最新刊期
卷 42 ， 期 3 ， 2026
- 摘要：Intelligent recognition technology for digital modulation types can play an important role among cooperative senders, cooperative receivers, interference senders, and non-cooperative receivers, and it can enhance the robustness of transmission in cooperative communication mode and enhance the ability of information countermeasures in non-cooperative communication mode. The latest research results of intelligent recognition for digital modulation types, as well as the research background and meaning of intelligent recognition technology for modulation types, were detailed in the review, statistical learning based on the likelihood ratio test method, and feature extraction based on the pattern recognition method were described, the technology routes of intelligent recognition for digital modulation types were combed and compared in detail, and the present challenges were identified and the future blueprint on the development of the intelligent recognition for digital modulation types was proposed. This research provides references to grasp the research status and development trend of intelligent recognition for digital modulation types in the recent years.关键词：digital signal;modulation recognition;review;statistical learning;pattern recognition10|12|0更新时间：2026-04-08
Review
- 摘要：Benes network can achieve low construction costs and high throughput on high-radix-switching scenarios. However, existing route-resolving algorithms corresponding to Benes network do not guarantee rearrangeable non-blocking (RNB) switching when suffering failure of some of its internal switch units. An unbalanced Benes network was proposed and its RNB switching trait could be guaranteed in certain conditions. A trimming method to convert a Benes network to unbalanced Benes network was proposed, through which failed switching units could be blocked, and fault-tolerant switching was achieved. When solving yield rate problems of the switch array, the trimming method demonstrated advantages over the conventional method on RNB switching radix, with 56.05% in average and 93.75% in maximum. When solving high reliability switching problems of the switch array, the trimming method had 12.5% to 21.9% higher switching radix when tolerating maximum 3 faulty switch units. Moreover, a fast route-resolving method for unbalanced Benes network was proposed and verified via field programmable gate array (FPGA), and the result shows that it doesn’t become the bottleneck of the system. Furthermore, through the trimming method, controllable partial reconfiguration of Benes network can be achieved, so Crossbar-like switching form based on partial reconfiguration is also supported by Benes network.关键词：Benes network;non-blocking network;fault-tolerance;route-resolving;hardware accelerator6|8|0更新时间：2026-04-08
- 摘要：To meet the stringent requirements for communication determinism of time-critical services carried by the train real-time data protocol(TRDP), a time-sensitive train communication network scheme was proposed based on integration of time-sensitive networking(TSN) and TRDP. This scheme constructs a hierarchical time‑sensitive train communication network architecture, in which diverse service requirements of onboard train application systems are mapped through TRDP onto the TSN. The time‑synchronization and traffic‑scheduling capabilities of TSN were employed to guarantee end-to‑end determinism for time‑sensitive train traffic. Based on this architecture, a deadline-aware gate control list(GCL) generation algorithm was designed to ensure the schedulability of critical traffic. Simulation results demonstrate that the proposed scheme significantly improves the transmission time accuracy and periodic stability of TRDP periodic data, while substantially reducing end-to-end delay jitter under mixed background traffic, thereby providing theoretical support for the construction of next-generation, highly reliable and efficient train communication networks.关键词：TSN;TRDP;traffic scheduling;next-generation train communication network3|0|0更新时间：2026-04-08
- 摘要：In response to the high sensitivity, strong correlation, and real‑time requirements of industrial data, an adaptive differential privacy (ADP) framework was proposed as a new paradigm for privacy protection in dynamic environments. The theoretical evolution and industrial adaptability of ADP were systematically elaborated. Three core technical pathways were distilled: dynamic privacy budget scheduling, correlation‑aware sensitivity estimation, and privacy‑utility balance optimization. The effectiveness of the approach was validated through three typical industrial scenarios: industrial control systems, supply‑chain collaboration, and predictive maintenance. The results demonstrate that ADP can overcome the “utility‑privacy” trade‑off inherent in static differential privacy, enabling synergistic optimization of privacy preservation and data value extraction in complex industrial settings. Finally, it was pointed out that compact privacy analysis, efficient processing of high‑dimensional heterogeneous data, robustness design, and the construction of a standardized ecosystem represented key future directions for research and application in industrial data security.关键词：industrial data security;adaptive differential privacy;privacy-utility trade-off;dynamic budget scheduling;correlated sensitivity;industrial Internet13|5|0更新时间：2026-04-08
