain resource interactions difficult to capture, posing challenges to large-scale satellite network resource coordination and degrading the quality of service (QoS) for task flows. To address this, a multi-level hypergraph (MLH) is introduced to represent domain resources across temporal, resource type and spatial dimensions. MLH consolidates similar resource features via hyperedges, reducing redundant connections. The cross-domain resource coordination problem is then modeled as a mixed-integer linear programming (MILP) problem to maximize the sum of the minimum priorities of scheduled tasks. Furthermore, by leveraging MLHs topological nesting, a multi-dimensional resource dual-level scheduling algorithm (MRDSA) is proposed, decomposing the problem into two subproblems solved using the hyperpath scheduling algorithm (HSA) and internal path scheduling algorithm (IPSA). Simulations demonstrate that the proposed method enhances QoS, reduces computational complexity and improves resource utilization ratio in LEO satellite networks.

In recent years, uncrewed aerial vehicles (UAVs) have become increasingly prevalent for collecting environmental data from various wireless sensors. However, existing research on employing UAVs to collect data from wireless sensors has often ignored the heterogeneous requirements of sensors. In this paper, we investigate joint deployment, user association, and power allocation for data collection in the UAV-assisted wireless sensor network to accommodate the heterogeneous requirements of sensors, where a novel satisfaction function is designed for three types of sensors, including sensors with delay requirements, sensors with energy consumption requirements, and sensors with both delay and energy consumption requirements. Leveraging the satisfaction function, we formulate the optimization problem aimed at jointly optimizing the positions of UAVs, the association between sensors and UAVs, and the power allocation of sensors to maximize overall satisfaction of sensors. In order to effectively address the considered problem, we decompose it into two subproblems, i.e., joint UAV deployment and user association subproblem, and transmission power allocation subproblem. An enhanced human evolutionary algorithm is developed to tackle the joint UAV deployment and user association subproblem, and the Lagrange dual method and gradient descent method are employed to solve the transmission power allocation subproblem. The suboptimal solution is achieved by iteratively addressing the two subproblems until convergence of the proposed enhanced Lagrange and gradient descent-based human evolutionary optimization algorithm is attained. Extensive simulations demonstrate the effectiveness of the proposed algorithm in enhancing overall satisfaction of sensors, underscoring its significant advantages in managing heterogeneous network environments.

Satellite communications are widely utilized in the Internet of Remote Things (IoRT) to achieve large-scale coverage and efficient data collection. This paper investigates an efficient data acquisition scheme in which Low Earth Orbit (LEO) satellites collect data from terrestrial low-power sensor networks via gateway stations (GSs). Specifically, ground-based sensors employ short-packet transmissions to relay sensed data to GSs, which subsequently aggregate the received data and upload it to LEO satellites passing over the relevant area. To ensure efficient data collection, our objective is to minimize the energy consumption of terrestrial sensors while achieving the lowest possible data collection delay, thereby extending the sensor lifetime. To address this joint optimization problem, we propose a delay-energy-efficient data aggregation and scheduling algorithm based on proactive network configuration and a novel hypergraph-based link scheduling approach. Finally, extensive numerical simulations are conducted to evaluate the performance of the proposed algorithm. The experimental results demonstrate that, compared to existing benchmark algorithms, the proposed method significantly reduces both data collection delay and sensor energy consumption.

Integrated Energy Systems (IES) are critical infrastructures enabling multi-energy synergy and low-carbon transitions. However, their distributed and uncertain nature poses significant challenges for adaptive and efficient scheduling. To address these issues, this paper proposes a software, edge-intelligent, and cross-layer orchestration algorithm based on Multi-Agent Multi-Objective Proximal Policy Optimization (MAMOPPO). First, a cooperative multi-agent scheduling model is established, where distributed agents operate at the edge and coordinate through shared global information. Then, a multi-value network is employed to decouple and optimize operational costs, carbon emissions, and renewable energy utilization. Finally, a mirror learning strategy is introduced to enhance policy stability under uncertainty and facilitate cross-layer coordination between energy control and communication layers. Simulation results show that the proposed approach reduces system cost by 42.8%, lowers emissions by 44.6%, and maintains high renewable energy utilization, demonstrating its effectiveness in orchestrating intelligent edge-based services in next-generation smart grid scenarios.

Telemedicine significantly reduces patients' medical treatment time and costs. Medical images play an important role in telemedicine services, which contain patients' private information and face the risk of illegal access when transmitted over the network, leading to patient privacy leakage. This paper proposes a dual-adversarial medical image encryption network (DAEN) for protecting patient privacy. We formulate the image encryption problem as an image conversion task and design two paired adversarial networks for image transformation. Furthermore, we analyze the characteristics of an ideal cipher image and construct pseudo-ciphertext to guide network training. The networks generate keys with strong randomness and sensitivity through adversarial training and dynamically encrypt the medical images. Experiments are conducted on the Chest X-ray dataset, and the results show that the DAEN can encrypt medical images into meaningless images, achieving high information entropy, NPCR, and UCAI compared to advanced encryption algorithms, which can effectively secure medical image transmission.

This review delves into the applications and prospects of nonterrestrial networks (NTNs) in the field of information and communication. NTNs utilize aerial or space platforms as critical components of the communication network, including high-altitude unmanned systems, low-altitude unmanned systems, and satellites. Compared to traditional terrestrial cellular networks, NTNs offer advantages, such as wider coverage, flexible deployment, and resistance to ground-based disasters. Therefore, NTNs have broad application prospects in industries, such as transportation, public safety, media entertainment, healthcare, energy, agriculture, and finance. This review focuses on the network architecture and key technologies that support NTNs, including the system’s composition architecture, key technologies, application case analysis, challenges, potential solutions, and future outlook, aiming to provide beneficial reference and guidance for the promotion and application of NTNs. The review also examines the support of international organizations for NTNs’ standardization, as well as related research progress and future challenges.

