最新刊期
卷 41 ， 期 10 ， 2025
- 摘要：With the development of generative AI technology, AI agents driven by large language model (LLM) start to emerge, and the communication of AI agents becomes one of the most important directions for the future network. It needs to build a new infrastructure based on the current network, generate the new access and authentication technologies, new networking and addressing technologies, new data transmission and control technologies. It will bring the technological innovation across protocol layers in the network, and even trigger a new paradigm change in communication. Based on the application scenario of agent communication, the impact of agents communication on wide area IP network was analyzed, the networking, architecture and key technologies of Internet of agents were designed, to support the implementation of end-to-end agent communication.关键词：Internet of agents;networking and architecture;wide area interconnection5130|5368|0更新时间：2025-11-06
Review
- 摘要：Underwater acoustic (UWA) communication is currently the only medium and long-range underwater wireless data transmission technology. The inherent physical characteristics of UWA channels, such as narrow bandwidth, high delay extension, harsh multipath, high dynamism, and loud noise, make UWA channels one of the complex channels. The key technologies of the physical layer in UWA communication include channel modeling, modulation techniques, channel estimation, and channel equalization. With the development of the data-driven approach, data-driven enhanced deep learning algorithms have achieved excellent results in UWA scenarios. An overview of the research progress in key technologies of the physical layer for UWA communication was provided. The data-driven receiver methods were summarized. Finally, the main domestic and foreign research directions of underwater acoustic communication were concluded, and the prospects of UWA communication technology were discussed.关键词：underwater acoustic communication;channel estimation;channel equalization;deep learning4688|5908|0更新时间：2025-11-06
- 摘要：With the expanding application of unmanned underwater vehicle (UUV) in underwater operations and target detection, underwater acoustic communication networking technologies designed to support UUV clusters have been widely investigated as a key field in underwater information acquisition. MAC protocols, which serve as core technologies for coordinating data transmission among nodes and optimizing network performance, play an important role in improving the efficiency of underwater acoustic communication network. A systematic review of the research progress on MAC protocols for acoustic networks supporting UUV clusters was conducted, covering developments from classical protocol architectures to recent cluster-oriented adaptation schemes. Existing MAC protocols were analyzed in depth based on two dimensions: channel access mechanisms and application scenarios. The technical characteristics and applicable conditions of various methods were summarized, and future development trends of UUV cluster-oriented underwater acoustic network MAC protocols were prospected. This review aims to provide references and insights for the innovative design and engineering application of UUV cluster underwater acoustic network MAC protocols.关键词：UUV;MAC;access mechanism;application scenarios4155|3967|0更新时间：2025-11-06
- 摘要：In the context of in-band full-duplex (IBFD) underwater acoustic communication systems, self-interference (SI) suppression is challenging due to the limited underwater computational resources. Traditional sparse channel estimation algorithms struggle to cope with time-varying, multipath, and noise-dense channel characteristics, failing to balance convergence speed and estimation accuracy. To address this, an error adaptively compensated shrinkage affine projection algorithm (EA-CS-APA) for underwater SI suppression was proposed. In this method, an error energy-based selective update mechanism was introduced to suppress ineffective parameter disturbances and a nonlinear mapping between error and step size was constructed to achieve adaptive step size adjustment, effectively balancing convergence speed and steady-state accuracy. Experimental results demonstrate that, compared with the compensated shrinkage affine projection algorithm (CS-APA), the proposed method achieves approximately 20%, 10%, and 40% improvements in normalized mean square deviation, SI suppression performance, and computational efficiency, respectively. It exhibits more robust performance advantages in complex time-varying multipath environments, providing an effective SI suppression solution for underwater communication devices with limited computational resources.关键词：underwater acoustic communication;self-interference suppression;affine projection algorithm;adaptive filtering;sparse channel estimation3335|344|0更新时间：2025-11-06
