最新刊期
卷 41 ， 期 4 ， 2025
- 摘要：As a critical computing infrastructure for large-scale model applications, the efficient operation of artificial intelligence data center relies on high-performance optical network transmission infrastructure. However, the optical networks interconnecting artificial intelligence data center face numerous challenges in meeting the demands of high real-time, high burstiness, and high reliability. Based on this, real-time resource allocation aims to overcome localized scheduling limitations in optical networks interconnecting artificial intelligence data center and reduce transmission and scheduling delays. Adaptive and collaborative optimization enables a continuous progression from passive adjustment to active collaboration in response to dynamic traffic bursts. Proactive failure recovery aims to achieve an orderly evolution from passive restoration to active intervention for reliability in optical networks interconnecting artificial intelligence data center. Future developments in large-scale real-time scheduling, deep computing-network convergence, and AI digital twins will drive new advancements in artificial intelligence data center interconnections.关键词：artificial intelligent data center;resource allocation;network optimization;failure recovery1029|1676|0更新时间：2025-10-31
- 摘要：Against the backdrop of the vigorous development of artificial intelligence (AI) large models, the demand for the data center optical module rate continues to rise. To meet the requirements of high rate, low power consumption, low latency, and high integration for ultra-high-speed short-reach optical interconnect, the development trend of the optical module industry was introduced, and relevant equalization and coding technologies were discussed. In terms of equalization technology, a low-complexity Volterra equalization method based on the least angle regression strategy (LaNLE) and a deep neural network equalization method based on hidden feature extraction (HFE) were proposed. Under the 120 Gbit/s PAM-8 signal transmission, the complexity of LaNLE was reduced by 70.1% compared with traditional methods at the same bit error rate; HFE improved the neural network training efficiency and achieved 288 Gbit/s PAM-8 signal transmission. In terms of coding technology, the optimization method of QC-LDPC codes in the optical interconnection system and the LDPC joint equalization and decoding method based on deep learning were introduced. Experimental results show that the proposed methods can effectively optimize the bit error rate and improve the system performance. At the same time, the future research directions were prospected.关键词：optical interconnect;signal processing;equalization;coding1034|986|0更新时间：2025-10-31
- 摘要：With the explosive growth in the scale of artificial intelligence data centers (AIDC), traditional AIDC networks characterized by “data center optical interconnection (DCI) + electrical packet switching (EPS)” are increasingly challenged in terms of power consumption, communication latency, and reliability. To address this issue, photonic technologies have been introduced in recent years to reduce the power consumption and enhance the scalability, flexibility, and reliability of AIDCs. Two types of network architectures—“DCI + EPS + optical circuit switching (OCS)” and “DCI + fast optical switching (FOS)”—that had been studied were reviewed. Combining the practices of leading enterprises and academic institutions, the technical pathways, performance advantages, and issues yet to be studied of these proposals were discussed. Insights were provided to guide the design of future large-scale AIDC networks.关键词：artificial intelligence data center;optical-electrical switching network;computing power cluster541|2213|0更新时间：2025-10-31
- 摘要：The rapid development of artificial intelligence has posed significant challenges to intelligent computing centers, especially in the collaborative scheduling of computational and networking resources. To address the deployment optimization issues of distributed intelligent computing services (Ring Allreduce), a novel technology called the computing power-wavelength plane (CWP) which enhanced the traditional waveplane framework to enable integrated virtual management of computational and network resources was firstly proposed. Based on the CWP, an efficient routing, wavelength, computing power, and time slot assignment (RWCTA) algorithm was proposed for Ring Allreduce service deployment. Simulation results demonstrate that, compared to conventional waveplane-based algorithms, the RWCTA algorithm based on the CWP effectively reduces the overall task completion time by 62.4% and the average task computation time by 54.5%.关键词：optical network of intelligent computing centers;integrated communication and computing;Ring Allreduce service;RWCTA;computing power-wavelength plane307|370|0更新时间：2025-10-31