The integration of vehicular networks with Machine Learning (ML) is driving the advancement and intelligence of future vehicular systems. As a core technology in the IoVs, Vehicular Edge Computing (VEC) leverages the computational and communication resources of both vehicles and edge servers, enabling model training closer to the data source. Federal Learning (FL) has shown great promise in training large-scale ML models without exposing raw data. However, many vehicles are reluctant to participate in FL training due to high resource demands and the inherent mobility challenges of vehicular networks. To address this issue, this paper proposes an FL auction framework that incentivizes vehicle participation by maximizing the utility of the FL platform. Specifically, our approach factors in the basic utility, average reward and tolerance delay of dynamic vehicles to determine their bidding intent. Additionally, an online auction-based client selection algorithm is proposed that ensures individual rationality for vehicles, coupled with a reward function based on model accuracy to further encourage participation in FL training. Simulation results demonstrate the effectiveness of the proposed algorithm, showing a 38.9% improvement in platform utility and a 31.3% reduction in average payments compared to the Online Auction (OA) algorithm.

Non-terrestrial networks (NTNs) are integrated with terrestrial networks to form space-air-ground integrated networks (SAGINs), providing seamless global coverage and supporting the development of the digital economy. However, when it comes to the actual design and deployment of SAGINs, the heterogeneity, self-organization, and flexibility of SAGIN pose challenges for precise modeling and quantitative analysis. In this regard, this paper proposes a multi-dimensional analysis model based on stochastic geometry for SAGIN, which considers the randomness of ground users’ (GUs) distribution and the multi-dimensional coverage characteristics of NTN nodes. The model determines the policies for GUs to access NTNs by adopting the maximum received average signal-to-interference-plus-noise ratio (SINR) association policy (AP) and the balanced satellite load AP. Specifically, we analyze the interference distribution of different links in the terahertz (THz) band and their Laplace transforms, then derive the uplink connectivity expressions of ground-to-space links with/without aerial relays under the two APs. Numerical results validate the accuracy of the theoretical model and explore the impact of APs, SINR thresholds, THz channel propagation coefficients, and aerial relay numbers on SAGIN connectivity, providing theoretical guidance for the deployment of THz SAGINs.

As the most vulnerable part of the traffic scenario, it is vital to ensure the safety of pedestrians. Accurately predicting the future trajectory of pedestrians not only ensures the safety of pedestrians but also improves the efficiency of traffic operations. In light of this, this paper presents a novel target-driven method for pedestrian trajectory prediction. The method uses bidirectional long short-term memory (Bi-LSTM) to predict the approximate position of the pedestrian at different time intervals and uses the short-term location as a short-term target for pedestrians. Meanwhile, our method takes into account the influence of surrounding vehicles on the future trajectory of pedestrians. We predict the speed of vehicles around the pedestrian and apply it to the pedestrian trajectory prediction to make the prediction more accurate. We evaluated our method on two public datasets, PIE and JAAD, and demonstrated its superiority over the benchmark methods. The ablation experiments further reveal that the target estimation and speed estimation modules reduced prediction errors by approximately 11% and 30% respectively.

Pedestrian trajectory prediction in autonomous driving is a critical and complex issue with important implications for road traffic safety. The primary challenges of this task stem from: 1) difficulty in simultaneous learning of individual and group motion behaviors within dense crowd scenarios and 2) limited interpretability of the model. In this work, we implement the social force model to the Conditional Variational Auto Encoder (CVAE) trajectory prediction framework and propose a trajectory prediction method in dense crowd scenarios, called GroupGCN. Specifically, firstly, pedestrians are grouped using the grouping module. Secondly, Graph Convolutional Network (GCN) is utilized to analyze pedestrian movement from both intra-group and inter-group perspectives. Intra-group involves understanding interactions among individuals while considering their preferences. Inter-group entails studying interactions between pedestrian groups at the group level, then the trajectory collision problem between groups is addressed by the collision prediction network. The performance of GroupGCN is evaluated compared with several public benchmarks. Experimental results show that GroupGCN improves performance by 61.8% on the ETH and UCY datasets compared to the prediction methods that consider pedestrian groups, which shows that GroupGCN will have better performance in dense crowd scenarios.

Internet of Medical Things (IoMT) is increasingly gaining attentions in fall detection because of its ability to sense, monitor and analyze, which can provide proper assistance to the elderly with fragile health conditions. As fall events are infrequent, its important to timely detect its occurrence in order to alleviate the harmless. This paper presents a Lightweight Attention Network Fall Detection (LA-FD) framework, which detects falls by analyzing gait acceleration signal (GAS). Then, we can take appropriate measures to mitigate the impact. LA-FD introduces depth-separated convolution to the Lightweight Attention Network module to reduce computational costs and model parameters. Additionally, relative positional offsets are incorporated into each self-attention module to enhance attention mechanisms' expressiveness. The result shows that LA-FD significantly reduces model size by 99.3% compared to the Transformer and 95.1% compared to CNN-LSTM, while only a 5% accuracy drop compared to the Transformer and 2% compared to CNN-LSTM.

Autonomous vehicles may mistakenly predict the future trajectories of neighboring vehicles when the trajectory prediction model is under attack. Recent works utilize adversarial training to mitigate the prediction errors of the trajectory prediction model under attacks. However, adversarial training exhibits high training costs and poor generality for different attack methods. Meanwhile, adversarial training improves the trajectory prediction performance under attacks by learning the adversarial examples, which leads to greater performance degradation in normal (without attacks) cases. In this paper, to ensure the driving safety of autonomous vehicles, we propose a causal robust trajectory prediction method named CausalRobTra, which employs Total Direct Effect (TDE) inference to defend trajectory predictors against adversarial attacks from the perspective of causal inference theory. First, we propose four directional metrics to evaluate the prediction errors of the trajectory prediction model under attacks. Then, we construct the causal graph of trajectory prediction under attacks and analyze the causalities among the nodes. Next, we conduct the counterfactual intervention on the history trajectory by replacing the history trajectory with the counterfactual trajectory to cut off the link between the history trajectory and the adversarial perturbation. Finally, we calculate TDE by subtracting the counterfactual prediction from the factual prediction to eliminate the impact of adversarial perturbation on the final prediction. Compared with no defense case, our method improves the performance by 13.4% under attacks and at the cost of 7.7% performance degradation on clean data. In addition, our method improves the performance by 20.6% on clean data compared with adversarial training and has a similar performance to adversarial training under attacks. Such an improvement can ensure the safety of autonomous vehicles under attacks and avoid many traffic accidents. Our CausalRobTra is a plug-and-play defense method that can be easily applied to any other trajectory prediction model. Extensive experiments demonstrate that our method effectively improves the adversarial robustness of the trajectory prediction model under attacks at the expense of lower performance degradation in normal (without attacks) cases.