- 摘要：Underwater wireless optical communication(UWOC) has been widely used for high-speed and high-capacity underwater data transmission because of the lowest link delay, the highest transmission rate and the lowest implementation cost. At high transmission rates, even a short period of blockage or occlusion can lead to sudden communication interruption, which results in the loss of a large amount of transmitted information. To address the issue of communication link blockage and improve system transmission stability, a RIS-assisted UWOC system was proposed. The specific research process was as follows: (1) the cascaded turbulent channel fading coefficients from source to destination were modeled as Gamma-Gamma distributions, and the pointing errors caused by beam jitter and RIS jitter were considered. (2) The UWOC system employed intensity modulation with direct detection, and the transmitting light source could be switched to the optimal one for communication based on feedback from the receiver. (3) The probability density function of the instantaneous received signal-to-noise ratio(SNR) of the RIS-assisted system was derived. Based on the probability density function, a closed-form expression for the interruption probability was provided. (4) Closed-form expressions for the finite SNR diversity order and the asymptotic diversity order at high SNR were derived. (5) Monte Carlo simulations were conducted to verify the correctness of the derived results. Theory analysis and simulation experimects show that the transmission stability of the system can be significantly improved by increasing the number of emitting light sources. Moreover, when the product of the number of transmitting light sources and RIS blocks was constant, the system achieved the same asymptotic diversity order.关键词：UWOC;RIS;ocean turbulence;optimal selection104|225|0更新时间：2025-11-06
- 摘要：Backscattered photons in bidirectional underwater wireless optical communication (UWOC) systems can interfere with optical signals, reduce the signal-to-noise ratio (SNR) at the receiver, and even cause optical saturation in the detector. Therefore, in-depth research on the power distribution and arrival angle distribution of backscattered photons holds significant importance for bidirectional UWOC systems. An improved Monte Carlo (MC) simulation method was proposed. Instead of directly enlarging the detector’s target area, photon characteristics within a larger annular region were statistically analyzed to approximate those within the detector’s target area. This approach was capable of reducing the simulation time by more than 58%. The power distribution and photon arrival angle distribution of backscattered photons for different water types and light source divergence angles were comprehensively analyzed. An experimental platform was also established for validation. Simulation results demonstrate that when the light source divergence angle is below 60°, there is no significant difference in the backscattered light power distribution or photon arrival angle distribution. Experimental results indicate that as magnesium hydroxide concentration increases, the backscattered light power becomes relatively higher. When the light source divergence angles are 3° and 48°, their backscattered light power distribution curves nearly overlap, aligning with simulation results. Through simulations and experiments, this study confirms that water types are the primary factor influencing backscattered light distribution, while the light source divergence angle has limited impact on system backscattering characteristics. This finding provides valuable references for the design of receivers in bidirectional UWOC systems.关键词：bidirectional UWOC;MC method;backscattering;optical power distribution;photon arrival angle distribution68|104|0更新时间：2025-11-06
- 摘要：According to the current limitations of underwater information transmission network in meeting operational requirements, domestic and international development trends in underwater information network were analyzed, a water-air domain integrated information transmission network was designed which had the advantages of cable/wireless, fixed/maneuver, water-air cross-domain. Key technologies including underwater grid network, submarine primary base station, cross-domain water-air communication gateways, and multi-source underwater information fusion were developed to construct an underwater information superhighway, the deep integration of underwater trunk cable, wireless fixed and mobile networks was realized, the bearer network for a variety of services was provided, such as underwater communication, detection, early warning, observation and navigation, rapid perception and dynamic response capabilities for multi-source underwater information were enhanced, to provide technological support for China’s marine scientific research, marine resource development and utilization, and national defense security.关键词：water-air cross-domain;integrated information transmission;information fusion179|114|0更新时间：2025-11-06
