Network traffic research relies on large-scale, high-quality traffic data. However, obtaining such data remains difficult because of privacy constraints, collection costs, class imbalance, and continuous updates. These challenges have increased researchers' interest in traffic generation. Although many generation methods have been proposed, existing studies and surveys often overlook two key questions: what form of traffic is generated and what practical objectives it can support. Based on 113 candidate records published from 2019 to 2026, this survey provides a detailed analysis of 39 representative network traffic generation studies through the lenses of representation levels and objective consistency. We organize existing methods into four representation levels and analyze how generated data relate to usage scenarios. We find that many methods preserve information that is useful for downstream tasks such as classification and intrusion detection, but task usefulness does not guarantee replayability or usability in real network environments. High-level representations are easier to model, yet they often discard protocol semantics, packet dependencies, and communication logic. We therefore distinguish task consistency from protocol consistency and show that the latter remains underexplored. We further summarize evaluation practices, discuss level-specific metrics, and highlight future directions including controllable generation, protocol-aware state-consistent synthesis, and engineering-oriented evaluation.
Contrastive learning has received extensive attention because it can effectively strengthen inter-modal alignment and information complementarity in multi-modal clustering. However, there are two main problems with existing contrastive multi-modal clustering methods: (1) Current methods usually perform contrastive learning at only one or two stages and have imprecise selection strategies for negative sample pairs. (2) Most methods focus only on contrastive learning and ignore the retention of consistency information. To address the above challenges, we propose a framework of multi-stage contrastive multi-modal clustering with consistency retained. First, we design a systematic contrastive learning framework, which includes early-middle-late contrastive learning between single features, fused features, and clusters, respectively, which improves the semantic integrity and information fidelity of the fused representation. Second, we propose a new negative sample pair selection strategy to discriminately select the correct negative sample pair for each sample. Finally, we add a new loss term to late-stage contrastive learning, which considers the consistency of clustering results of the same sample under different modalities for the first time. Therefore, compared to the baseline model, this module has improved the accuracy by an average of 7.4 percentage points (PPs). In addition, we design a consistency retained module to ensure that the model can learn the information of each modality without being disturbed by high/low quality modalities, which improves the accuracy by an average of 2.2 PPs. Experiments conducted on multiple multi-modal datasets demonstrate that the accuracy of our method is 4.1 PPs higher on average compared to the second-best multi-modal clustering method, proving the excellent potential of our method.
In practical reverse engineering, stripped and optimized binaries lack high-level semantics, hindering automated function understanding. This paper adopts binary function name prediction as a benchmark to evaluate open-source large language models (LLMs) for function-level semantic inference under realistic conditions. We systematically examine key factors affecting performance, including the pretraining domain, model scale, architecture and optimization settings, prompting, and contextual information. Experiments on a large-scale multi-project binary dataset reveal clear limitations in recovering function semantics from stripped binaries. Code-oriented LLMs consistently outperform general-purpose models, while increasing the model size alone does not yield monotonic gains. We further show that realistically recoverable contextual signals, especially structurally recoverable cross-function contexts, substantially mitigate semantic sparsity and improve prediction quality. These results delineate current capability boundaries of LLM-based binary semantic understanding and suggest future directions in contextual modeling and domain-adaptive techniques for practical reverse engineering.
Automated dental age estimation is essential for clinical dentistry. However, traditional methods like Demirjian's tooth development stage (TDS) require time-consuming manual annotation by trained experts. Moreover, population differences in dental development patterns can significantly affect the accuracy of age estimation, highlighting the need for an ethnicity-specific reference data set (RDS). This study aims to use deep neural networks (DNNs) for automated estimation of Demirjian's TDS of molars on digital panoramic radiographs, integrating the most complete southern Chinese RDS for ethnicity-appropriate dental age assessment. Panoramic radiographs from individuals aged 2 to 25 years are annotated for molar TDS and used to train eight DNN architectures, including AlexNet, DenseNet-201, and ResNet-50. DenseNet-201 achieves the highest accuracy of 93% in classifying molar TDS. Most misclassifications involve adjacent stages. The integrated mean dental age (IMDA) is obtained by mapping the predicted molar TDS using the southern Chinese RDS. The mean difference and correlation coefficient between the estimated IMDA from the best-performing DNN (AlexNet) and chronological age are -0.063 years (-3.3 weeks) and r=0.898 (p< 0.001), respectively. These findings demonstrate that combining DNN for TDS estimation with ethnic-specific RDS enables accurate and reliable dental age assessment.
Accurate and efficient estimation of aboveground biomass (AGB) is important for peanut phenotyping and field management. This study evaluates spectral and textural features derived from unmanned aerial vehicle (UAV) multispectral imagery for nondestructive and high-throughput estimation of peanut AGB. Vegetation indices (VIs) and gray-level co-occurrence matrix (GLCM) texture features (TFs) are derived from the green, red, red-edge, and near-infrared bands and assessed via six regression models. The results reveal that near-infrared TFs exhibit the highest sensitivity to AGB. Among the GLCM configurations, a setting of a 7×7 window size, a 45° orientation, and 16 gray levels produces the most stable texture representation. Although combining all VIs and TFs slightly improves prediction accuracy, the relatively large differences between the coefficient of determination (R2) and adjusted R2 in some models suggest that the full feature set contains redundant predictors and has limited model parsimony. An extreme gradient boosting (XGBoost)-Shapley additive explanation (SHAP) feature selection strategy is therefore used to identify five key variables: difference vegetation index (DVI), variance, mean, energy, and modified soil-adjusted vegetation index (MSAVI). This compact feature set substantially reduces predictor dimensionality, limits the differences between R2 and adjusted R2 to below 0.020, and achieves consistent predictive accuracy, with R2 above 0.840 and the root mean square error (RMSE) below 0.055 kg/m2. The proposed approach provides a practical tool for high-throughput peanut biomass monitoring, phenotyping, and production management.