Image matching is a fundamental and critical task of multisource remote sensing image (RSI) applications. However, RSIs are susceptible to various noises. Accordingly, how to effectively achieve accurate matching in noise images is a challenging problem. To solve this issue, we propose a robust multisource RSI matching method utilizing attention and feature enhancement against noise interference. In the first stage, we combine deep convolution with the attention mechanism of the transformer to perform dense feature extraction, constructing feature descriptors with higher discriminability and robustness. Subsequently, we employ a coarse-to-fine matching strategy to achieve dense matches. In the second stage, we introduce an outlier removal network based on a binary classification mechanism, which can establish effective and geometrically consistent correspondences between images; through weighting for each correspondence, inliers versus outliers classification are performed, as well as removing outliers from dense matches. Ultimately, we can accomplish more efficient and accurate matches. To validate the performance of the proposed method, we conduct experiments using multisource RSI datasets for comparison with other state-of-the-art methods under different scenarios, including noise-free, additive random noise, and periodic stripe noise. Comparative results indicate that the proposed method has a more well-balanced performance and robustness. The proposed method contributes a valuable reference for solving the difficult problem of noise image matching. The code is available athttps://github.com/liyuan-repo/RMmodel.

Psychophysiological computing can be utilized to analyze heterogeneous physiological signals with psychological behaviors in the Internet of Medical Things (IoMT). Since IoMT devices are generally limited by power, storage, and computing resources, it's very challenging to process the physiological signal securely and efficiently. In this work, we design a novel scheme named Heterogeneous Compression and Encryption Neural Network (HCEN), which aims to protect signal security and reduce the required resources in processing heterogeneous physiological signals. The proposed HCEN is designed as an integrated structure that introduces the adversarial properties of Generative Adversarial Networks (GAN) and the feature extraction functionality of Autoencoder (AE). Moreover, we conduct simulations to validate the performance of HCEN using the MIMIC-III waveform dataset. Electrocardiogram (ECG) and Photoplethysmography (PPG) signals are extracted in the simulation. The results reveal that the proposed HCEN can effectively encrypt floating-point signals. Meanwhile, the compression performance outperforms baseline compression methods.

To support diverse Internet of vehicles (IoV) services with different quality of service (QoS) requirements, network slicing is applied in vehicular networks to establish multiple logically isolated networks on common physical network infrastructure. However, dynamic and efficient radio access network (RAN) slicing adapting to the dynamics of vehicular networks remains challenging. The diverse applications make multi-dimensional resource requirements, which will result in the resource allocation more complicated. In addition, the system needs to frequently adjust the resources of slices, which will cause additional slicing overhead. Thus, to solve the above problems, we propose a resource allocation strategy by using multi-agent reinforcement learning to allocate resources in vehicular networks. Firstly, the cost composition of RAN slicing is analyzed, and the optimization problem is formulated to minimize the long-term system cost. Then, we transform the resource allocation problem into a partially observable Markov decision process. Finally, we propose a multi-agent deep deterministic policy gradient based resource allocation algorithm to solve it. All base stations are treated as independent agents, and they cooperatively allocate spectrum and computing resources. Simulation results show that the proposed strategy reduces the system cost effectively compared to the benchmarks, and the average QoS satisfaction rate achieves 96.5%.

Mobile Crowd Sensing (MCS) is a data collection technology in which workers finish tasks and get payment. In MCS, some tasks are not preferred workers due to their remote locations or cheap prices, which leads to a huge proportion of unpopular tasks. Although increasing tasks payment is an effective to increase task popularity, however, it may decrease platform utility. In this work, we introduce bundling into MCS to solve this problem. Specially, a Task Bundling Reorganization Mechanism (TBRM) is proposed. In TBRM, unpopular tasks are properly bundled with popular tasks to maximize the minimum of both the number of task completions and expected profit. The TBRM is separated into two phases: the area selection phase and the rule selection phase. First, the randomly generated solution is input into the area selection phase, which selects the portion of the bundle that needs to be reorganized; then, the results of the area selection phase is regarded as input of the rule selection phase, which selects the appropriate task to reorganize; finally, the TBRM repeats this process until convergence. Experimental results demonstrate the effectiveness of the TBRM mechanism.

In this paper, we investigate the joint unmanned aerial vehicle (UAV) deployment and resource allocation problem to minimize the latency of multi-UAV-assisted computation offloading in air-ground integrated millimeter-wave (mmWave) networks, in which UAVs have both computing and relaying capabilities, thereby providing more opportunities for ground user equipments (UEs) to access the moble edge computing (MEC) servers with rich computing resources. Moreover, the study also takes into account the dynamic interference experienced by UEs due to different uploading completion times during the computation offloading process. To efficiently address the considered non-convex problem, we split it into four subproblems, i.e., UAV deployment, MEC server selection, computation resource and task ratio allocation, and power allocation subproblems, and solve them iteratively. Specifically, the first one is solved by three-dimensional-strategy iterative weekly acyclic game, the second one is addressed by Markov Approximation approach in which the third one is solved by the interior point method at each iteration, and the last one is solved by whale optimization algorithm (WOA). Finally, extensive simulations are provided to demonstrate the effectiveness of the proposed approach, and results have shown the approach can effectively mitigate the effect of blockage on mmWave transmissions and reduce the total latency of all UEs, particularly in scenarios where the communication bandwidth is limited or data volumes of tasks are large.