-
Distributed reinforcement learning-based AUV 3D underwater current target tracking control algorithm
摘要：To address the challenges of high dimensionality, dynamic disturbances, and sparse rewards in autonomous underwater vehicle (AUV) target tracking within complex three-dimensional ocean current environments, a distributed reinforcement learning-based control algorithm was proposed. Firstly, realistic 3D ocean current data was incorporated to design dynamic target tracking scenarios, accurately modeling the AUV’s motion dynamics. Secondly, a distributed reinforcement learning framework was constructed by integrating the dueling deep Q-network (Dueling DQN) architecture with quantile regression methods. This framework quantified the uncertainty of Q-values to mitigate overestimation risks in 3D turbulent environments, enhancing the policy’s adaptability to dynamic disturbances. Finally, the prioritized experience replay mechanism was introduced, along with a constraint-optimized reward function, to optimize data sampling strategies and accelerate model convergence. Experimental results demonstrate superior performance of the proposed algorithm compared with the baseline methods, such as deep Q-network (DQN), double deep Q-network (DDQN), and Dueling DQN, in complex current conditions, achieving significant improvements in convergence speed, target tracking accuracy, and robustness.关键词：reinforcement learning;AUV;target tracking;quantile regression143|533|0更新时间：2025-11-06 - 摘要：Mobile edge computing is considered as an important solution to reduce backhaul pressure and improve quality of service, yet existing resource management strategies are poorly adapted in highly dynamic ocean environments. To address this problem, a task offloading and resource allocation algorithm based on an improved twin-delayed deep deterministic policy gradient was proposed. The algorithm was designed to systematically coordinate servo UAV deployment with edge node resources to jointly optimize communication resource allocation and computational task scheduling, while taking into account the energy constraints of ocean edge nodes and the time-varying characteristics of ocean networks. Specifically, the problem was formulated as a non-convex optimization framework with the objective of maximizing throughput under stringent quality of service requirements of user devices. The proposed algorithm dynamically adapted to the changing ocean environment through resource coordination, effectively balancing delay and energy consumption. Simulation results show that the proposed algorithm significantly outperforms existing benchmark methods in highly dynamic maritime communication scenarios, demonstrating the effectiveness and feasibility of the approach.关键词：mobile edge computing;resource allocation;task offloading;maritime communication;twin-delayed deep deterministic policy gradient106|482|0更新时间：2025-11-06
Topic: Technologies for Underwater Communication Networks
- 摘要：In a multipath transmission environment, a two-dimensional direction of arrival (DOA) estimation method for dynamic metasurface antenna based on beam-space time smoothing algorithm was proposed, addressing the issue of decreased estimation accuracy in DOA caused by aperture loss in existing spatial smoothing algorithms. Firstly, by leveraging the agility and phase-shifting characteristics of the L-shaped dynamic metasurface antenna (DMA), multipath signals were observed comprehensively across multiple dimensions, including beamspace, spatial domain, and temporal domain. As a result, signal components of all dimensions were obtained without aperture loss. Next, eigenvalue decomposition was performed on the beamspace data for each coherent time period to extract the signal subspace data and calculate the cross-covariance matrix across multiple periods. Subsequently, the cross-correlation matrices of these cross-covariance matrices were averaged to form the modified beamspace temporal smoothing matrix. Then, the 1D beamspace multiple signal classification (MUSIC) algorithm was employed to separately estimate the incident angles in the horizontal and vertical directions. Finally, the maximum likelihood algorithm is applied for pairing to obtain the 2D DOA. Monte Carlo simulations validated the theoretical analysis results. Compared to existing DOA estimation methods based on smoothing algorithms, this method offers stronger decorrelation and smaller spectral peak fluctuations, effectively enhancing estimation accuracy.关键词：beamspace;temporal smoothing;DMA68|222|0更新时间：2025-11-06