- 摘要：Unmanned aerial vehicle (UAV) plays a crucial role in disaster early warning and battlefield reconnaissance, yet single-UAV operation suffers from limited sensing coverage and computational capacity. To overcome these constraints, a joint detection and computation (JDC) optimization algorithm inspired by grey wolf optimization (GWO) within swarm intelligence was proposed. The JDC optimization algorithm integrated a dynamic grid-based detection model, which fused target existence probability and environmental uncertainty for adaptive deployment, with an airborne collaborative computing framework that leveraged distributed task offloading and a block coordinate descent (BCD) operator for efficient resource allocation. The simulation experiment results show that JDC reduces overall latency by 33.35% and achieved 76.47% faster convergence than conventional swarm intelligence methods. The effectiveness of detection and computation (EDC) metric improves by 10.01%～28.74% compared with benchmark algorithms. These results demonstrate the superiority of the proposed JDC optimization algorithm in ground station–denied environments, providing theoretical support for autonomous collaborative surveillance and practical guidance for real-time, high-precision target detection in UAV swarm systems.关键词：UAV swarm;collaborative computing;resource scheduling;optimization deployment;swarm intelligence algorithm64|217|0更新时间：2026-04-08
- 摘要：In real-world application scenarios, attackers often add additive noise or reverberation and other interferences to the spoofing speech, which will cause the performance of the detection system trained with clean speech to drop sharply. Therefore, an activation function was designed to replace the skip connection in the residual network, thereby proposing a synthetic speech detection system with noise-robust. After analyzing the influence of different activation functions on the skip connection of the residual block, the input features were divided into non-significant features, significant features and undetermined features, and a novel activation function was proposed. The optimal parameters of the activation function were determined through the method of variance growth. Experimental results show that compared with existing methods, the method proposed not only significantly reduces the equal error rate of the system, but also has good robustness to noise interference.关键词：spoofing speech detection;synthetic speech detection;activation function;noise-robust14|28|0更新时间：2026-04-08
- 摘要：Multi-view subspace clustering (MVSC) is widely recognized for its effectiveness in integrating multi-source heterogeneous data, yet two critical challenges are identified: insufficient robustness of affinity matrices against noise and limited capability in capturing cross-view consistent information. To address these issues, a multi-view subspace clustering algorithm was proposed by integrating dual denoising mechanisms with pairwise similarity principles. Firstly, a dual denoising strategy was designed during the initial data processing stage, where smooth denoising was achieved through low-pass filters, followed by low-rank denoising via a multi-step matrix decomposition framework. Subsequently, pairwise similarity principles were introduced, and inter-view consistency constraints were systematically constructed to effectively explore shared information across views. Finally, an optimization framework was developed using the augmented Lagrangian method (ALM) . The proposed algorithm demonstrates significant advantages across seven datasets from different domains. Compared with other algorithms, it achieves average improvements of 17.79% in accuracy, 20.72% in normalized mutual information, and 16.37% in F-score metrics. Particularly on the CESC dataset, the algorithm attains an ACC score of 0.862 9, representing a 16.13% improvement over the efficient and effective one-step multiview clustering (EEOMVC), which fully demonstrates its superior performance in multi-view data fusion.关键词：multi-view learning;subspace clustering;matrix decomposition;low-rank representation40|102|0更新时间：2026-04-08