- 摘要：With the rapid development of the global AI industry, the computational power demands of large-scale models continued to grow, prompting major technology companies worldwide to actively construct ultra-large-scale clusters exceeding 10 000 or even 100 000 GPU. Limited by natural resource supply, construction investment, and other constraints, the construction of a multi-cluster interconnected fundamental network through a high-speed all-optical network is an important potential solution for achieving efficient collaborative training across clusters. To meet the ultra-large bandwidth, ultra-low latency, and ultra-high reliability requirements of intelligent computing interconnection, a hitless intelligent computing optical transport network (HIC-OTN) and its key technological solutions were proposed. Based on HIC-OTN, the first field trial of 104 km cross-cluster pipeline parallelism (PP) training had been demonstrated, verifying the feasibility of 100 km-class cross-cluster PP training. Based on the 800 Gbit/s HIC-OTN interconnection, highly efficient collaborative training was achieved in two scenarios (52 km and 104 km clusters), delivering over 98% of the single-node training efficiency. Moreover, hitless and imperceptible optical network protection switching was demonstrated, ensuring zero impact on training performance.关键词：cross-cluster training;HIC-OTN;non-destructive protection;ultra-large bandwidth;ultra low latency614|813|0更新时间：2025-10-31
Topic: Optical Interconnection for Intelligent Computing
- 摘要：Swarm intelligence systems are recognized for their significant potential in advancing the intelligent development of social production and lifestyle, owing to their distributed architecture, high self-organization, and strong robustness. However, security and privacy protection issues were identified as critical factors limiting their stable operation and widespread application, directly impacting user trust and the large-scale deployment of the technology. Consequently, ensuring the operational security of swarm intelligence systems, achieving data privacy protection, and enhancing their anti-attack capabilities and robustness in complex environments were highlighted as urgent challenges to be addressed. In this context, their definitions, characteristics, general structures, and application scenarios of swarm intelligence systems were comprehensively described. The security objectives covering data, communication, system reliability, robustness, and trust management, as well as the privacy protection objectives related to data, identity, and intent, were explicitly proposed. Additionally, the main attack methods and related defense techniques were analyzed. Based on this, current mainstream solutions were systematically reviewed to address the security and privacy protection issues in swarm intelligence systems. Finally, the core challenges faced by swarm intelligence systems in this field were thoroughly discussed, and potential future development directions were explored, aiming to provide theoretical support and technical guidance for subsequent research.关键词：swarm intelligence;swarm intelligence system;security defense mechanism;privacy protection380|1455|0更新时间：2025-10-31
- 摘要：With the development of artificial intelligence (AI) large models progresses, the demand for computing resources in distributed training has increased significantly. This leads to a substantial increase in the amount of data communicated between clusters,which poses severe challenges to collective communication libraries in intelligent computing scenarios.. Focusing on the performance bottlenecks and business requirements of computing tasks in intelligent computing scenarios, the technical difficulties faced by current collective communication libraries were analyzed. At the same time, the future development trends of these libraries were also envisioned, such as developing more efficient communication algorithms, implementing more flexible scheduling mechanisms, and enhancing cross-platform compatibility, with the aim to provide valuable references and support for research and practical applications in the field of intelligent computing.关键词：AI;large model;intelligent computing scenario;collective communication library477|1511|0更新时间：2025-10-31
Review
- 摘要：In recent years, rapid advancements in communication technology have highlighted the shortcomings of the traditional transmission control protocol/Internet protocol (TCP/IP) network architecture. New network architecture protocols have been proposed in response to the evolving landscape, where multiple modal networks are expected to coexist for an extended period. Traditional network forwarding solutions struggle to adapt to the dynamic nature of polymorphic network, leading to issues such as high latency and costs. Based on this, a flow data-driven polymorphic network forwarding scheme and a reinforcement learning-based algorithm tailored for polymorphic network were proposed. Addressing the dynamic nature of polymorphic network, a polymorphic network forwarding scheme based on programmable data planes was