As an essential part of the future wireless networks, low earth orbit satellite networks (LEO-SN) is expected to achieve ubiquitous global networks access, in which the multibeam transmission is widely used to meet the increasing rate demand. However, in multibeam LEO-SN systems, the multiuser access and inter-group interference are crucial issues that need to be addressed urgently. To this end, we investigate the system weighted sum rate (WSR) maximization problem under the constraints of the user terminals (UTs) grouping, satellite total power budget, and minimum transmission rate requirements. For solving the problem, we propose a multiuser space-ground multicast group control (MU-SGMGC) scheme. Specifically, we first group all UTs into multiple multicast groups based on the channel correlation coefficient. Then, the group centers determination algorithm based on user distribution is proposed to ensure that each beam can cover all user groups. Finally, the beamformers designing problem is transformed into a difference-of-convex (DC) programming problem by utilizing auxiliary variables, and an iterative algorithm based on convex-concave procedure (CCP) is presented to solve the problem. Simulation results show that our proposed MU-SGMGC scheme has significant superiority in system WSR compared with the benchmark algorithms.

The proliferation of satellite launch technology has prompted the rise of Low Earth Orbit (LEO) satellite constellations (LSC) as an effective complement to improve network coverage. However, few LSC are designed to guarantee quality of service (QoS) within link budget constraints, leading to wasteful resource utilization. In this paper, we consider QoS and link budget as constraints and formulate the LSC design problem as a multi-objective optimization problem (MoP). We focus on optimizing the configuration of the LSC to achieve maximum seamless multi-coverage and link capacity while minimizing its cost. To improve the population diversity and approach the optimal solution, we propose an Improved Non-Dominated Sorting Genetic Algorithm-II (INSGA-II) to solve this MoP, then the optimal LSC is designed considering different elevation angle constraints. Furthermore, a performance comparison with existing state-of-the-art constellations is presented. The results show that the designed LSC exhibits comparable and even superior performance to Telesat and Kepler, while the constellation scale is only 0.64 times that of the Kepler system.

Within smart home systems, the heating, ventilation and air conditioning (HVAC) system plays a critical role in regulating the indoor thermal environment, however, HVAC, which consume 40% of total building energy, are very energy-intensive. It is essential to design HVAC control strategy that reduce energy consumption while maintaining a satisfactory thermal environment. Considering that the human ability to control the thermal environment is underutilized in the existing research, this paper proposes a framework that involves human in the loop (HITL) and reinforcement learning (RL) to improve HVAC strategy under human guidance. We formulate the optimization problem as a problem of minimizing energy cost and thermal comfort cost. A human behavior model is then created to simulate the human behavior in a variety of thermal environments. We propose a HITL- deep deterministic policy gradient (DDPG) algorithm for thermal comfort and HVAC energy optimization based on human guidance and DDPG. We built a simulation environment based on proposed framework for strategy train and performance evaluation. The results show that HITL-DDPG can reduce HVAC energy consumption by 38.7% while improving occupant thermal comfort by 31.1%.

As the Medical Internet of Things (MIoT) is rapidly evolving, increasing wearable devices are applied to collect various physiological signals for the purpose of medical applications. Currently, it's challenging to process multi-modal physiological signals on wearable devices because of their limitations of energy and computing resources, as well as the threat of data privacy leakage. To address aheadmentioned problems, this paper designs a novel scheme to process multi-modal physiological signals, namely Multi-modal Compression and Encryption Network (M-CEN) with GAN and AE. The proposed MCEN accomplishes the encryption and compression of data by the adversarial property of GAN and the feature extraction property of AE. The proposed scheme aims to protect user privacy and reduce the consumption of energy and computing resources. In addition, the proposed MCEN is simulated using the Pulse Transit Time PPG dataset. The results show that the proposed scheme can well encrypt the data, and the compression performance is improved by 36.3% and 61.3% compared to the CAE and WT Algorithm respectively.

Alzheimer's disease (AD) is a common neurodegenerative disease that damages the health of the aged. To precisely diagnose Alzheimer's disease, a widely-accepted approach is to extract the features of resting-state functional Magnetic Resonance Imaging (rs-fMRI). Existing work fails to effectively explore the spatial dependency among brain regions and the temporal dynamics of brain activity. Furthermore, these methods are also limited by the scale of the dataset. This paper proposes a novel transformer-based method, namely, spatiotemporal graph transformer network (STGTN), which can effectively extract spatiotemporal features of rs-fMRI for accurate diagnosis. STGT-N unites temporal transformer and spatial transformer, which incorporates functional connectivity (FC) of rs-fMRI as edge features in the constructed brain graph. To break the limitation of sample size, we use adversarial training to generate adversarial examples (AEs) by the proposed STGTN. Experiments are conducted based on the ADNI dataset. The results show that the proposed model achieves accuracy of 92.58% for the task of normal control (NC) vs. AD classification, and 85.27 % for the task of early mild cognitive impairment (eMCI) vs. late mild cognitive impairment (lMCI) classification, respectively.

Vehicular networks require massive communication connections between vehicles and infrastructure to support the high data rate vehicular service applications. However, due to the obstruction of buildings in urban areas, the channel capacity of vehicle-to-infrastructure (V2I) links will be deteriorated. Thus, intelligent reflecting surface (IRS) is introduced to aid vehicular communications to increase the channel capacity of V2I links. In this paper, we aim to maximize the sum V2I capacity by jointly optimizing the transmit beamforming matrix at the base station (BS) and the phase shifts at the IRS. Most of the existing works adopt alternating optimization-based iterative algorithms to tackle the joint beamforming and phase shift optimization problem, which suffer from high computational complexity. Therefore, we propose an unsupervised learning (UL)-based algorithm with a two-stage network architecture to address the joint optimization problem. The network architecture consists of a two-stage transformer network, which can implicitly learn the spatial and temporal features of historical channels to further improve the learning performance. Simulation results show that the proposed UL-based algorithm can obtain the comparable performance with much lower computational complexity compared with the conventional alternating optimization-based iterative algorithm.