- 摘要：With the rapid development of Internet technology, the task of intrusion detection of the field of network security has become more important. Aiming at the problems of high feature dimension, imbalance of data categories and low model detection rate of single classifiers in current intrusion detection, a dynamic integrated intrusion detection model based on Powershap and hybrid sampling was proposed. Firstly, the Powershap algorithm was used for feature selection of the dataset. Subsequently, the hybrid RENN-BorderlineSMOTE sampling algorithm was applied to address the category imbalance in the dataset by under-sampling and over-sampling specific categories of data. Finally, the optimal combination was filtered from multiple base classifiers based on Generalization Diversity and integrated into the dynamic integration framework KNORAE to combine the advantages of multiple base classifiers. The model was validated on the CIC-IDS2017 dataset, which confirmed the superiority of the model in network traffic detection.关键词：intrusion detection;class imbalanced;ensemble learning;mixed sampling39|161|0更新时间：2025-11-06
- 摘要：With the widespread application of Internet of things technology, non-orthogonal multiple access (NOMA) has been recognized as an efficient multiple access technique, offering significant advantages in large-scale connectivity and spectral efficiency. However, impulsive noise, a common source of interference, has been found to severely degrade the performance of NOMA systems. To address this issue, a deep learning-based compressed sensing method was proposed for impulsive noise suppression. By exploiting the temporal sparsity of impulsive noise in NOMA systems, a data-driven approach was employed, with the network being used to estimate the impulsive noise. Firstly, a linear mapping network was used to replace the compressed sensing algorithm in computing the pseudoinverse, providing an initial estimate of the impulsive noise. This estimate was then fed into a compressed sensing reconstruction WaveNet (CSR-WaveNet), which could learn the sparse characteristics of the impulsive noise for more accurate noise estimation. Simulation results show that, compared to existing methods, the proposed approach can achieve lower bit error rates.关键词：deep learning;compressed sensing;impulsive noise;non-orthogonal multiple access64|84|0更新时间：2025-11-06
- 摘要：To address the bottlenecks of resource scheduling and data transmission in traditional slicing and mapping methods for deep neural networks (DNN) , an efficient dynamic slicing and intelligent mapping optimization algorithm was proposed based on a network on chip (NoC) accelerator. The algorithm was designed to flexibly divide DNN computing tasks through dynamic slicing and optimize task and data flow management in the NoC architecture. Experimental results show that the proposed algorithm significantly outperforms traditional methods in computing throughput, NoC transmission delay, external memory accesses, and energy efficiency, especially for complex models.关键词：NoC accelerator;DNN slicing;smart mapping58|159|0更新时间：2025-11-06
-
A large language model-enhanced few-shot learning algorithm for network complaint intent recognition
摘要：Addressing the escalating challenge of network complaint handling in the 5G era, driven by increasingly complex network architectures and growing user bases, the aim is to solve the prevalent issue of few-shot intent recognition for complaint categories with scarce samples. Traditional small-parameter natural language processing (NLP) models exhibit insufficient generalization and suboptimal performance in such scenarios due to their heavy reliance on data quantity and quality. While large language models (LLM) possess strong general semantic understanding, their limited domain-specific knowledge in telecommunications and substantial computational demands hinder direct, efficient application to few-shot network complaint intent recognition. To precisely tackle this, an LLM-enhanced few-shot intent recognition algorithm for network complaints was proposed. This innovative algorithm integrated the powerful text generation capabilities of LLM with the efficient inference characteristics of small models. It leveraged LLM to iteratively generate high-quality simulated training samples for few-shot categories, incorporating a small model for quality evaluation and feedback to continuously optimize the sample generation process. This mechanism significantly boosted small model training and specifically enhanced the recognition accuracy of few-shot complaint intents. Test results on actual network complaint data demonstrated a 21% improvement in intent recognition accuracy for 3 types of few-shot network complaints, and a 9% overall improvement in recognition accuracy across all 8 complaint types.