最新刊期
卷 9 ， 期 4 ， 2025
- 摘要：Mobile edge computing (MEC) has become a key technology in 6G by bringing computing and storage capabilities from centralized data centers to the network edge, effectively meeting the demands of multiple application scenarios. However, in the complex environments, the line-of-sight links between users and MEC servers are prone to being blocked by obstacles, significantly limiting offloading capabilities. Integrating reconfigurable intelligent surface (RIS) into MEC systems enhances the wireless signal transmission environment, thereby offering an effective solution to this challenge and positioning RIS-assisted MEC as an important research focus in future communication technologies. Firstly, the basic concepts and application scenarios of MEC were outlined, and the recent developments of RIS were reviewed. Then, the optimization objectives, research methods, and applications for RIS-assisted MEC systems were discussed in detail. Finally, the future research trend of combining RIS with MEC was envisioned and analyzed in-depth in combination with the advanced technologies such as integrated sensing, computation, and communication and low-altitude economy, aiming to provide valuable references and insights for researchers in related areas.关键词：RIS;MEC;computation offloading;resource allocation197|70|0更新时间：2026-01-04
Review
- 摘要：The application of alliance chain in vehicle networking to achieve road traffic observation information management can take into account the decentralization of public chain and the efficient characteristics of private chain, and improve the transparency of communication between vehicles. However, existing schemes do not adequately consider the risk of malicious vehicles and malicious road side unit (RSU) nodes spreading false information, which can lead to urban traffic paralysis and accidents. In addition, the transmission of massive vehicle data and repeated incident reports increases the computing and communication overhead, which affects the feasibility and scalability of the system. Aiming at the above problems, a control transmission and efficient and secure sharing scheme for interactive information in the vehicle-connected scene was proposed. Firstly, an information transmission control strategy based on angular difference self-transforming area code was designed. The amount of information forwarded was limited by setting the threshold of vehicle area code distance, and Bloom Filter was adopted to filter information to reduce the transmission of redundant information. Secondly, a management mechanism based on credit points was established to deal with the damage to the system caused by malicious nodes spreading false information. Finally, the DMML-Hotstuff (dynamically manage multiple leader Hotstuff) consensus algorithm was proposed, which effectively solved the performance bottleneck of single master node. The system throughput and Byzantine tolerance were improved by dynamic node adjustment. The simulation results show that the proposed scheme reduces the vehicle information transmission by 23% and 39% compared with the existing methods, and can evaluate the credit score in real time and quickly identify and isolate malicious nodes. In terms of consensus algorithms, the throughput is improved by 19%, 52% and 188%, respectively, compared to the three mainstream Byzantine fault-tolerant algorithms.关键词：Internet of vehicle;alliance chain;consensus algorithm;reputation mechanism100|113|0更新时间：2026-01-04
- 摘要：With the rapid development of industrial Internet of thing (IIoT), the relationship between all entities of the network elements became more complex, dynamic and multidimensional. It was difficult to accurately describe the high-density, large-scale and complex interaction relationship of all elements in IIoT through existing information interaction technology. Focusing on the interrelation of resource allocation among network entities in macrocell scenarios, sum-rate maximizing problem in IIoT macrocell based on hypergraph was studied and constructed. Non-orthogonal multiple access (NOMA) technology was considered in this scenario to support multiple access between industrial Internet devices. Device-to-device (D2D) communication technology was introduced to reduce interference between devices and improve spectral efficiency. When constructing the interference model among femtocells and their internal devices, the femtocell access points (FAP) were defined as vertices, and the interference relationships were defined as edges and hyperedges. By analyzing the hyperedge relationships in the hypergraph, base stations causing less interference were allowed to share the same spectrum, and a maximum vertex-weight cluster algorithm based on the hypergraph was used to achieve optimal spectrum resource allocation and spectrum sharing matching. Simulation results demonstrated that, compared to graph-based resource schemes, the proposed method significantly improves spectral efficiency and exhibits better adaptability and scalability in high-density, large-scale, and complex environments, providing an effective solution for resource allocation in IIoT under ultra-dense wireless network conditions.关键词：industrial Internet of thing;non-orthogonal multiple access;device-to-device cluster;hypergraph;maximum clique problem;resource allocation200|97|0更新时间：2026-01-04