The growth of the Internet of Things has given rise to various innovative services such as self-driving vehicles, remote operations, immersive technologies, tactile Internet and automated factories. Machine Type Communication (MTC) data, primarily transmitted through short packets, is the backbone of these applications. However, short packets have limited error correction capabilities, thus it is necessitating the use of retransmission techniques to increase the reliability of Short Packet Communication (SPC). While retransmissions exploit time diversity to lower the bit error rate, excessive retransmissions can degrade system performance. Consequently, refining retransmission policies is a crucial research area for SPC systems. In this paper, an SPC system model with Incremental Redundancy Hybrid Automatic Repeat reQuest (IR-HARQ) is studied to optimize the number of retransmissions under various conditions by determining an ideal block-length selection policy. A Probabilistic Q-learning (PQL) algorithm based on the two-factor theory is proposed to determine the optimal block length selection by striving to optimize the retransmission. Simulation results show that the proposed algorithm can effectively improve the performance of retransmission system.

In the Healthcare Internet of Things (H-IoT), Federated Learning (FL) is a promising solution for processing huge amounts of medical data. At present, FL applied in H-IoT still faces many challenges such as low training efficiency and high data privacy risk. In this work, we develop a three-layer FL architecture, which introduces split learning to both prevent the leakage of medical data and improve training efficiency according to the inherent properties of Neural Networks (NN). Moreover, we formulate a long-term optimization problem with the goal of accelerating training speed of models in H-IoT. Then, an online model partitioning algorithm namely Privacyaware Model Partitioning Algorithm (PMPA) is derived based on Lyapunov optimization theory that enables mobile devices of the FL architecture to efficiently train local models and protect the data privacy. Furthermore, the simulation results show that compared with traditional FL, the local training delay of the proposed algorithm can be reduced by 28.94% and 39.89%, respectively.

Mobile crowdsensing (MCS) has become an attractive issue in recent years. Most existing researches either select opportunistic sensing or participatory sensing for task execution, which will lead to the problem of restricted task locations or high cost. In this work, we propose a complementary hybrid worker selection method for MCS, where workers complete tasks in different sensing modes, namely opportunistic and participatory sensing. The proposed worker selection method contains two phases. In the opportunistic worker selection phase, an updated iterative algorithm is designed to select a low-cost and high-coverage opportunistic worker set. Specifically, when an opportunistic worker is selected, the algorithm will update the coverage of the remaining candidate opportunistic workers on the sensing task. In the participatory worker selection phase, we design an algorithm that combines group and match to solve the problem of restricted task locations. Specifically, we group the sensing tasks that opportunistic workers have failed to cover and recruit participatory workers to complete the sensing tasks in the groups. Experiments on a real dataset prove that the proposed method outperforms other benchmark methods.

In this paper, we investigate the reliability of intelligent reflecting surface (IRS)-assisted unmanned aerial vehicle (UAV) communications in the case of limited UAV energy. Under constraints of the UAV energy and the channel decoding error rate, we formulate a reliability maximization problem by jointly optimizing the IRS’s scheduling, the UAV’s trajectory, the IRS’s phase shift, and the UAV’s transmit power. Since the partial constraints of the problem are strictly nonconvex and its variables are coupling, the problem is difficult to convert to a nonconvex problem. Therefore, we propose a chaotic adaptation hybrid whale optimization algorithm (CAHWOA) to solve the problem. CAHWOA is implemented by using alternately the chaotic adaptation whale optimization algorithm (CAWOA) and the binary optimization algorithm (BWOA). Simulation results demonstrate that the joint optimization of IRS and UAV can improve the system communication reliability by almost 32% compared with the two baseline schemes. CAHWOA can improve the convergence rate by nearly 20% and enhance the optimization-seeking accuracy by about 0.04 compared with the three baseline algorithms.

Low earth orbit (LEO) satellite networks have become one of the hot research areas as an essential part of satellite communication networks. The dynamic topology and unbalanced traffic demand may lead to intersatellite link (ISL) congestion; thus, improving network load balancing performance is one of the key issues to be addressed in LEO satellite networks. We propose a load-balanced collaborative offloading (LBCO) strategy to achieve a balanced traffic distribution in LEO satellite networks. LBCO strategy consists of two algorithms, namely, channel-aware gradient fair association (CAGFA) algorithm and ISLs collaborative offloading (ISLCO) algorithm. The CAGFA algorithm aims to maximize the aggregate weighted utility, and the ISLCO algorithm aims to achieve the traffic offloading and download observation data from the LEO satellite network. Specifically, we first determine the actual downloading satellite set and neighboring satellite set by constructing an earth station (ES) time-share graph and a space–time topology graph. Then, the LBCO strategy uses the CAGFA algorithm to obtain the optimal satellite terminal association indicator and the load of downloading satellites. Finally, the ISLCO algorithm is proposed to achieve proportional offloading of traffic among the neighboring satellites and download massive observation data. Simulations show that the proposed CAGFA algorithm improves the weighted utility by 3.3% and the convergence by 47.6% compared with the benchmark stochastic gradient descent-based association (SGDA) algorithm. We also validate the performance of the LBCO strategy by data download throughput, which performs better than the other benchmark algorithms under three different load scenarios.

Intelligent logistics empowered by artificial intelligence (AI) has become an inevitable trend in the development of modern logistics, thus, a convenient and efficient logistics system has attracted widespread attention. However, how to use AI to execute computation-intensive applications on resource-constrained logistics vehicles (LVs) still faces enormous challenges. For the dependent applications in intelligent logistics, this article investigates a task dynamic offloading strategy for LVs-edge collaboration with multiple dependent tasks, considering the intertask dependency, to guarantee the Quality-of-Service (QoS) requirements of LVs. First, the dependent application ARCore is modeled and transformed into a model with a linear execution sequence. Then, based on this task model, the joint task offloading and resource allocation problem is formulated. The goal is to minimize the weighted sum cost of the execution delay and energy consumption while guaranteeing the delay tolerance and computing resource constraints of the tasks. Furthermore, we propose a federated LVs-edge collaborative computation (FECC) offloading framework to solve the optimization problem, which only requires each agent to share its model parameters without sharing local training data, thereby reducing the computation complexity and signaling overhead of the multiagent training process. Numerical results show that the proposed strategy has significant advantages in terms of total system cost compared to the baseline strategy.