proposed. This scheme comprised the in-band network telemetry (INT) collection module responsible for network information gathering, the network flow processing module for parsing and filtering network flows, the modal processing module for identifying and routing polymorphic network flows, the service function module for providing network functions, the data forwarding module for routing data, the cache module for data storage, and the node control module for generating and deploying forwarding strategies. Leveraging the protocol-independent nature of the programming protocol-independent packet processors (P4) programming language, this scheme facilitated the coexistence of multiple network modes and enhances forwarding efficiency through programmable switch hardware. Furthermore, considering the data update mechanism of programmable switches using tables, a reinforcement learning-based algorithm for polymorphic network forwarding was proposed. Adapting to the constraints of programmable hardware, this algorithm selects appropriate forwarding ports based on the state of polymorphic network flows to achieve efficient forwarding. Finally, the implementation and testing of the forwarding algorithm was conducted in China Environment for Network Innovation (CENI), validating the effectiveness of the algorithm.关键词：polymorphic network;network forwarding;programable data plane263|371|0更新时间：2025-10-31
- 摘要：With the rapid development of 6G communication networks and mobile edge computing(MEC) technology, the deployment density of base stations equipped with edge servers has been continuously increasing, and computational tasks show a growing trend toward diversification. To address the impact of differentiated quality of experience (QoE) for users and unbalanced resource allocation on system performance, an efficient matching offloading scheme was proposed. The scheme comprehensively considered both offloading decisions and computational resource allocation, and a mixed-integer nonlinear programming problem was formulated with the objective of maximizing system utility. By decomposing the original problem, an iterative optimization algorithm was designed based on bilateral matching theory for solution. Simulation experiments were conducted using public datasets from the central business district (CBD) of Melbourne, Australia. The results demonstrate that, compared with existing schemes, the proposed scheme shows significant advantages in improving system utility.关键词：MEC;QoE;resource allocation;task offloading235|430|0更新时间：2025-10-31
- 摘要：Direction of arrival (DOA) estimation is a critical area in array signal processing, with angle of arrival (AOA) being its core parameter. AOA estimation was taken as the main research objective to enhance accuracy in arbitrary structured arrays. By analyzing the system models of arbitrary structured arrays in the existing literature, a new subspace approximation iterative algorithm was proposed based on E-subspace theory, combined with the maximum likelihood criterion and Lloyd-like iterative algorithm. Furthermore, to reduce computational complexity and improve angle estimation accuracy, an iterative improvement algorithm based on the MUSIC peak range was proposed in conjunction with the MUSIC algorithm. Finally, simulation results demonstrate the effectiveness of the proposed methods in millimeter-wave channel estimation, showing significant performance improvements compared to the traditional MUSIC algorithm.关键词：AOA estimation;arbitrary structure array;E-subspace;Lloyd-like iterative algorithm171|390|0更新时间：2025-10-31
- 摘要：Designing efficient packet transmission over lossy links with a large bandwidth-delay product (i.e., long-fat) is important in the next-generation wireless networks, which are envisioned to be of a space-air-ground integrated scale. Transmission control protocol (TCP), the dominant transport layer protocol in the past decades, is known to have had performance issues in such links. A new transport approach was proposed using user datagram protocol (UDP) along with an application-layer forward erasure correction (FEC) code called streaming code to provide retransmission-free reliability. A novel approach based on reinforcement learning (RL) was proposed to jointly perform congestion control and FEC repair transmissions. Proof-of-concept simulations demonstrate that the approach can achieve high and smooth goodput in long-fat lossy links, which can be attractive for performance-enhancing proxy (PEP). It also highlights several future research directions that can be exploited based on the proposed approach.关键词：forward erasure correction;congestion control;reinforcement learning361|619|0更新时间：2025-10-31