- 摘要：To improve the performance of cell-free (CF) networks, the hybrid precoding design was studied for reconfigurable intelligent surface (RIS)-aided CF networks, of which the capacity of the fronthaul link between the access points and central processing unit was limited. A weighted sum rate maximization problem was formulated by jointly optimizing the hybrid precoding matrices, RIS phase shift coefficients, and quantization noise covariance matrices while satisfying the users′ quality of service (QoS) constraints and fronthaul link capacity constraints. By applying the alternating optimization method, the optimization problem was divided into four sub-problems and solved in an iterative manner. To address the non-convex QoS constraints, the quadratic transform was adopted to approximate the constraints. For the non-convex constant-modulus constraints, a manifold optimization method was proposed under the alternating direction method of multipliers framework. The simulation results show that the proposed algorithm significantly outperforms the baseline algorithms and the deployment of RIS can effectively enhance the rate of the users.关键词：CF network;RIS;hybrid precoding;quantization and compression137|81|0更新时间：2026-01-04
- 摘要：Ultra-wideband (UWB) technology is recognized for its ability to provide high-precision positioning information, offering significant advantages in indoor positioning. Angle of arrival (AOA) estimation, as one of the key technologies in UWB positioning, is considered crucial for improving positioning accuracy. However, in practical applications, UWB AOA estimation is faced with numerous challenges, particularly in complex environments such as non-line-of-sight (NLOS) conditions, antenna hardware impairments and environmental variations. These factors are known to cause signal distortion and measurement deviations, thereby reducing estimation accuracy. Traditional modeling methods are often found inadequate in effectively addressing these nonlinear issues. Therefore, a feature fusion-based UWB AOA estimation method was designed. This approach integrated channel impulse response (CIR) data and dual-antenna received signal features, while a Transformer encoder was introduced to deeply analyze complex signals, thereby enhancing AOA estimation accuracy. Experimental results demonstrated that the proposed method could achieve high-precision AOA estimation under NLOS conditions.关键词：UWB signal;AOA estimation;deep learning;feature fusion;NLOS100|143|0更新时间：2026-01-04
- 摘要：Radio frequency (RF)-based indoor positioning technology is recognized as one of the important research directions in the sixth generation wireless communication (6G) systems. With the advancement of artificial intelligence (AI), deep learning-based indoor fingerprint localization methods have achieved significant improvements in positioning performance. However, these methods still face the following challenges, including lengthy RF data collection periods and high annotation costs, which lead to poor environmental generalization capability of existing deep learning algorithms across different scenarios. To address this issue, a few-shot transfer learning indoor fingerprint localization method based on a Siamese graph convolutional network (Siamese GCN) was proposed. The Siamese GCN model was combined with a maximum mean discrepancy-based domain adaptation approach, requiring only a small number of channel state information samples to be collected in the current environment. Pre-trained network weights from other environments were reused, significantly reducing data collection and annotation costs in new environments. To validate the effectiveness of the proposed method, real environmental data were collected in two typical indoor scenarios: a laboratory and a corridor. Experimental results demonstrated that the proposed transfer learning method achieved satisfactory localization performance using only 30% of the labeled samples.关键词：Siamese GCN;indoor localization;transfer learning;channel state information80|73|0更新时间：2026-01-04
- 摘要：Collaborative crowdsensing of heterogeneous entities has emerged as a pivotal research area in the field of collective intelligence in recent years, primarily focusing on how diverse agents, such as humans, drones, and unmanned vehicles, can collaboratively perceive and interpret their environments, which holds significant promise for applications in the emergency rescue and urban operations. However, existing task allocation algorithms for collaborative crowdsensing often struggle to balance allocation effectiveness with solution efficiency, and they frequently fail to facilitate deep collaboration among diverse entities. To address the partially observable conditions of environmental states during emergency rescue operations, a “hard collaboration” model for heterogeneous entities was introduced and a model for task allocation problem was formulated. To address this challenge, a multi-agent collaboration framework was developed and a task allocation algorithm for heterogeneous entities with partially observable environment states was proposed. Extensive experiments across four scenarios show that the proposed method outperforms the baseline algorithms in task completion rates, achieving an average of 84.40% ± 4.74% compared with the best baseline′s 65.98% ± 4.97%. Moreover, the proposed method exhibits strong robustness, maintaining commendable task completion rates even amid changes in perception scenarios and underscoring its potential for deployment in dynamic environments.关键词：emergency response;collaborative crowdsensing of heterogeneous entity;task allocation;multi-agent reinforcement learning73|67|0更新时间：2026-01-04