This paper investigates user grouping and beam-forming design in multi-beam low earth orbit (LEO) satellite communication (SATCOM) systems. To serve a great many user terminals (UTs) with a limited number of beams and improve the system performance, we formulate the weighted sum rate (WSR) maximization problem subject to the constraints of the UTs grouping, the satellite total power, and the minimum rate requirements of UTs. For solving this problem, we propose a multi-group multi-beamforming (MGMBF) scheme. In this scheme, all UTs are firstly adaptively grouped based on the channel correlation coefficients. Further, the beam centers are determined to ensure that all UTs are covered. After UTs grouping, slack variables are introduced to convert the beamforming design into a difference-of-convex (DC) programming problem. Moreover, an iterative algorithm is presented to solve the problem based on the convex-concave procedure (CCP), in which the beamforming vectors and slack variables are updated jointly by solving the convex sub-problem. Simulation results demonstrate that the MGMBF scheme improves the WSR by 25.1% compared with the MBIM algorithm, verifying the significant advantages of the proposed scheme.

Epilepsy is one of the most common neurological diseases in the world. Feature extraction of electroencephalogram (EEG) is very important for predicting epileptic seizures. Conventional technologies of EEG signals analysis mostly utilized supervised learning methods with a mass of labeled data. However, annotating data is a time-consuming and expensive process. In this paper, we propose a novel unsupervised feature learning method based on variational autoencoder, namely, residual convolution variational autoencoder (RCVAE), which aims to improve the accuracy of epileptic seizure prediction. RCVAE automatically extracts important features and reconstructs the spatiotemporal EEG signal, reducing the learning difficulty with residual network structure. In addition, this work also utilizes the Pearson correlation coefficient and the reconstructed loss function, which are used to evaluate the quality of the reconstructed signal. Finally, the performance of the proposed model is verified on the CHBMIT dataset, the accuracy rate is up to 96.17%, and the false alarm rate is only 0.015.

To address the impact caused by the large increase of data traffic in internet of vehicles (IoV), vehicular edge caching, as an effective technology to alleviate the above issue, attracts extensive attention. However, most existing studies in vehicular edge caching only considered the temporal feature of content popularity, which will impact its accuracy. Thus, we propose a spatial-temporal correlation multi-agent caching policy (STC-MACP) to dynamically determine where to cache and what to cache. Firstly, we predict the content popularity based on the spatial-temporal correlation of the historical content requests. Secondly, multi-agent reinforcement learning (MARL) is applied to solve the caching decision optimization problem to obtain the optimal caching policy with maximizing delay reduction. Finally, we conduct the IoV simulation environment, and simulation results show that the STC-MACP can effectively reduce the content access delay. Compared with the most popular caching (MPC), the content access delay of STC-MACP is decreased by 28% when the Zipf parameter is 0.8.

Low earth orbit (LEO) satellites have a pivotal role in global data monitoring. However, one of the most challenges is load balancing between LEO satellite networks due to the frequent topology changes and uneven distribution of global ground users. Thus, we propose a channel-aware gradient fair association (CAGFA) strategy to maximize the aggregate weighted utility in a dynamic satellite environment. Specifically, we first determine the downloading and neighboring satellites by constructing an earth station (ES) time-share graph and a space-time topology graph, respectively. Then, the CAGFA strategy is designed to obtain the optimal satellite terminal association indicator, and intersatellite links (ISLs) collaborative offloading is used to achieve the traffic balance for the LEO satellite networks. Simulation results indicate that the proposed CAGFA strategy increases the weighted utility by 6.5% and the convergence by 34.7% compared with the root mean square propagation-based association (RMSPA) strategy.

In 6G Internet of Vehicles (IoV) system, Federated Learning (FL) is usually used to structure the joint training model between vehicles and RSU. However, due to the mobility of the vehicles, the link between vehicles is unstable and the parameters trained by FL are exchanged frequently, which may increase the communication overheads. Therefore, we propose a communication-efficient Federated Double Distillation (FedDD) framework in this paper. In particular, the cluster-heads are dynamically selected as the distributed learning clients combined with three-dimensional attributes to improve the collaborative transmission efficiency. Then, the knowledge distillation is further integrated into the federated learning framework to reduce communication overheads caused by the frequent parameters exchange in the instable link. The experimental results show that, compared with the benchmark FedAvg algorithm, the FedDD reduces the communication overheads by three orders of magnitude. Moreover, the FedDD improves the communication efficiency of FL while sacrificing only a small amount of accuracy.

The mobile crowdsensing (MCS) is an emerging sensing paradigm based on the mobile device. For location-dependent sensing tasks (LDSTs), when tasks are farther with low payment from workers, they can be difficult to complete. The completion rate of this unpopular task has always been an issue. Most existing researches mainly focus on how to increase payment for unpopular tasks, but the platform may suffer from it, because an incorrect increase results in an inability to raise the number of completed tasks. In this paper, we present a task bundling reorganized mechanism (TBRM) to improve the platform utility of MCS system. In the proposed mechanism, the unpopular and popular tasks are properly bundled to improve the platform utility. To decrease searching time for suitable bundles, two sub-policies are respectively utilized to design TBRM based on reinforcement learning: the area selection policy and the rule selection policy. Experimental results demonstrate that TBRM outperforms the three benchmark mechanisms, which reveals that TBRM can effectively bundle unpopular tasks and improve platform utility.

In the field of mobile crowd sensing (MCS), worker selection is a key research issue and has progressively gained considerable interests in the academic community in recent years. The goal of worker selection is to choose the superior workers for tasks that require high-performance characteristics. To solve the problems of long delay and poor perceived quality, we present a worker selection architecture for a recommendation system applied to the MCS system. A worker selection algorithm with high quality of service (QoS) is designed within the architecture, which considers the worker’s reputation and willingness attributes to address the challenge of efficiently selecting excellent workers. Based on these two attributes, we then compute worker QoS and develop a three-dimensional tensor to optimize the worker’s service. Finally, we get a continuously updated list of workers. Extensive experiments on real-world datasets show that the proposed algorithm performs better than the benchmarks, including random, greedy, and matrix-based algorithm. The results indicate that the proposed algorithm’s efficiency has risen by 31% compared to the matrix-based algorithm.

