最新刊期
卷 41 ， 期 11 ， 2025
- 摘要：Edge computing power, which delivers core capabilities such as computing and intelligence in proximity, has progressively evolved into critical infrastructure for vertical industries to achieve digital transformation and intelligentization. China has been placing high strategic emphasis on advancing technological innovation, fostering industrial development, and exploring practical applications in the field of edge computing power. A systematic analysis on the holistic development landscape of edge computing power was conducted. The conceptual definitions and distinctive characteristics of edge computing power were elucidated, its technical architecture was systematically organized, prototypical application scenarios were summarized, and developmental challenges alongside strategic recommendations were proposed. These research outcomes provides critical references for advancing technological innovation, infrastructure deployment, and application ecosystems in the domain of edge computing power.关键词：edge computing power;cloud-edge synergy;industrial Internet;co-construction and co-sharing5525|5399|0更新时间：2026-01-20
Review
- 摘要：In view of the pain points faced by the complex architecture and slow introduction of new services of the current real-time communication network , and the future evolution trend of intelligent real-time communication,an edge intelligent architecture of intelligent agile real-time communication network was proposed, including an unified control plane, a unified intelligent plane and an edge agile media plane, so that the intelligent communication network can evolve from “call entrance” to “application entrance” and “super entrance”. The intrinsic network artificial intelligence(AI) system was studied, and the basic strategy of edge intelligence and the communication architecture of a single AI agent were proposed. The key technologies of collaborative inference of edge-cloud for real-time communication were analyzed. The construction of distributed AI model management, collaborative inference mechanisms of end-edge-cloud and user’s knowledge maps were proposed.关键词：real-time communication;edge intelligent;AI agent;edge-cloud collaborative inference;multi-agent interaction5555|3986|0更新时间：2026-01-20
- 摘要：To achieve efficient management of 6G cell-free radio access network (RAN) slice resources, a layered collaborative optimization resource allocation algorithm based on reinforcement learning (RL) was proposed. This algorithm adopted a multi-timescale hierarchical collaborative architecture. At the upper layer with a large timescale, the resource matching degree of each slice was used as the optimization feedback metric, and the dual-depth Q-network (DDQN) algorithm was employed to dynamically adjust the resource configuration of slices. At the lower-level small time scale, a user access decision-making mechanism based on a proximity policy optimization (PPO) algorithm was established. Under the premise of meeting user quality of service （QoS）requirements, the algorithm minimized slice resource consumption through user collaboration cluster selection and resource allocation mechanisms. Simulation results demonstrate that the proposed algorithm significantly reduces resource consumption through user-centric dynamic access decisions and enhances resource matching through periodic reconfiguration of slice resources, thereby achieving efficient utilization of system resources.关键词：cell-free;network slicing;deep reinforcement learning;resource allocation;multi-timescale179|84|0更新时间：2026-01-20
- 摘要：Artificial intelligent (AI) agents based on large language models (LLM) possess fundamental capabilities such as thinking, reasoning, reflection, and planning. Moreover, these agents are equipped with robust self-learning and optimization abilities, enabling them to autonomously execute tasks and achieve efficient iteration. This significantly extends the application boundaries of large language models. Focusing on the architectural design and application exploration of AI agent platform for government and enterprise sectors, a novel architecture for an AI agent platform tailored to government and enterprise industries was proposed. An agent template layer was innovatively incorporated by the platform, which included AI-native applications and moduled for digital platform reconstruction, as well as a diverse range of industry-specific tools. Additionally, core advantages were featured such as private deployment, domestic adaptation, and multi-agent collaboration, achieving precise alignment with the needs of creating intelligent agents for specific industries. Case studies in the fields of government affairs, law enforcement, construction, and transportation demonstrate that the platform can significantly enhance the efficiency of AI agent construction and application, providing an effective solution for the digital and intelligent transformation of government and enterprise sectors.关键词：AI agent platform;government and enterprise industry application;AI-native application;digital platform reconstruction198|69|0更新时间：2026-01-20
- 摘要：With the widespread application of vision-language model (VLM) in tasks such as image-text retrieval, image captioning, and visual question answering, efficiently training large-scale models under the constraints of cross-industry data privacy and limited computational resources was recognized as a critical challenge for telecom operators. Federated learning (FL) was employed to address data silos and privacy concerns by enabling distributed collaborative training without sharing raw data. However, the massive number of parameters in VLM was found to lead to high communication and computation costs during training. Additionally, the strong data heterogeneity in FL environments was observed to limit the generalization capability of global models. To address these challenges, federated personalized low-rank mixture of experts (FedLRM), a personalized mixture-of-experts federated fine-tuning framework for VLM was proposed. It combined the parameter-efficient tuning method low-rank adaptation (LoRA) with a gating mechanism to build a local mixture of experts (MoE) architecture that fuses global and local features, enhancing fine-grained personalization. Experiments results show that in heterogeneous data scenarios, FedLRM improves accuracy by up to 1.88% compared to baseline methods, verifying that it provids an effective solution for the federated optimization of personalized vision-language models.关键词：vision-language model;low-rank adaptation;personalized federated learning;mixture-of-experts57|35|0更新时间：2026-01-20