- 摘要：To address the “data silo” problem caused by data privacy protection, federated learning offers a technological framework that enables multiple parties to collaboratively train models without sharing raw data. However, to attract high-quality data owners and ensure the long-term stable operation of the system, a fair and reasonable value evaluation and revenue distribution mechanism is essential. Although the Shapley value has been widely used in federated learning for contribution evaluation, relying on a single approach limits the multi-perspective assessment of data contribution and value distribution. For this purpose, the Least core method was introduced, providing a new perspective for value allocation by minimizing the maximum deficit. However, the computational complexity of the Least core method, similar to that of the Shapley value, posed certain computational limitations. By further combining various sampling methods to approximate the allocation results, it significantly reduced the computational cost in large-scale scenarios while ensuring the accuracy of the allocation. Experimental results demonstrate that the mechanism ensures fairness and system stability while effectively improving computational efficiency, providing a feasible solution for diverse federated learning scenarios.关键词：federated learning;game theory;least core;contribution evaluation;incentive mechanism6|0|0更新时间：2026-04-08
- 摘要：Airborne network intrusion detection may face the dual challenges of scarce abnormal samples and unbalanced data distribution. Traditional methods are difficult to simultaneously guarantee detection accuracy and generalization ability. Therefore, combined with the data augmentation method of multi-view contrastive sparse autoencoder (MCSAE), a joint optimization method with improved hierarchical sampling ensemble learning was proposed. Firstly, for the problem of missing abnormal samples, a MCSAE was designed. Through multi-view data augmentation and contrastive learning strategies, a more discriminative latent representation was learned under the framework of sparse autoencoder, and the quality of abnormal sample generation was optimized by the re-input contrast mechanism, effectively alleviating the model deviation caused by data sparsity. Secondly, for the problem of class imbalance, an improved hierarchical sampling strategy was proposed. On the basis of traditional hierarchical sampling, a global feature retention mechanism was introduced to avoid the distortion of the majority class distribution caused by local sampling, ensuring that the classifier can learn the complete statistical characteristics of the data. Finally, combined with F1 score adaptive weighted ensemble learning, diverse base classifiers such as random forest and LSTM were integrated to dynamically adjust the model weights, further improving the detection ability for minority class attacks. The experimental results show that, compared with the existing methods on the airborne network dataset, the proposed method has a 5.2% increase in recall rate and a 3.7% increase in F1 score. This provides a reliable solution for intrusion detection in complex network environments.关键词：distribution imbalance;multi-view contrastive sparse autoencoder;stratified sampling;ensemble learning42|115|0更新时间：2026-04-08
- 摘要：In recent years, millimeter-wave radar-based kick motion recognition has attracted significant attention as a hands-free human-computer interaction technology, demonstrating considerable value in applications such as smart home systems and automotive interfaces. However, in complex environments, both static interference (e.g., walls, pillars, and antenna coupling effects) and dynamic interference (e.g., pedestrian movements and subtle limb motions) continue to present substantial challenges to recognition accuracy. To achieve high-precision and robust recognition, an integrated method combining interference suppression and deep learning was proposed. The method employed vector mean cancellation and moving target indication (MTI) to suppress static interference. In contrast, dynamic interference was mitigated through a combination of Doppler weighting, constant false alarm rate (CFAR) detection, density-based spatial clustering of applications with noise (DBSCAN), and connected region constraint. Subsequently, multi-frame range-Doppler maps (RDM) and range-angle maps (RAM) were extracted to serve as model inputs. A dual-stream CNN-MHSA-STCN architecture was constructed, incorporating a convolutional neural network (CNN), multi-head self-attention (MHSA), and a simplified temporal convolutional network (STCN) for comprehensive motion recognition. Experimental results demonstrate that the proposed method achieves a recognition accuracy exceeding 98% on a self-collected dataset, while maintaining high precision and robustness in complex operational environments.关键词：millimeter-wave radar;kick motion recognition;interference suppression;dual-stream CNN-MHSA-STCN31|48|0更新时间：2026-04-08
