最新刊期
卷 41 ， 期 8 ， 2025
- 摘要：Opto-photonics information refers to the use of light (photons) instead of electricity (electrons) for communication, sensing, computing, and display, offering advantages such as low power consumption, high bandwidth, high speed, strong anti-interference capability, and high security. Based on this, the core components of opto-photonics information technology, including materials, devices, subsystems, and applications, were explored. Firstly, its fundamental concepts, historical development, and industry landscape were introduced. Then, the classification of opto-photonics information materials and their application requirements in devices were analyzed. Next, key devices such as emitters, optical sensors, display devices, and core chips were discussed, along with their integration and synergy. Finally, the integration and application of opto-photonics information subsystems were summarized, focusing on wireless optical communication, optical sensing, optical computing, and light-field display, while highlighting typical applications such as 6G communication, smart cities, and intelligent manufacturing. In the future, opto-photonics information technology is expected to play a crucial role in a broader range of fields.关键词：opto-photonics information;optical communication;optical sensing;optical computing;optical display3952|4231|0更新时间：2025-09-04
Review
- 摘要：As model parameters surpass the trillion-scale mark, intelligent computing interconnection faces technical challenges including ultra-large-scale networking, low-latency communication, and high-bandwidth synchronization multidimensional evaluation. The framework incorporating key metrics such as throughput, latency, and scaling ratio was established, the distinctive requirements of three major application scenarios: large-scale model training, AI inference, and edge computing, were analyzed. Through comparative analysis of solutions from leading technology enterprises, innovative practices were summarized including CLOS architecture and Fat-Tree topology, with discussions focused on critical technologies like interconnection protocols, network topologies, and congestion control. Future development directions such as open protocols and optoelectronic integration were also outlined. The findings demonstrate that continuous innovation in intelligent computing interconnection technologies will provide crucial infrastructure support for AI development.关键词：AI;intelligent computing interconnection;large-scale model training;network topology;optoelectronic integration3888|3897|0更新时间：2025-09-04
- 摘要：To address the issues of insufficient collaboration among computing resources and poor adaptability to task requirements in computing power networks, the computing power routing problem was modeled as a sequential decision problem. A deep reinforcement learning-based computing-aware routing algorithm was proposed for dynamic routing scheduling of computing network collaboration. The idea of hybrid expert models was drawn on and a differentiated expert network was designed based on an encoder-decoder structure for specialized optimization in three typical scenarios: delay-sensitive, ordinary, and computationally intensive. The routing selection space was constrained through an action masking mechanism to achieve efficient hop-by-hop decision-making and output a path containing the optimal computing node. The simulation experiment results show that compared with other routing scheduling algorithms, the proposed algorithm improves service success rate by about 17%, reduces end-to-end latency, optimizes load balancing between nodes, demonstrates good network topology adaptability, and can effectively meet the differentiated needs of diverse computing tasks.关键词：computing-aware routing;computing-network integration;multi-scenario optimization;sequential decision-making;deep reinforcement learning3592|471|0更新时间：2025-09-04
- 摘要：The intergenerational heterogeneity of computing supply and the demand for supply chain security have made the driving forces behind heterogeneous computing becoming an emerging trend in AI infrastructure. However, in heterogeneous hybrid training scenarios, the RoCEv2 (RDMA over converged Ethernet version 2) solution suffered from deficiencies in load balancing and congestion control, resulting in suboptimal parallel communication performance during model training. Meanwhile, existing high-performance homogeneous intelligent computing network solutions were faced with deployment barriers due to the heterogeneity of devices and closed-source CCL (collective communication library). To address these challenges, ICE (intelligent control Ethernet), a high-performance intelligent computing network solution for heterogeneous computing scenarios, was proposed. Based on the RoCEv2 protocol framework, ICE was designed to avoid deep customization of devices and CCL. Through a combination of heterogeneous communication library information collection, a centralized controller, and autonomous control at the end side, global optimal path planning and global active congestion control were achieved, significantly enhancing heterogeneous parallel communication performance. Experiments conducted in real-world physical environments demonstrate that ICE improves performance by up to 47%. Thus, ICE presents as a pioneering and easily deployable solution for constructing heterogeneous intelligent computing networks.关键词：heterogeneous computing power;intelligent computing network;RoCEv2;communication scheduling;congestion control;load balancing156|1081|0更新时间：2025-09-04