The computation-intensive applications pose unprecedented demands on the Internet of Vehicles (IoVs). How to address the delay constraint to execute the computation tasks effectively becomes a significant issue for this scenario. Compared with remote cloud, edge servers reduce the delay by being deployed close to vehicles. However, most edge servers are connected to fixed access points, which leads to the inflexible edge computing architecture. Considering the dynamics of vehicles’ location and service request, it is a promising paradigm that multi-vehicle compute the task collaboratively by utilizing the vehicles’ available computing resources. In this paper, by jointly considering the local execution, V2V offloading, and multi-vehicle collaboration, we determine the optimal task partition ratio after the cooperative vehicles are selected. Then, double deep Q-network (DDQN) is used to take the optimal dual actions. Finally, we develop a multi-vehicle intelligent collaborative computing strategy (MV-ICCS) to minimize the total system delay. Simulation results show the advantage of the proposed strategy and evaluate the system performance.

The ever-increasing user requests for video services have posed a great challenge to the cellular networks, and the emergence of various new services also puts forward higher requirements to the mobile networks. By caching video contents in cache-enabled heterogeneous networks, the delivery delay of content and the stress of backhaul links can be improved conspicuously. However, how to store diverse contents has got much attentions in the past decade. In this paper, considering the time-varying user requests, a hierarchical cooperative caching strategy with user preference is proposed. Firstly, the caching of content is modeled as a delay optimization problem. Secondly, the historical request data is used to predict the user preference, and the singular value decomposition (SVD) model is further used to predict the missing rating data. Thirdly, both the user preference and rating matrix are used to optimize the caching strategy. Finally, the proposed caching strategy is validated using the MovieLens dataset, the results reveal that the proposed strategy improves the delay performance by at least 35.3% compared with the benchmark strategies.

To support the diversified Quality of Service (QoS) requirements of application scenarios, network slicing has been introduced in the mobile cellular network. It allows mobile cellular network operators to accomplish the creation of multiple logically isolated networks on common network infrastructure flexibly depending on specified demands. Meanwhile, in Internet of Vehicles (IoV), it is very intractable to supply a stable QoS for the vehicles, especially for the dynamic vehicular environments. Thus, we investigate the IoV slicing problem in this article, and propose a QoS guaranteed network slicing orchestration, namely, the long short-term memory-based deep deterministic policy gradient algorithm (LSTM-DDPG), to ensure the stable performance for the slices. Specifically, we first decouple the resource allocation problem into two subproblems. After that, the deep learning and reinforcement learning (RL) are used to allocate resources collaboratively to solve these two questions. We use deep learning LSTM to track the characteristic of the long-term vehicular environment changing, and the RL algorithm DDPG is utilized for online resource tuning. Extensive simulations have proved the effectiveness of the LSTM-DDPG, which can offer stable QoS to the vehicles with a probability greater than 92%. We also demonstrated the adaptiveness of the proposed orchestration with different slicing environments, and the performance is always optimal compared to that of other algorithms.

Multi-access mobile edge computing (MEC) is envisioned as a key enabling technology to support compute-intensive and delay-sensitive applications in railway Internet of Things (RIoT) networks. However, the time-varying channel variations in RIoT scenarios make it challenging to achieve efficient resource allocation. The emerging deep reinforcement learning (DRL) is able to respond to the above-mentioned challenge. In this paper, with the aim of reducing the total computational cost (weighted sum of consumed energy and delay), we investigate the dynamic resource management issue of joint subcarrier assignment, offloading ratio, power allocation and computation resource allocation in multi-access MEC assisted RIoT networks. To address this intractable mixed integer nonlinear programming issue, we put forward a hybrid DRL (HDRL) scheme, which is an integration of deep double Q-learning (DDQN) and deep deterministic policy gradient (DDPG). The HDRL algorithm is capable of learning the advisable strategies for actions including discrete-continuous hybrid variables. In HDRL algorithm, DDQN plays the role of making subcarrier assignment decision, and DDPG plays the role of making offloading ratio, power allocation as well as computation resource allocation decisions. Numerical results demonstrate that HDRL scheme can yield much less computational cost than the existing baselines for multi-access MEC assisted RIoT networks. In addition, the HDRL scheme is close to the near-optimal performance with comparatively low execution time.

Collaborative caching in edge-user architecture has been regarded as one of the most promising technologies to release the pressure of core networks and reduce the content download delay. However, the caching resources at edge servers and devices are limited, so how to utilize their cache space efficiently has become a significant issue. This paper introduces a three-tier caching framework consisting of a macro-cell base station (MBS), multiple small-cell base stations (SBSs), and user equipments (UEs). In this framework, by combining physical and social attributes, we propose a directed graph-based edge-user collaborative caching (DG-EUCC) strategy to minimize the content download delay. Specifically, the wireless communication networks between the different types of nodes at the SBSs tier and the UEs tier are simplified to a one-tier directed graph (DG) with social attributes, to realize the simplification of the system model. Further, we design a DG-based collaborative caching strategy to minimize the content download delay, where each node caches the most popular contents according to the weighted content popularity set. Simulation results show that, compared with the benchmark strategies, the proposed DG-EUCC strategy can effectively reduce average download delay.

Mobile edge computing (MEC) enables resource-constrained mobile devices (MDs) to offload their tasks onto nearby edge servers. However, there exists a profit allocation problem between users and edge nodes (ENs) due to the limi-tations of ENs computing capacity and spectrum resources. In this paper, we propose an auction pricing-based MEC offloading strategy to maximize the profit of ENs. Firstly, we design an overall auction process using the binary offloading model by considering MDs battery capacity, basic profit, and tasks tolerable delay. Secondly, the bidding willingness of MDs in each round of auction are given on the premise of effectively ensuring users rationality. Finally, an auction pricing-based task offloading strat-egy is proposed, in which the winner of a single-round auction can offload its computation task to the ES. Simulation results verify the performance of the proposed strategy. Compared with the VA algorithm, the profit obtained by ENs has increased by 23.8%.