- 摘要：With the development of smart grids, power line communication (PLC) faces challenges such as surging service data and increasing transmission rates, making high-precision channel estimation crucial for enhancing communication performance. Traditional compressed sensing methods rely on discretized dictionaries, which suffer from basis mismatch issues that limit multipath delay resolution. Addressing the significant multipath effects characteristic of power line channels, a continuous-domain sparse recovery framework for channel estimation was proposed. Leveraging the uniform spacing of subcarrier frequencies in orthogonal frequency division multiplexing (OFDM) systems, the framework mapped multipath delays into a frequency-domain response with a Vandermonde matrix structure, achieving sparse representation. The atomic norm minimization (ANM) algorithm was employed to solve the problem in the continuous parameter space, avoiding discrete grid errors and significantly improving delay resolution. Combined with the exponential decay characteristic of power line channels, the multipath weights were derived analytically through frequency-domain attenuation factors, enabling closed-form reconstruction of the channel response. Simulations demonstrate that the proposed method achieves super-resolution performance in delay resolution, showing significant advantages over algorithms such as orthogonal matching pursuit (OMP).关键词：PLC;channel estimation;ANM;Vandermonde structure;super-resolution estimation37|16|0更新时间：2026-04-08
- 摘要：Wi-Fi channel state information (CSI)-driven human activity recognition (HAR) has been widely studied for applications in activity monitoring. However, challenges such as high deployment costs and limited sensing range remain. To address these, long term evolution attention-guided ConvNeXt (LTE-ACN) was proposed, which used mobile communication LTE signals for activity recognition. Firstly, effective features of channel state information were extracted based on cell reference signals. Then, signal processing methods including noise filtering, Savitzky-Golay smoothing, peak-valley enhancement, and Gramian angular field transformation were applied to construct the motion dataset. Finally, the enhanced image data was input into the LTE-ACN model, in which the improved ConvNeXt architecture effectively reduced the feature information loss while the incorporated convolutional block attention module (CBAM) attention mechanism enhanced the expressive capability of key features and strengthened the feature correlations in the spatial domain. Experimental results demonstrate that the proposed method achieves an average accuracy of 96.44% in the six actions recognition, verifying the feasibility of LTE signal-based human motion recognition.关键词：mobile communication network signal;CSI;HAR;attention mechanism;convolutional neural network22|124|0更新时间：2026-04-08
- 摘要：In satellite Internet of things (IoT) communications, both pilot and storage resources are limited such that conventional synchronization scheme and its hardware implementation have faced great challenges. To address this issue, a minimal synchronization scheme was proposed based on double correlation operators, including an auto-correlation based frequency-phase decoupled estimation and a cross-correlation based crorelation blockwise data compensation. In specific, an auto-correlation frequency offset estimator robust was provided by the former to the Rice fading and a low-complexity phase offset estimator with a strong ability of anti-frequency offset. A small amount of cross-correlation based data compensation values were used by the latter to greatly lower the impact of residual frequency offset. Simulation results show that with a maximum pilot overhead of only five percents, the proposed minimal synchronization scheme can implement the frequency-to-phase decoupling, and can reduce storage space occupation by 98% and achieve near error performance at the low signal-to-noise ratio (SNR) compared with the conventional synchronization scheme, which has certain practical significance for engineering implementation of future satellite IoT communication systems.关键词：pilot resource;storage resource;double correlation operator;blockwise compensation;satellite IoT communication47|35|0更新时间：2026-04-08
- 摘要：In the context of digital transformation, accurate modeling of power industry business processes is considered crucial for achieving intelligent operation and standardized management. To this end, a domain-adaptive language model named ElecBPM-LoRA was proposed for power workflow standardization. By incorporating low-rank adaptation (LoRA) and a structured prompting mechanism, the model was enabled to accurately transform natural language into structured statements and automatically generate business process diagrams compliant with the BPMN 2.0 specification. Experimental results demonstrate that ElecBPM-LoRA outperforms mainstream large language models in identifying power industry process elements, and exhibits significant advantages in the completeness of process structure modeling and semantic consistency. The proposed approach provides reliable support for process automation and standardized modeling in the field of critical infrastructure.关键词：large language model;BPM;chinese text processing;automatic modeling;LoRA;prompt engineering22|33|0更新时间：2026-04-08
Research and Development