- 摘要：As intelligent computing centers become core infrastructure supporting the high-quality development of the digital economy, the training of hundred-billion-parameter large models imposes stringent requirements on network performance. Traditional monitoring methods struggle to address communication bottlenecks in ten-thousand-card clusters due to insufficient sampling precision and lack of fine-grained observation. A sub-millisecond-level network monitoring system (sMon) was proposed, which integrated intelligent counters and dynamic bandwidth analysis modules into the workflow processing pipeline to enable real-time tracking of NIC port queue depth and traffic fluctuations. By implementing dynamic bandwidth calculation through sliding window algorithms, sub-millisecond temporal accuracy was maintained while reducing system overhead to 0.8% CPU utilization via asynchronous log collection mechanisms. Testing on a 128-node A100 cluster, the system’s capability was demonstrated to capture sub-millisecond traffic details at network ports. Experimental results show a two-order-of-magnitude improvement in data granularity compared with conventional monitoring solutions. The proposed system provides real-time monitoring and performance assurance for constructing “ultra-large-scale, ultra-high-bandwidth, ultra-reliable” intelligent computing center networks through high-precision network state perception, effectively supporting the requirements of large-scale AI training tasks.关键词：intelligent computing centers network;network monitoring;sub-millisecond-level158|300|0更新时间：2025-09-04
Intelligent Computing Network
- 摘要：Aiming at the problem that a single reflected link model cannot accurately measure the environment of extremely large-scale reconfigurable intelligent suface (XL-RIS) system, a communication model of superimposed XL-RIS near-field region and BS far-field region was constructed. On this basis, an efficient beam training scheme was proposed to reduce the training cost. Firstly, considering the path gain of near-field spherical beam and far-field planar beam, the influence coefficient F was introduced to derive the channel model suitable for XL-RIS near-field and BS far-field superposition region. In addition, in order to improve the received signal power and make the direct beam and the reflected beam in phase superposition, the phase correction parameter was introduced, and the training codebook matching the hybrid near-far field channel was constructed. Finally, a spatial layering scheme with variable step size was designed for the model. Specifically, the sampling interval of each layer increases from the origin along the radius. The simulation results show that the dual link hybrid near-far field model can achieve 93.7% rate performance under perfect channel condition when the SNR is 0, and the rate performance is improved by 57.6% and 205.4% compared with the near-field reflection model and the far-field reflection model respectively. The average reachable rate error of the new spatial layering scheme is less than 1% compared with the traditional layering scheme, but the training cost is reduced by 63.6%.关键词：XL-RIS;hybrid near-far field;beam training;hierarchical space185|305|0更新时间：2025-09-04
- 摘要：Container technology, as a lightweight virtualization solution, remains a core supporting technology for multi-cloud network architectures. However, the high cost of cross-cloud communication remains a critical challenge. The key innovation of models of the resource scheduling problem in multi-cloud networks as a quadratic assignment problem (QAP) and an improved discrete marine predators algorithm (DMPA) was proposed. The DMPA was significantly enhanced performance through the following strategies: a hybrid population initialization strategy based on uniform distribution and pseudo-reverse learning, which avoided the initial solution from falling into local optima; an adaptive convergence factor was introduced to dynamically adjust the search range, balancing global exploration and local exploitation; a 2-exchange mutation strategy with variable exchange intervals was employed to enhance population diversity; the integration of tabu search to optimize elite solutions and avoid repetitive searches. Experiments show that DMPA achieves known optimal solutions in 22 out of 30 QAP instances, with an average deviation rate of less than 3%. Under real-world multi-cloud network datasets, the optimization effect is significant, and communication costs are reduced. Compared to algorithms such as MPA and HPSO, DMPA demonstrates outstanding performance in both optimization accuracy and stability, providing an efficient solution for cost optimization in multi-cloud networks.关键词：multi-cloud network;cloud native;secondary distribution;swarm intelligence optimization algorithm88|168|0更新时间：2025-09-04