- 摘要：With the widespread application of large language models (LLM) in natural language processing tasks, improving their performance in domain-specific question answering has become a key research focus. To address the limitations of traditional methods in complex multi-hop reasoning tasks, a knowledge-graph-enhanced multi-hop question answering approach, LLMKG, was proposed. Factual knowledge from knowledge graphs was injected into Prompt to enhance the reasoning capability of LLM in vertical domains. Comparative experiments conducted on the COKG-DATA dataset show that LLMKG outperforms the best baseline by 3.5% in terms of Hits@1. The method operats in a zero-shot setting, requires no additional parameter updates, and is applicable to various types of LLM. Temporal knowledge enhancement and multimodal knowledge fusion strategies were further explored, and a multi-modal knowledge graph (MMKG) was proposed as a future direction. This approach offers a novel and effective pathway for advancing intelligent question answering systems in specialized domains.关键词：large language model;knowledge graph;multi-hop question answering;multi-hop reasoning81|25|0更新时间：2026-01-20
- 摘要：In the field of cloud-network operation and maintenance, network stability and security are of utmost importance. Apart from software and hardware failures of equipment, 70% of cloud-network failures are caused by non-standard configurations. Therefore, it is particularly important to regularly audit the device configurations. However, the traditional auditing method of writing rules manually and checking the configuration text line by line is inefficient to meet the actual needs. For this purpose, a cloud-network configuration auditing system based on the reinforcement learning fine-tuning large language model was designed and developed. This model could automatically detect and correct non-standard behaviors in network configurations, thereby enhancing the stability and security of cloud-network operation and maintenance. The test results show that this model has achieved remarkable results in improving the auditing efficiency, reducing the occurrence rate of network failures, and cutting down the operation and maintenance costs. It provides an innovative solution for cloud-network configuration auditing and lays a foundation for subsequent research in model optimization, expanding application scenarios, and integrating with emerging network technologies.关键词：configuration audit;fine-tuning;reinforcement learning;large language model;prompt engineering46|18|0更新时间：2026-01-20
- 摘要：As 5G evolves toward 5G-Advanced, the intelligentization of the core network is recognized as a key path for operators to achieve differentiated experience assurance. Two implementation schemes were investigated to enable the NWDAF network function to collect wireless data: the wireless OMC-based scheme and the data sharing platform-based scheme. Experimental validation was conducted in a live network environment. The results demonstrated that both schemes can support NWDAF in performing wireless-side cell congestion prediction and user assurance functions.However, significant differences were observed in terms of data timeliness, operational complexity, and system scalability. Through comprehensive analysis, the solution that collaborates with the data sharing platform was recommended prioritize, to promote the efficient deployment of core network intelligence for operators.关键词：core network;NWDAF;intelligent control plane;wireless OMC;data-sharing platform56|93|0更新时间：2026-01-20
Topic: AI-Empowered Communication Networks
- 摘要：Space-air-ground integrated technology is one of the core enabling technologies for realizing the vision of ubiquitous connectivity in 6G networks. Efficient over-the-air (OTA) testing solutions are urgently needed for the verification of key technologies and the performance evaluation of devices. Firstly, the development status, network architecture, and channel characteristics of space-air-ground integrated systems were analyzed. Then, based on the specific characteristics and testing requirements of 6G space-air-ground integrated systems, a satellite-ground integrated OTA testing system was proposed to align with the channel environment characteristics of the communication devices in space-air-ground integrated systems, thereby enabling highly efficient OTA testing. Finally, the standardization and technical challenges that future OTA testing research for 6G space-air-ground integrated systems will encounter were summarized, and potential research directions were outlined, providing a theoretical foundation and guidance for future studies.关键词：6G;space-air-ground integrated system;OTA performance testing;MPAC;channel emulation78|67|0更新时间：2026-01-20