- 摘要：The self organizing manufacturing model was combined with the existing industrial interconnection technology, wireless network technology, distributed computing technology, and artificial intelligence technology, which can encapsulate traditional manufacturing resources into manufacturing units with high autonomy, adaptability, and functionality. Through interaction with other manufacturing units, self-organizing negotiation and allocation of manufacturing tasks were completed. To develop a rapid response mechanism for the intelligent allocation of manufacturing tasks and self-organizing resource distribution, the multi-agent contract net protocol (CNP) was integrated with the Monte Carlo tree search (MCTS) algorithm. Based on this, dynamic matching between manufacturing tasks and resources was achieved. Apart from this, a self-organizing operational mechanism and an optimization method were accomplished for the scheduling of manufacturing workshop control systems. Finally, the practical feasibility of the proposed method was verified through a discrete workshop task allocation case. The experimental results indicate that the method can better achieve the goal of "one-step negotiation and global optimization".关键词：multi-agent system;self organizing operation;Monte Carlo tree search;manufacturing task allocation;intelligent manufacturing59|44|0更新时间：2026-01-04
- 摘要：The optimization of Wi-Fi network performance typically constitutes a multi-parameter, multi-objective dynamic optimization problem, which presents significant challenges in mathematical modeling. Deep reinforcement learning (DRL), which obviates the need for complex mathematical formulations, has been widely applied in recent years to optimize Wi-Fi network performance. Meanwhile, generative diffusion models (GDMs) have achieved remarkable progress in modeling complex data distributions across various domains. Therefore, integrating DRL with GDMs can further enhance its capabilities in optimizing Wi-Fi network performance. The typical medium access control (MAC) mechanism in Wi-Fi network is the distributed coordination function (DCF), whose performance significantly degrades as the number of contending terminals increases. A deep diffusion deterministic policy gradient (D3PG) algorithm was proposed, which integrated diffusion models with the deep deterministic policy gradient (DDPG) framework to optimize Wi-Fi network performance. In addition, an access mechanism that jointly adjusted the contention window and the aggregation frame length based on the D3PG algorithm was proposed. Simulations have demonstrated that this mechanism significantly outperforms existing Wi-Fi standards in dense Wi-Fi scenarios, maintaining throughput performance even as the number of users increases sharply.关键词：Wi-Fi network;generative diffusion model;deep reinforcement learning;performance optimization;access control54|61|0更新时间：2026-01-04
- 摘要：Orthogonal time frequency space (OTFS) as one of the key candidate technologies for 6G, is recognized for its ability to effectively combat the effects of doubly-selective fading channels. However, channel estimation in OTFS systems has remained a major focus and challenge in academic research. In recent years, deep learning-based OTFS channel estimation schemes were proposed, which utilized artificial intelligence techniques to rapidly capture channel variations. Nevertheless, these existing algorithms were generally characterized by large network scales, making it difficult to meet the lightweight requirements of mobile terminals. To address this issue, an OTFS channel estimation algorithm based on a lightweight parallel denoising network was proposed with the aim of improving computational efficiency and reducing device power consumption. By integrating image denoising and data-driven concepts, the algorithm retained the strong generalization capability of deep learning methods while reducing the computational cost on mobile devices through optimized network architecture and reduced pilot power, thereby providing a new solution for lightweight terminal communication in high-mobility scenarios. The parameter quantity of the proposed algorithm was only 15% of that of the existing denoising convolutional neural network (DnCNN), significantly reducing both the network parameter scale and computational complexity. Simulation results demonstrated that, thanks to its unique parallel structure design, the proposed algorithm compensated for the estimation performance loss caused by lightweight design. Under a five-path fast time-varying channel, a performance gain of 4 dB was achieved compared to DnCNN.关键词：OTFS;channel estimation;deep learning;denoising network;lightweight60|57|0更新时间：2026-01-04