- 摘要：Satellite communication, as a key component of the global communication network, is irreplaceable in providing coverage in remote areas and for emergency communications. However, the development of emerging applications such as direct satellite connections for mobile phones has raised higher requirements for system capacity, urgently necessitating solutions to technical challenges such as scalable beam development, increased user numbers within beams, and improved user rates. To address this, 3GPP had initiated research on coverage enhancement and capacity enhancement technologies starting from the Rel-18 version. Based on the uplink capacity enhancement scheme using orthogonal covering codes (OCC) that is currently under research by 3GPP, the technical pathway was systematically analyzed and its performance was evaluated through simulations. Additionally, the standardization direction for downlink capacity enhancement was explored, providing technical reference for the subsequent evolution of 5G non-terrestrial network (NTN). Furthermore, a non-orthogonal capacity enhancement strategy based on pattern division multiple access (PDMA) was proposed, breaking through the traditional orthogonal multiple access limitations from the perspectives of multi-user access efficiency and spectrum resource utilization, and laying a theoretical and technical foundation for meeting the ultra-large capacity demands of 6G NTN. The research results indicate that through the optimization of resource utilization, the capacity of satellite communication systems can be significantly enhanced, effectively supporting the large-scale deployment of NTN in the future.关键词：satellite communication;capacity enhancement;OCC;PDMA138|781|0更新时间：2025-09-04
- 摘要：In response to the significant estimation errors in hop period and hop frequency under low signal-to-noise ratio (SNR) conditions, a method for hop period and hop frequency estimation based on maximum entropy binarization of time-frequency maps and a detection and localization (DL)-YOLOv5s model was proposed. Firstly, the maximum entropy thresholding method combined with morphological filtering was utilized to process the time-frequency map, resulting in a clear maximum entropy binarized time-frequency map. Then, the proposed DL-YOLOv5s model was appllied to detect and localize the hop frequency signals within the maximum entropy binarized time-frequency map. By incorporating the ASPP module and BiFPN module, the precision of edge and corner detection for hop frequency signals was enhanced. Additionally, a multi-head self-attention mechanism was introduced to improve the localization accuracy of hop frequency signals by the BOT3 module. Ultimately, the coordinates of the hop frequency signals were obtained, and through the correlation of these coordinates, the estimation of hop period and frequency was completed. The experimental results show that compared to the YOLOv5s model, the proposed DL-YOLOv5s model improves precision (P) by 5%, recall (R) by 2.2%, and the mean average precision (mAP) at 0.5 and mAP at 0.5:0.9 by 5.1% and 4.2% respectively. In comparison to other models such as YOLOv7 and YOLOv8, the proposed DL-YOLOv5s model is smaller in size, making it more suitable for resource-constrained environments like embedded devices commonly used for frequency-hopping signal parameter estimation. Additionally, compared to traditional methods for frequency-hopping signal parameter estimation, the proposed method effectively reduces the estimation errors of hopping period and hopping frequency under low signal-to-noise ratio conditions.关键词：frequency-hopping signal;low signal-to-noise ratio;parameter estimation;YOLOv5s81|182|0更新时间：2025-09-04