- 摘要：In large-scale satellite communication networks, the dynamic and uncertain nature of inter-satellite links significantly impacts network performance. Due to the high-speed movement of satellites, orbital variations, and disturbances from the space environment, traditional static routing algorithms struggle to adapt to frequent network topology changes caused by link establishment and disconnection. To address this issue, a link survival time prediction framework based on a Markov model was constructed, systematically analyzing the relationship between link availability, recovery rate, and failure rate, while also examining the impact of solar conjunction on link stability. Based on this framework, two optimization strategies were proposed: a reliable routing algorithm based on constrained shortest path first with link survival time (CSPF-LST), where link failure probability was introduced to dynamically adjust path selection, improving link survival time; a dynamic weighted shortest path algorithm based on link survival time, which integrated link survival time and latency factors for comprehensive optimization. Simulation results demonstrate that the proposed algorithms significantly improve routing reliability and data transmission efficiency in satellite networks, providing theoretical and technical support for the efficient operation of large-scale satellite constellation networks.关键词：satellite communication network;link survival time;reliable routing algorithm;Markov model34|16|0更新时间：2026-01-20
- 摘要：A multi-platform synthetic aperture positioning (SAP) method based on eigenvalue decomposition was proposed to address the mismatched resolution and low positioning accuracy in two dimensions of single-platform SAP method. A cost function for non-coherent accumulation positioning from multi-station data was established. The cost function was solved using eigenvalue decomposition to improve the local signal-to-noise ratio and positioning accuracy of the positioning image. To further enhance the computational efficiency of the multi-platform SAP method, particle swarm optimization (PSO) algorithm was introduced to accelerate the eigenvalue decomposition process, thereby improving the computational efficiency of the positioning method. Simulation results show that the root mean square error (RMSE) of the proposed positioning algorithm is lower than that of traditional two-step positioning methods and existing direct multi-platform positioning methods. After the introduction of the improved PSO algorithm, the positioning time was approximately reduced to 1/20 of that before the improvement, significantly enhancing computational efficiency.关键词：multi-platform;eigenvalue decomposition;incoherent accumulation;particle swarm optimization algorithm40|44|0更新时间：2026-01-20
- 摘要：In order to solve the contradiction between the limited spectrum resources of wireless networks and the exponential increase of Internet of things devices, an enhanced cognitive non-orthogonal multiple access (NOMA) communication network with mixed cellular transmission and D2D communication was proposed. Based on the premise of improving the coverage of cellular communication, the D2D node was authorized to use the spectrum resources of cellular network to complete the communication task. In view of the advantages of high spectral efficiency of NOMA technology, a dynamic cognitive NOMA transmission scheme was designed. Considering channel estimation error was existed, the optimal power allocation that minimized the system outage probability was found, while the outage probabilities and system throughput of cellular system and D2D communication system were analyzed, respectively. The simulation results verify the correctness of the theoretical derivation, and show that the proposed scheme demonstrates significantly improved reliability compared to existing schemes, achieving a throughput gain of up to 12.3% in the low-to-medium SNR region while maintaining comparable performance in the high SNR region.关键词：cognitive radio;non-orthogonal multiple access;D2D communication;power allocation;outage probability;channel estimation error33|14|0更新时间：2026-01-20
- 摘要：The future low earth orbit (LEO) satellite communication system requires a more flexible physical layer, thus giving rise to generalized frequency division multiplexing (GFDM) technology. The training symbols composed of repeated weighted sequences were studied, avoiding the crosstalk problem in time-frequency offset estimation, and a new time-frequency offset estimation algorithm was proposed. This algorithm utilizes the time-domain symmetric conjugation of the training code and its excellent autocorrelation characteristics in the time domain to achieve timed synchronization through piecewise moving correlation. The decimal frequency offset was estimated through the phase difference of two repeatedly trained symbols, and the integer frequency offset was estimated by taking advantage of the good autocorrelation property in the frequency domain and the low energy property at zero. The simulation results show that, compared with other algorithms, the frequency offset estimation error of this algorithm is reduced by two orders of magnitude, the time offset estimation error is reduced by three orders of magnitude, and the received bit error rate is reduced by seven orders of magnitude. Therefore, the time-frequency offset estimation and anti-frequency offset performance of this algorithm are better, which can effectively improve the overall performance of the GFDM system.关键词：low earth orbit satellite;generalized frequency division multiplexing;time-frequency offset estimation;anti-frequency offset performance34|21|0更新时间：2026-01-20
- 摘要：With the popularization of cloud computing and big data technology, the energy consumption problem of data centers has become increasingly prominent. How to achieve energy efficiency optimization through intelligent algorithms has become a key issue in the field of green computing. Traditional machine learning models are difficult to effectively capture the temporal dependencies and multidimensional coupling characteristics of energy consumption data in data centers, resulting in insufficient accuracy in energy efficiency optimization. To this end, an ensemble algorithm based on dilated convolution optimized bi-directional long short-term memory (DC-Bi-LSTM) network was proposed, which combined the bi-directional capture ability of recurrent neural networks with the error correction mechanism of ensemble learning to construct high-precision energy consumption prediction and energy efficiency optimization models. The experimental results show that compared to the current best prediction methods, the DC-Bi-LSTM network integrated algorithm reduces mean absolute error (MAE) by 0.22, mean absolute percentage error(MAPE) by 0.43%, and root mean squared error (RMSE) by 0.23. It can effectively overcome the interference of data noise and uncertainty in prediction and improve the effectiveness from data center energy efficiency prediction.关键词：data center;DC-Bi-LSTM;energy efficiency predict41|19|0更新时间：2026-01-20