关键词：large language model;few-shot learning;network complaint intent recognition65|149|0更新时间：2025-11-06 - 摘要：Session-based recommendation predicts the next interaction item by analyzing anonymous users’ historical interaction data. Because of the sparsity of user behavior data, modeling session representations from a single perspective may fail to fully capture users’ true intentions. Moreover, existing session-based recommendation models primarily focus on the sequential order of session items while overlooking the higher-order relationships between them. To address this, the cross-view contrastive learning and multi-channel hypergraph convolution （CLMHC）model that constructs two complementary views was proposed: a hypergraph and a global hypergraph. By leveraging a channel-mixing attention mechanism, the model adaptively integrated the user intentions captured from both hypergraph channels while employing contrastive learning to refine user representations across different views. Extensive experiments on three datasets demonstrate that our model significantly outperforms existing approaches in recommendation accuracy. The ablation studies further confirm that both views contribute positively to performance improvement.关键词：session-based recommendation;hypergraph;multi-channel;attention mechanism;contrastive learning95|114|0更新时间：2025-11-06
Research and Development
- 摘要：Indirect network sharing is considered as an innovative network sharing method, which has a broad application prospect and important social value, when achieves the problems in 5G-Advanced (5G-A) network sharing development and promotes the further technology evolution. The current status of the development of network sharing technical standards was introduced, the challenges faced by network sharing operators in the 5G-A phase were analyzed, and the design principles such as new network sharing methods, solutions, and architectural functions were proposed. The features between it and existing technologies in terms of architecture, application scenarios, and sharing modes were compared, and the basic requirements for indirect network sharing applicable to different regions worldwide were put forward. All these were aimed at providing useful references for future research on technologies and business models as well as wider applications.关键词：5G-Advanced;5G cross-network roaming technology;indirect network sharing;core network85|335|0更新时间：2025-11-06
- 摘要：With data gradually becoming a form of asset for enterprises, structured data is increasingly being shared both internally and externally to better leverage its value and avoid data silos. However, due to the fact that most existing digital watermarking schemes can only add watermark information once, repeated additions will overwrite the original watermark information, making it impossible to achieve multi-level traceability of data. In response to the above issues, a multi-level database image watermark embedding scheme based on the Chinese remainder theorem (MIWC) was proposed. This scheme used image watermarking as a carrier, preprocesses the image through Harr wavelet transform, Bloom filter and other methods to form pixel Bloom filter, and secretly segmented the watermark information according to the properties of the Chinese remainder theorem. The watermark information could be added step by step in a hierarchical manner, achieving the recording and traceability of the entire data flow chain. Functional analysis proves that the solution has the ability to embed multi-level watermarks that existing solutions do not have, and has high practical application value. Experimental results have shown that this scheme has reversibility and high efficiency in embedding and extracting watermarks. At the same time, the scheme has strong robustness, it can resist attacks against watermarks.关键词：database watermark;data traceability;Chinese remainder theorem;Harr wavelet transform66|234|0更新时间：2025-11-06
- 摘要：A prediction algorithm based on feature correlation analysis and improved CNN (convolutional neural network)-LSTM (long short-term memory) network was proposed based on the power digital space technology system. Firstly, in the attention layer, the user correlation was analyzed, and the influencing factors strongly related to power load users were weighted and averaged. The weight of these categories in user power load prediction was adjusted to strengthen the influence of few and important power sample features on power load calculation and avoid the interference of redundant features, achieving feature selection. Secondly, in the CNN layer, by constructing a two-layer neural network, the spatial features of the samples are extracted and inputted into LSTM. Finally, LSTM extract the temporal features of the sequence through the LSTM layer and output the prediction results. The experiment was based on confidential data from six important users supplying power resources in a northern city for prediction. The experimental results show that the MAPE based on correlation analysis and improved CNN-LSTM model was lower than the error of LSTM, random forest, and BP neural network in five types of user tests.关键词：unbalanced sample;LSTM;power load forecasting;CNN;correlation analysis58|529|0更新时间：2025-11-06
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