- 摘要：The core functions of low-altitude Internet of things (IoT), such as communication and navigation heavily rely on third-party libraries. Vulnerabilities in third-party libraries can lead to significant risks such as drone loss of control and data leakage. To address the limitations of existing vulnerability identification methods, such as difficul-ties in promptly detecting vulnerabilities in newly migrated libraries and inefficiencies when running on resource-constrained IoT devices, a migration library vulnerability identification method based on time factor optimization was proposed. By deeply mining migration information from open-source projects, six metrics, including temporal support and label support, were constructed to screen novel and lightweight migration libraries. A streamlined transformer model was employed to detect vulnerabilities in the selected libraries, which reduced the computational burden on edge devices and enabled light-weight yet accurate vulnerability identification. Experimental results demonstrated that the proposed method achieved an average F1-score of 0.78 in vulnerability identification tasks, outperforming mainstream approaches by more than 10%. Training time was reduced by approximately 58%, and the average prediction time was only 4.7 ms. The method effectively enhanced both the security and real-time performance of library migration in low-altitude scenarios, providing efficient protection for low-altitude IoT devices.关键词：low-altitude IoT;third-party libraries;vulnerability identification;time factor55|43|0更新时间：2026-01-04
- 摘要：Healthcare insurance fraud is becoming increasingly serious worldwide, posing significant threats to both the economy and the healthcare system. Compared with the widely studied prescription stage, the drug collection stage in the resale of returned drugs is the most covert and challenging key link in the entire fraud supervision chain. To address this issue, the concept of mobile crowd-sensing was adopted in the Internet of things (IoT), treating the suspect's mobile phone as a sensor to automatically collect multimodal data, including WeChat chat logs, which were then analyzed to construct a social Internet of things (SIoT) for the suspect. However, during the data processing and analysis phase, current relationship extraction methods face challenges in accuracy because of the multimodal heterogeneity of the data and variations in suspects' communication habits. To overcome these challenges, a suspect relationship extraction method based on large language models (LLMs) was proposed, applying LLMs for the first time to extract suspect relationships from social media chat logs in the drug resale process. Prompt templates and interaction optimization for the model were specifically designed for chat logs. This method has been applied in the healthcare insurance fraud supervision activities of the Shijingshan District People's Procuratorate in Beijing. Extensive experiments on the real-world data demonstrate that this approach, leveraging multiple mobile sensing nodes in the SIoT, significantly improves the processing capabilities of complex multimodal data and enhances the accuracy and efficiency of suspect relationship extraction. This method facilitates the in-depth investigation of key individuals involved in the upstream and downstream links of healthcare insurance fraud.关键词：large language models;health insurance fraud supervision;WeChat chat records;relation extraction52|86|0更新时间：2026-01-04
- 摘要：The cell-free massive multiple-input multiple-output (CF-MMIMO) system integrates the advantages of distributed antenna systems and massive MIMO, enabling significant improvements in user coverage and spectrum efficiency through the collaboration of multiple access points (AP). It is a highly promising architecture for future 6G communication networks. However, existing detection algorithms for CF-MMIMO systems fail to achieve a good balance between complexity and detection performance. To address these challenges, a low-complexity optimal signal detection algorithm was proposed for uplink CF-MMIMO systems, called Richardson semi-iterative network (RSI-Net), based on deep unfolding networks. The Richardson semi-iterative (SI) theory was introduced, and replaced the existing parameter estimation scheme with a deep unfolding network (DUN) trained with hidden layer parameters to achieve adaptive parameter estimation in response to changing channel statistics. Additionally, to accelerate convergence, a scaling factor was introduced to improve the distribution of eigenvalues in the iterative matrix. Simulation results demonstrated that RSI-Net algorithm maintained low computational costs and excellent detection performance in CF-MMIMO systems with weakened channel hardening characteristics, regardless of changes in the number of users or AP.关键词：cell-free;massive multiple-input multiple-output;signal detection;semi-iterative;deep unfolding66|42|0更新时间：2026-01-04
- 摘要：With the rapid development of mobile network, cloud services and other technologies, the number of services in the Internet of Things environment is gradually increasing, the types are gradually diverse, and the correlation between the services is more complex. This correlation will affect the quality of service(QoS) performance of services, so it is necessary to consider the correlation between services in the process of service composition. In order to solve this problem, services were modeled by the intention and the context. The service relationship network was described by constructing the hypergraph. The set of clustered services and the correlation between the services are represented by hyperedges. On this basis, the NSGA-Ⅲ multi-objective optimization algorithm is used to complete the service composition, which covers the influence of the correlation between the services represented by hyperedges on QoS performance. The experiment proves that the hypergraph model can describe the service and the correlation between the services well, and improve the QoS quality of the service combination solution solved by the multi-objective optimization algorithm.关键词：hypergraph;service modeling;correlation;QoS;service composition30|47|0更新时间：2026-01-04