- 摘要：To solve the problems of missed detection, false detection, and poor accuracy in infrared road pedestrian and vehicle target detection in traffic safety systems, the basic YOLOv8 was improved and an improved SDL-YOLO algorithm was proposed. Firstly, by introducing spatial depth transformation convolution SPD-Conv to enhance multi-scale feature capture, the problem of low image resolution and difficult object detection could be solved. Subsequently, a lightweight upsampling operator DySample was introduced to optimize complex scene feature maps through dynamic sampling strategies, effectively mitigating target boundary blurring and environmental occlusion issues. Finally, the detection head would be changed to a lightweight shared convolutional detection head LSCD to reduce model parameters and computational complexity, while ensuring lightweight while improving detection accuracy. At the same time, the algorithm was comprehensively evaluated using quantitative evaluation indicators accuracy P, recall R, average accuracy mAP, and FLOPS on the publicly available infrared night scene dataset. The experimental results show that compared with the original algorithm, SDL-YOLO has a 1.5% increase in accuracy P, a 1.5% increase in recall R, and an 1.5% and 1.7% increase in average accuracy mAP50 and mAP50-95, respectively. At the same time, it has lower model complexity FLOPS. The effectiveness of SDL-YOLO algorithm in improving detection accuracy, correcting false positives and false negatives has been further demonstrated through visual analysis of detection results.关键词：deep learning;object detection;infrared image;YOLOv8;pedestrian and vehicle45|21|0更新时间：2026-01-20
Research and Development
- 摘要：With the increasing demand for ubiquitous broadband communication and the continuous evolution of space-terrestrial integrated communication technology, satellite communication and terrestrial mobile communication networks have gradually shifted from independent development to deep integration. The key technologies of satellite-terrestrial spectrum sharing in the field of satellite-terrestrial integration were introduced. Firstly, the typical scenarios of the satellite-terrestrial spectrum sharing system and the compatibility analysis method were elaborated. Secondly, an intelligent cooperative access network architecture for satellite-terrestrial spectrum sharing was put forward. Based on this architecture, a beam-position-level satellite-terrestrial dynamic spectrum sharing scheme was proposed. References and insights are provided for the subsequent research on the spectrum sharing technology of the satellite-terrestrial integrated network in the industry.关键词：satellite-terrestrial integrated network;spectrum sharing;network architecture;interference management42|29|0更新时间：2026-01-20
- 摘要：In order to meet the needs of wide coverage, low delay and efficient resource utilization of service communication in new power system, a new power communication network with wide coverage architecture and a virtual network mapping optimization method based on this architecture were proposed. Firstly, the satellite Internet technology was integrated into electric power communication network architecture to form a new electric power communication network architecture, the coverage of the electric power communication network was expanded, and stable communication was achieved in extreme climates, natural disasters or remote areas. Secondly, based on this new architecture, a virtual network mapping model for power business was proposed. In this model, the resource utilization and delay parameters of the communication network were taken as the core optimization indexes. By improving the traditional simulated annealing algorithm, the dual optimization of low delay and high resource utilization in virtual network mapping was realized. Finally, the simulation results show that the proposed method can maximize the resource utilization while ensuring the low delay of the wide coverage communication network, thus improving the operating efficiency and stability of the power communication system.关键词：satellite Internet;wide coverage;virtual network mapping;electric power communication network;low delay;high resource utilization rate;improved simulated annealing algorithm22|22|0更新时间：2026-01-20
- 摘要：Nuclear power bases have the characteristics of wide areas, a large number of personnel, and dynamic changes in personnel distribution. Real time understanding of the distribution of personnel in the base is crucial for emergency management of the nuclear power base. To address this, an integrated personnel counting system solution was proposed. This solution involved the software and hardware from various personnel counting systems, including license plate recognition systems, facial recognition systems, and information release systems. The goal of achieving centralized collection, management, and visual display of personnel access or assembly data at the perimeter of the nuclear power base, the perimeter of internal areas, and emergency assembly points was accomplished. Based on real-time personnel distribution data, management could quickly formulate precise response strategies during emergencies, thereby significantly improving the emergency response capability of the nuclear power base.关键词：personnel dynamic headcount;emergency personnel statistics;digital technology;nuclear power base24|17|0更新时间：2026-01-20
Engineering and Application
- 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：778476 Visits Today：1966
0