Network slicing can support the diverse use cases with heterogeneous requirements, and has been considered as one of the key roles in future networks. However, as the dynamic traffic demands and the mobility in vehicular networks, how to perform RAN slicing efficiently to provide stable quality of service (QoS) for connected vehicles is still a challenge. In order to meet the diversified service request of vehicles in such a dynamic vehicular environment, in this paper, we propose a two-timescale radio resource allocation scheme, namely, LSTM-DDPG, to provide stable service for vehicles. Specifically, for the long-term dynamic characteristics of service request from vehicles, we use long short-term memory (LSTM) to follow the tracks, such that the dedicated resource allocation is executed in a long timescale by using historical data. On the other hand, for the impacts of channel changes caused by high-speed movement in a short period, a deep reinforcement learning (DRL) algorithm, i.e., deep deterministic policy gradient (DDPG), is leveraged to adjust the allocated resources. We prove the effectiveness of the proposed LSTM-DDPG with simulation results, the cumulative probability that the slice supplies a stable performance to the served vehicle within the resource scheduling interval can reach more than 90%. Compared with the conventional deep Q-networks (DQN), the average cumulative probability has increased by 27.8%.

Mobile CrowdSensing (MCS) aims to accomplish task requesters sensing tasks by recruiting quantities of workers. Thus, it is crucial to match and distribute sensing tasks to workers efficiently. We consider the influence of transmission outage probability to obtain the best matches and maximize overall social welfare. The workers cannot receive sensing tasks if the outage probability is large, in this case, the social welfare will be smaller. Thus, we maximize overall social welfare via four steps: winner selecting, matching, transmitting and pricing, which is developed as maximize social welfare (MSW) mechanism. Experiment results show that compared with benchmark algorithms, our mechanism transmit sensing tasks to far workers with a lower outage probability and achieve larger social welfare. Furthermore, outage probability is decreased by 19% and overall social welfare created by accomplishing sensing tasks is improved by 15%.

In order to mitigate the impact of explosively increasing data traffic on content request services in the Internet of Vehicles (IoV), edge caching technology is implemented in IoV to accelerate the response process of content requests and release the backhaul burden of the base station. However, the content popularity obtained by the traditional content popularity method cannot capture the requests of vehicles accurately due to the time-varying characteristics of the content popularity, which results in a relatively low cache hit ratio. Thus, this article proposes a cooperative caching strategy with content request prediction (CCCRP) in IoV, which precaches the contents requested by vehicles with greater probability in other vehicles or the roadside unit (RSU) to reduce the content acquisition delay. Specifically, vehicles are first clustered using the K-means method to simplify the process of vehicle requesting and content transmission. Then, content requests from vehicles are predicted using the long short-term memory (LSTM) networks according to the historical content request information. Finally, reinforcement learning method is adopted to solve the objective function to obtain the optimal caching decision, which improves the Quality of Service (QoS) of vehicle requests. Simulation results demonstrate that CCCRP can improve the cache hit ratio and reduce content acquisition delay effectively. For example, the cache hit ratio of CCCRP can be increased by 5% and 7% compared to the traditional LFU and LRU caching strategies when the Zipf parameter equals 0.7, respectively.

To fulfill the diversified requirements of vehicular network services, we design an intelligent slice coordinator in this paper, which consists of two parts, service clustering and slice scheduling. In the first part, service clustering captures the Service Level Agreement (SLA) of services and clusters them based on K-means++ clustering algorithm according to the similarity of service requirement. Meanwhile, the services will be mapped into different slices. In slice scheduling module, we design the shared proportional fairness scheme (SPFS) to deal with the imbalance of radio resource utilization, and then further design the resource allocation algorithm based on linear programming obstacle method to solve the optimal slice weight distribution and maximize the slice load variation tolerance. Simulation results show that the SPFS has smaller average bit transmission delay (BTD) than the static slicing scheme, and the optimal slice weight distribution can be obtained under different user load distribution scenarios. The BTD gain achieves 1.5632 in the uniform user load scenario with 20 users per slice.

Based on the research of network and computing, while considering the delay of vehicular networks, this paper proposes an intelligent task offloading framework that can dynamically schedule network and computing resources to improve the performance of next-generation vehicular networks. Considering the task computing problem of the vehicles and the applications of the mobile edge computing (MEC), an intelligent task offloading joint optimization algorithm is proposed in this paper. The algorithm firstly uses the K-Nearest Neighbors (KNN) method to select the offloading platform (i.e., cloud computing, mobile edge computing, local computing) of the computing task. Considering the computing resource allocation problem and the complexity of the system, the algorithm secondly uses the reinforcement learning method to solve the resource allocation problem effectively. Simulation results show that, comparing to the baseline algorithm that all tasks are offloaded to the local or MEC server, the proposed algorithm achieves a significant reduction in latency cost. Compared with the Ful1 MEC” algorithm, the proposed algorithm can save 80% of the average system cost.

The diversified service requirements in vehicular networks have stimulated the investigation to develop suitable technologies to satisfy the demands of vehicles. In this context, network slicing has been considered as one of the most promising architectural techniques to cater to the various strict service requirements. However, the unpredictability of the service traffic of each slice caused by the complex communication environments leads to a weak utilization of the allocated slicing resources. Thus, in this paper, we use Long Short-Term Memory- (LSTM-) based resource allocation to reduce the total system delay. Specially, we first formulated the radio resource allocation problem as a convex optimization problem to minimize system delay. Secondly, to further reduce delay, we design a Convolutional LSTM- (ConvLSTM-) based traffic prediction to predict traffic of complex slice services in vehicular networks, which is used in the resource allocation processing. And three types of traffic are considered, that is, SMS, phone, and web traffic. Finally, based on the predicted results, i.e., the traffic of each slice and user load distribution, we exploit the primal-dual interior-point method to explore the optimal slice weight of resources. Numerical results show that the average error rates of predicted SMS, phone, and web traffic are 25.0%, 12.4%, and 12.2%, respectively, and the total delay is significantly reduced, which verifies the accuracy of the traffic prediction and the effectiveness of the proposed strategy.