- 摘要：As the domestic electricity market reform advanced in China, market participants are required to understand the electricity price trends to flexibly adjust production plans and procurement strategies. Thus, the demand for accurate electricity price forecasting increases steadily. In actual forecasting, abnormal electricity price data, including missing values and mislabeling, cause non-smooth model training. In response to these problems, the K-nearest neighbors (KNN)- random forest (RF) method was first applied. This method, capable of capturing global features, was used to accurately identify and replace abnormal data points. Then, variational mode decomposition (VMD) was employed to decompose electricity price data into multiple sub-modes. Finally, a convolutional neural networks (CNN)-bi-directional long short-term memory (BiLSTM) combined model was utilized for forecasting, and the results were integrated to obtain the final day-ahead electricity price forecast. The Simulation results show that, compared with the basic model, the combined KNN-RF-VMD-CNN-BiLSTM electricity price forecasting algorithm achieves relative improvements of 15.8%, 13.6%, 1.54%, and 32.4% in mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), and mean absolute percentage error (MAPE) indicators, respectively. Moreover, the inference time for a single epoch is within seconds. This algorithm effectively balances the forecasting efficiency and accuracy.关键词：day-ahead electricity price forecast;anomaly detection;CNN-BiLSTM;KNN;RF43|34|0更新时间：2026-01-04
- 摘要：Improper behavior of escalator passengers can easily lead to public safety accidents and property losses. Accurately identifying dangerous behaviors of escalator passengers based on surveillance videos is of great significance for ensuring public safety. However, existing behavior recognition methods rarely focus on the dangerous behaviors of passengers in escalator scenes, and lack modeling and analysis of spatial-temporal interactions between people and escalators. Therefore, spatio-temporal information from human skeleton and human-object interactions were extracted, and a two-stream human-object interaction graph convolutional network considering distance metrics to identify dangerous behaviors of escalator passengers was designed. Firstly, features from both human skeleton and escalator keypoints were extracted, supplementing scene information for human skeleton features using escalator keypoints. Secondly, distance metrics between humans and escalators to dynamically capture changes in human-object relationships within dangerous behaviors was utilized, enhancing the model's modeling of spatio-temporal interaction information in dangerous behaviors. Finally, to fill the gap in existing publicly available datasets regarding videos of dangerous behaviors on escalators, a dataset called ESC-Danger for escalator passenger dangerous behaviors was constructed. This dataset contains eight classes of escalator passenger dangerous behaviors, including lean, climb, crouch, reach out, poke head out, retention, retrograde, and run. The recognition ccuracy of the proposed model on the ESC-Danger dataset is 95.06%, demonstrating higher recognition accuracy and good generalization performance compared to other state-of-the-art algorithms.关键词：dangerous behavior recognition;human-object interaction;two-steam graph convolutional network;skeleton sequence;escalator65|86|0更新时间：2026-01-04
- 摘要：Federated learning (FL), as a flexible and scalable distributed machine learning approach, has been widely applied in the industrial Internet of things (IIoT) to achieve low-latency, low communication overhead, and high-accuracy model training while preserving data privacy. However, due to the heterogeneity in computing and communication capabilities among edge devices in IIoT, traditional synchronous FL suffers from the "straggler effect", where the server must wait for all clients to upload their local model parameters, significantly reducing training efficiency and making it difficult to meet the low-latency service requirements of IIoT. To address this issue and mitigate the training delay caused by device heterogeneity, a semi-synchronous heterogeneous industrial FL framework was proposed. Based on this framework, a client selection scheme was designed, leveraging training latency efficiency scores to enhance training efficiency. Furthermore, to improve network spectrum utilization, an adaptive bandwidth allocation mechanism based on a mathematical relationship was proposed that ensured equal global training latency per round, optimizing the model upload strategy of selected clients. Extensive simulation results demonstrate that, compared with benchmark schemes such as FedAvg and FedCS, the proposed approach achieves significant advantages in model accuracy, system latency, and spectrum efficiency.关键词：industrial Internet of things;federated learning;straggler effect;client scheduling;resource allocation98|75|0更新时间：2026-01-04
Theory and Technology
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
- Total Visits：212817 Visits Today：640
0