- 摘要：Node classification methods of complex network are mostly realized based on node representation learned by the graph neural network, the graph neural network encodes local structure information of complex networks through neighborhood aggregation. However, the over-smoothing problem of the graph neural network limits the node classification performance of complex network. In view of this problem, a node classification method of complex networks based on enhanced graph neural networks and contrastive learning was proposed. In the proposed method, not only the attention was introduced to the neighborhood nodes, in order to differentiate the importance of each neighbor node, but also the feature of each edge was constructed with combination of the local neighborhood overlap and the global neighborhood overlap, so as to expand the information of the node representation. Finally, contrastive learning was introduced to train the neural networks, so that the network’s global node priori information was utilized to jointly optimize the node representation. Experiments were performed on Cora, Citeseer, PubMed and Chameleon public network datasets. The results demonstrate that compared to the other advanced methods, the proposed method achieves better node classification performance, moreover, the effectiveness of the proposed method is verified through ablation study.关键词：network node classification;complex network;graph neural network;graph attention network;contrastive learning88|432|0更新时间：2025-09-04
- 摘要：The research of integrated simulation technologies for satellite communication networks was focused on, aiming to support the construction of a universal, open, and high reliable satellite communication network simulation system. In response to the multi-level simulation requirements of satellite communication networks, a software bus-based simulation integration architecture was proposed to enable flexible interactive integration of hierarchical granularity models and software components. Addressing the characteristics of satellite communication networks and their modeling/simulation tools, a comprehensive integration approach was designed combing data, models, and software to facilitate the development of loosely-coupled satellite communication network simulation systems. Considering the system equipment features of satellite communication networks, a vitual-real integrated simulation methodology was developed from three dimensions: physical link layer, network protocol layer, and operational control layer, providing technical support for virtual-real collaborative simulation verification. Finally, practical application was demonstrated through specific implementation projects.关键词：satellite communication network;multi-level simulation;simulation integration;virtuality and reality combination69|158|0更新时间：2025-09-04
- 摘要：Based on the theoretical framework of deep reinforcement learning, a hierarchical and progressive solution was proposed. Firstly, a heterogeneous data transmission architecture integrating edge computing nodes was constructed, and a multi-dimensional state space Markov decision process with time-varying characteristics was established. Secondly, the entropy regularization constraint term was embedded in the traditional deep Q-learning network (DQN) algorithm, and the experience replay mechanism of the same strategy was combined. An enhanced ESERDQN (improved DQN algorithm based on entropy and same-strategy experience replay) optimizer was formed. Finally, a five-dimensional evaluation index system (convergence rate, cumulative reward value, energy consumption, end-to-end delay, transmission cost) was designed to carry out multi-algorithm comparison experiments. The simulation results show that ESERDQN achieves stable convergence within 1 500 training cycles, which improves the convergence speed by 49.2%, 41.7%, 30.1% and 13.3% respectively compared with the benchmark greedy algorithm, random algorithm, DDPG algorithm and PPO. In terms of comprehensive business indicators, the unit energy cost was reduced by 27.8%, and the delay of key tasks is controlled within 12.3 ms, which verifies the technical superiority of the proposed method in complex transmission scenarios of smart cities.关键词：smart city;data transmission;entropy;same-strategy experience replay;deep reinforcement learning90|501|0更新时间：2025-09-04
- 摘要：Deep learning models rely on a large number of training samples in the modulation recognition task. However, in actual scenarios, the signal samples are limited, especially in complex noise environments, where the model performance is restricted. Therefore, a lightweight modulation recognition method based on local feature guidance was proposed. Firstly, a lightweight teacher network was constructed to extract local features from noisy modulated signals, and a local semantic feature optimization algorithm was designed to distill local knowledge into the student network. Secondly, aiming at the complex-domain characteristics of the modulated signal spectrum, a complex-domain Transformer was designed as the student network for global feature extraction, ultimately completing the recognition task. Experimental results show that the proposed model demonstrates higher recognition efficiency in small-sample scenarios compared with other deep learning models, and exhibits significant advantages in terms of computational complexity and real-time performance compared with existing methods.关键词：modulation recognition;knowledge distillation;small-sample dataset;Transformer;lightweight network115|446|0更新时间：2025-09-04