- 摘要：In the digital age, accurately grasping user needs is the key to acheive accurate marketing and personalized services. Aiming at this scenario, an improved singular spectrum analysis K-means (SSA-Kmeans) algorithm was proposed, which effectively solved the inefficient problem of manual selection of window length and feature components in the traditional SSA. Firstly, SSA was used to extract the core feature components of user access data in this algorithm. Then, K-means was combined for cluster analysis. The experimental results show that the clustering effect is significantly improved. The Davies-Bouldin index is 0.407 1 and 0.067 2 lower than that of direct clustering and wavelet transform denoising clustering, respectively. Moreover, the cluster division is more accurate. Based on the optimized clustering results, a differentiated operation strategy was further developed to provide customized services for different user groups. This method provides an efficient solution for website precision marketing and user retention, and has important practical application value.关键词：user clustering;SSA;K-means clustering algorithm;Davies-Bouldin index14|9|0更新时间：2026-04-08
- 摘要：The current data of electric power new energy engineering presents complex characteristics of multi-source heterogeneity, spatiotemporal coupling, and spatial non-stationarity, which make the boundaries of the full-scale data mining area fuzzy and lead to a decrease in the accuracy of full-scale data mining of electric power new energy engineering. Therefore, a multi-level grid based method for enhancing the mining of full-scale data of electric power new energy engineering based on spatial information was proposed. By quantifying the correlation between data through cosine similarity and Pearson correlation coefficient, a similarity driven grid density peak calculation method was adopted, combined with distance thresholding processing. Finally, a multi-level spatial information grid was used to achieve refined spatial partitioning of complex power new energy data. The row and column indexes of the partition results were calibrated as feature summaries, and the index labels were added to the grid cell feature values. After adding the index labels, the inverse distance weighting method was used to calculate the mining index threshold, in order to achieve full-scale data augmentation mining of power new energy engineering. The experimental data shows that the proposed method has high accuracy in engineering full-scale data augmentation mining, and the consistency between the mining results and the target results is strong. Therefore, the proposed method has strong practical value.关键词：multi-level grid of spatial information;electric power new energy engineering;full-scale data;data augmentation mining;cosine similarity23|65|0更新时间：2026-04-08
- 摘要：Customer escalated complaints serve as a critical indicator for measuring service quality and are vital for improving service quality and resolving customer issues. A hybrid learning approach LLM-RFE-XGBoost was proposed for early warning of potential escalated complaints. Firstly, large language models (LLM) was utilized to extract semantic features from customer call text. Then these features were integrated with original structured data, after which recursive feature elimination (RFE) was applied to select the optimal feature set. Finally, XGBoost was employed for prediction using all selected features. To validate the effectiveness of the model, predictive analysis was conducted using production data from a provincial telecom operator as the research subject. Empirical results demonstrate that the proposed LLM-RFE-XGBoost hybrid approach delivers optimal predictive performance. After practical application in a provincial telecom operator, escalated complaints decreased by 6.7%, which is of great significance for improvement of the service quality and customer satisfaction.关键词：escalated complaint;XGBoost;recursive feature elimination;large language model18|23|0更新时间：2026-04-08
- 摘要：Promoting the construction of 5G high-speed railway (HSR) dedicated networks plays an important role in improving the travel experience of HSR users, which has become a key focus of 5G strategy of operators. Firstly,the scenario-specific issues faced by the construction of 5G HSR dedicated networks were comprehensively analyzed, and then a new optimization solution was proposed to address the problems including frequent non-HSR user intrusion and the lack of comprehensive user attribute perception capabilities in the current HSR dedicated networks. By coordinating the wireless network and the core network, and leveraging the network data analytic function, the proposed method achieves, for the first time, performance optimization for the 5G HSR dedicated network based on intelligent mobile user profiling. The testing results indicate that the method can achieve an accuracy of over 90% for user attribute identification, and is expected to further reduce over 70% of the non-HSR user intrusion. Overall, the proposed solution provides an important reference schema for the application and implementation of network intelligent technology, and helps to promote the construction of AI-native networks, thus achieving more precise network user perception and intelligent scheduling, and supporting more diverse business scenarios.关键词：5G;high-speed railway;dedicated network;user profiling;intelligent scheduling45|149|0更新时间：2026-04-08
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：952130 Visits Today：1298
0