- 摘要：In the Internet of vehicles, the emergence of computationally intensive and latency-sensitive vehicular applications poses great challenges for computing time and power consumption. Vehicle edge computing (VEC) is therefore proposed to take advantages of the computing capabilities of vehicles and roadside unit (RSU). However, the efficient resource management for VEC systems to achieve optimised performance remains unsolved. Based on this, the issue of collaborative resource allocation and task offloading in the VEC system was studied, with partial tasks being offloaded through vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) communication modes. At the same time, the selection decisions, task offloading ratios, and the allocation of computing resources among vehicles and RSUs were collaboratively optimized, with the aim of minimizing the total system energy consumption while ensuring task completion. As the proble was a NP-hard mixed-integer nonlinear programming problem, the optimization problem was decoupled into multiple subproblems, and subsequently solved by using ant colony system (ACS), Lagrange multipliers and gradient method. Extensive simulation results demonstrate that the proposed scheme significantly outperforms the benchmarking algorithms in terms of task incomplete rate and overall system energy consumption.关键词：VEC;energy consumption;computation offloading;ACS;incomplete rate363|556|0更新时间：2025-10-31
- 摘要：Intrusion detection can actively identify Internet of things (IoT) traffic attacks, which is an important measure to maintain IoT security. Therefore, multiscale residual temporal convolutional networks-based intrusion detection model (MRID) was proposed. In MRID, a multiscale residual temporal convolutional module was utilized to enhance the network capability in learning spatiotemporal representations. An improved traffic attention mechanism was introduced to estimate the importance score that helps the model to concentrate on important information during leaning. The proposed MRID was easily integrated into a fog-enabled IoT to offer efficient real-time intrusion detection. Finally, empirical evaluations on two recent datasets (CICIDS2017 and CSE-CIC-IDS2018) were conducted, demonstrating that MRID improved the efficiency of intrusion detection and increased the robustness of model while maintaining computational efficiency.关键词：IoT;intrusion detection model;temporal convolutional network;multiscale residual;attention mechanism242|825|0更新时间：2025-10-31
Research and Development
- 摘要：To enhance the operational management efficiency of the power communication network, artificial intelligence technology was utilized to develop a digitalization method for distribution panels within the communication power system. By implementing object detection and text recognition, real-time monitoring of the power supply status for each subordinate branch in the distribution panel was achieved. Firstly, a multi-layer nested recognition network (MLNRN) architecture was proposed, which incorporated lightweight strategies to reduce the computational demands on terminal devices, allowing for efficient and accurate structured output of the distribution panel's power supply status. Secondly, an improved YOLOv5 network was introduced for the task of icon detection. By integrating ConvNext Block and bidirectional feature pynamid network (Bi-FPN) structures, the recognition accuracy for small targets, such as status lights, was significantly enhanced. Finally, a text recognition model targeting the labels of subordinate branches in the distribution panel was constructed on the basis of the convolutional recurrent neural network-connectionist temporal classification (CRNN-CTC). Transfer learning and image augmentation strategies were employed to improve recognition accuracy for text with multiple character types in non-standard distribution panel images. Simulation experiments demonstrate that the average accuracy of image recognition is 97.2%, whereas the accuracy of text recognition is 92%. These results validate the effectiveness and applicability of the proposed architecture in the digitalization of power distribution panel, which provides an effective solution for video inspection and intelligent maintenance in power communication networks.关键词：power communication network;digitization of power distribution panel;image identification;character recognition;YOLOv5;lightweight;CRNN205|908|0更新时间：2025-10-31
- 摘要：Based on 5G and IP multimedia subsystem (IMS) network architecture, the network architecture of the next generation new calling was put forward by introducing data channel (DC) and artificial intelligence (AI). As the network architecture became more complex, the problems such as many media nodes, circuitous routing, and service delay were particularly prominent. In order to simplify the network and improve efficiency, it is necessary to define the network evolution ideas. Firstly, the future evolution goals of the new call media node was outlined, and then the key technologies in the process of network evolution was conducted in-depth analysis. It provides a reference to the evolution of the new calling media node.关键词：new calling;network evolution;data channel;AI302|518|0更新时间：2025-10-31
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