- 摘要：The intrusion detection system (IDS) as a core component of IoT security defense, was directly impacted in its performance, which in turn affected the overall security of the network. However, the imbalanced distribution of class samples in intrusion detection datasets is found to reduce the detection performance of IDS for minority class samples. To address this issue, a dynamical class-weighted-based convolutional neural network intrusion detection (DCID) model was proposed. The DCID model utilized a one-dimensional convolutional neural network (1-D CNN) structure and introduced a dynamical class-weighted loss function, enabling the DCID model to not only maintain high detection performance for majority class samples but also significantly enhance the detection capability for minority class samples. To validate the effectiveness of the DCID model, experiments were conducted using the CICIDS 2017 dataset. The experimental results demonstrate that, compared to typical machine learning models, the DCID model exhibites significant advantages in terms of precision, recall, and F1-score. Additionally, the detection performance of the DCID model under different loss functions was compared, and the results indicated that the dynamical class-weighted loss function effectively improved the detection performance for minority class samples.关键词：intrusion detection system;imbalanced class distribution;convolution neural network;loss function;class-weighted79|106|0更新时间：2025-09-04
Research and Development
-
Research on the application of post-quantum cryptographic algorithms in the telecommunications field
摘要：As a vital strategic resource, cryptography plays a core role in the field of network and information security. In recent years, the rapid development of quantum computing has effectively promoted the research and application of post-quantum cryptography (PQC) algorithms in academia and industry. Based on this, the security threats posed by quantum computing to classical cryptographic algorithms were studied, the current development status of PQC was surveyed, the quantum security impacts on typical scenarios in the telecommunications field were analyzed, and finally, migration verification and testing analysis were conducted for the PQC signature algorithm (module-lattice-based digital signature standard) ML-DSA. The experimental results indicate that the ML-DSA algorithm can meet the requirements of high concurrency and high availability of the system.关键词：post-quantum cryptography;quantum security;network infrastructure;ML-DSA172|1368|0更新时间：2025-09-04 - 摘要：The edge computing fusion network sinks the computing resources to the user side and completes the computing tasks locally through the coordination of distributed edge computing nodes, which significantly reduces the cloud burden and transmission delay. However, with the increase of user access density and the complexity of scenarios, how to dynamically optimize network resources to cope with diversified service demands and large-scale data processing tasks has become a major challenge. Therefore, a resource management method for edge computing networks based on double deep Q network (DDQN) was proposed. Integrating the virtual network embedding (VNE) technology, the proposed method formulated a multi-constraint optimization model to maximize the long-term embedding revenue-to-cost ratio. By leveraging the online learning capabilities of the DDQN framework, it enabled dynamic decision-making through interaction and feedback with the environment. Simulation results demonstrate that the proposed method achieves average improvements of 13.3%, 25.7% and 8.5% of virtual network request (VNR) acceptance rate, long-term embedding revenue, and long-term revenue-to-cost ratio, respectively, compared with the existing methods.关键词：edge computing fusion network;computing shifting;DDQN;resource management96|358|0更新时间：2025-09-04
- 摘要：In response to the deficiencies in the operation and maintenance of China’s power grid regarding the replacement of expired smart meters and batch sampling rotation, a rotation strategy for estimating the operation error of smart meters based on data collected by the metering automation system was proposed. Firstly, a smart meter operation error estimation model was established on a per-transformer-area basis. Secondly, a regularization method was adopted to solve the ill-conditioned situation of the estimation model. Finally, based on rolling calculations over multiple time windows and a comprehensive evaluation mechanism, smart meters for rotation were selected for field verification. The effectiveness of the proposed strategy was verified through practical examples in the power grid.关键词：smart meter;rotation strategy;error estimation;regularization107|528|0更新时间：2025-09-04
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：1035510 Visits Today：3559
0