This paper proposes a novel continuous aperture array (CAPA)-assisted integrated communication and navigation (ICAN) framework for low Earth orbit (LEO) satellite constellations. Within this framework, an electromagnetic-based collaborative transmission model is developed, in which multiple satellites equipped with CAPAs simultaneously radiate downlink data streams and navigation reference signals over a shared spectrum. Building upon this, the achievable communication rate and the navigation Cramér-Rao bound (CRB) are derived, which explicitly characterize the intrinsic coupling between the dual-function beamformers and system performance. To improve the positioning accuracy with a communication quality of service guarantee, a joint beamforming optimization problem is formulated to minimize the average CRB subject to transmit power budgets and minimum rate constraints. To tackle the inherent infinite dimensionality of the CAPA beamformer design, an ICAN channel subspace is introduced to equivalently transform the formulation into a tractable finite-dimensional problem, which is then efficiently solved via an iterative convex optimization algorithm. Finally, numerical results demonstrate that the proposed CAPA-assisted beamforming design algorithm significantly outperforms conventional discrete phased array architectures and other benchmark schemes, yielding notable improvements in ICAN performance.
The rapid proliferation of marine economic activities has led to explosive growth in maritime communication demands. Consequently, the scarcity of spectrum resources, the highly dynamic spectrum environment, and diverse service requirements have become increasingly critical issues. While conventional deep reinforcement learning (DRL) methods perform well for specific training scenarios, they fail to generalize effectively to unknown environments. To address this challenge, this paper investigates a maritime dynamic spectrum access and service scheduling scheme based on adaptive meta-reinforcement learning. First, we construct a channel occupancy model that encompasses both Markov frequency-hopping mode and spread-spectrum frequency-hopping mode. By incorporating a dual-queue mechanism for urgent and normal data packets, we formulate the multi-agent cooperative decision-making problem as a decentralized partially observable Markov decision process (POMDP). Second, to overcome the limited generalization capability of traditional DRL algorithms, we propose a spectrum access and service scheduling algorithm based on a meta-learning paradigm (MetaSASS), which facilitates rapid policy transfer through meta-training across a distribution of tasks. Furthermore, to resolve the adaptation efficiency bottleneck caused by fixed inner-loop learning rates, we design an adaptive meta-reinforcement learning SASS algorithm (AMRLSASS). This algorithm employs a task encoding module to dynamically adjust learning rates, thereby balancing convergence speeds across tasks of varying complexities. Simulation results demonstrate that the proposed AMRLSASS algorithm outperforms baseline algorithms across a series of metrics within unknown environments and effectively validate the superiority of the proposed method in complex maritime environments. The algorithm proposed in this paper provides an effective solution for the rapid adaptation of intelligent spectrum access and service scheduling to unknown scenarios in maritime wireless communication systems.
This paper proposes a power amplifier (PA) linearization architecture with complementary filtering to realize high-linearity broadband signal transmission. The architecture fully leverages the advantages of analog and digital filtering techniques, ensuring high harmonic suppression within a compact size and reducing the demand for digital resources. A miniaturized triple-mode bandpass filter (BPF) with a wide stopband is integrated, which effectively suppresses the harmonic components of the PA output. Meanwhile, digital filtering is employed to mitigate intermodulation distortion (IMD) that cannot be effectively suppressed by the integrated analog filter, thus forming a complementary filtering digital predistortion (CF-DPD). As the CF-DPD herein focuses only on addressing the residual distortion left by analog filtering, it naturally reduces the system's sampling rate requirements. Experimental results demonstrate that the proposed filtering power amplifier with integrated feedback port (FPA-FB) delivers 38.7-40.5 dBm output power, 8.4-11.9 dB gain, and a peak power-added efficiency (PAE) of 42.6%-52.1%. The prototype occupies an overall size of 67 mm×32 mm, corresponding to 1.1λg×0.5λg, indicating a compact implementation. Moreover, when transmitting a 400-MHz 5G new radio (NR) signal, the system achieves a normalized mean square error (NMSE) of -32.2 dB and an adjacent channel leakage ratio (ACLR) of -45.1/-47.2 dBc at a reduced sampling rate of 850 mega samples per second (MSPS).
A highly integrated and compact dual circularly and linearly polarized metasurface antenna array operating in the X-band is developed in this study. By implementing an aperture-sharing configuration that enables two radiating patch elements to be shared between left- and right-hand circular polarization channels, significant miniaturization and efficient spatial utilization are achieved for multifunctional airborne meteorological radar and communication platforms. The introduction of a sequential-rotation feeding structure and metasurface patches enhances the circular polarization bandwidth characteristics of the antenna. Miniaturization of the dual circular polarization array and efficient spatial utilization are achieved by sharing two patch elements. Additionally, the simultaneous operation of the two circular polarization ports provides linear polarization radiation functionality. A microwave imaging example based on a back-projection algorithm is implemented to demonstrate the meteorological detection capability of the designed antenna. The antenna is simulated and physically designed, yielding a measured impedance bandwidth of 9.6-14 GHz (37.39%), a 3-dB gain bandwidth of 10.6-13.56 GHz (23.5%), and a peak gain of 12.83 dBi at 12 GHz. Stable dual circularly and linearly polarized radiation is exhibited, demonstrating versatile polarization-switchable characteristics that are ideal for airborne meteorological radar systems to detect weather conditions ahead of the aircraft and enable simultaneous communication capabilities.