Online FirstList of IssuesDocument Types

    Dynamic Side-information Reconstruction Method Guided by Spatial-Frequency Dual-Domain Constraints for Distributed Video Coding AI Introduction

    Hong Mo, Qiangyuan Xiao, Qingwang Wang, Jianhua Chen, Jiaxuan Li, Tao Shen
    DOI:10.11959/j.issn.1000-436x.TXXB260591
    Abstract:To address the degradation of side-information (SI) quality in distributed video coding caused by motion estimation mismatches in regions with complex motion and occlusion, a dynamic reconstruction method for unreliable SI regions with spatial-frequency dual-domain constraints is proposed. The Wyner-Ziv (WZ) encoder employs a low-complexity block mode decision mechanism to screen potential unreliable motion estimation regions and extracts a small number of pixels from them as spatial domain constraint information. The WZ decoder first exploits the estimated residual between the auxiliary pixel and the initial SI to refine the initial SI by combining the local structural correlation. Then, in the band-by-band Slepian-Wolf decoding process, the decoded DCT coefficients are introduced and the spatial domain residual is updated synchronously. The spatial domain and frequency domain dual constraints are jointly constructed to progressively reconstruct the unreliable SI regions. Experiment results demonstrate that, compared with mainstream predictive video coding schemes, the proposed method can effectively enhance SI and achieve competitive rate-distortion performance on low- and medium-motion video sequences.  
    Keywords:distributed video coding;side-information improvement;spatial-frequency dual-domain;motion estimation;low-complexity coding  
    26
    |
    36
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-11-10

    Multi-stage Attack Modeling and Trust Mechanism Vulnerability Analysis for Computing Power Networks AI Introduction

    Xue Yangguang, Zhu Kongni, Wang Yichuan, Tian Mengjie, Zhang Panpan
    DOI:10.11959/j.issn.1000-436x.TXXB260596
    Abstract:With the open access and collaborative scheduling of heterogeneous computing resources in computing power networks, trust evaluation mechanisms face strategic attacks driven by long-term behavior accumulation. To address the difficulty of characterizing multi-stage coordinated attacks that exploit trust evolution, this paper proposes an attack modeling method based on the discrete-time partially observable Markov decision process (DT-POMDP). The model divides attacks into latent, disguise, and pollution stages, jointly considers trust evolution, attack gain, detection risk, and execution cost, and captures the attacker's dynamic decisions under incomplete information. Reinforcement learning is used to solve attack strategies, analyzing the effects of trust thresholds, latency duration, and node scale on attack performance. The model is further extended to centralized scheduling and distributed consensus scenarios to study computing-power misreporting and Sybil attacks. Simulations based on OMNeT++ and Python show that trust accumulation mechanisms expand the latent attack window under certain conditions. In the supplementary task-level model with a fixed single attacker, resource-aware scheduling, and detection based on cumulative service time, increasing the node scale reduces the attack success rate, while across the four examined resource profiles, increased resource heterogeneity is accompanied by a higher attack success rate. Additionally, dynamic threshold adjustment and weighted consensus mechanisms effectively reduce the attack success probability, but at the cost of system availability and communication overhead. This work reveals the attack-defense game in trust-driven computing power networks and provides a theoretical basis for security mechanism design.  
    Keywords:computing power network;Attack modeling;DT-POMDP;Latent attack;Combined attack  
    38
    |
    54
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-11-10

    A Survey of OTFS Technologies for 6G Integrated Sensing and Communication AI Introduction

    LIAO Yong, CHEN Jinyang
    DOI:10.11959/j.issn.1000-436x.TXXB260595
    Abstract:Sixth-generation mobile communication systems (6G) must support reliable data transmission and high-precision environmental sensing in high-mobility environments with complex time-varying channels. Integrated sensing and communication (ISAC) has therefore become an important research direction for future wireless communications. Orthogonal time frequency space (OTFS) modulation represents communication channels and sensing echoes in the delay-Doppler domain, providing a common parameter representation for joint communication and sensing processing in highly dynamic scenarios. This paper reviews OTFS-ISAC research for 6G. It first introduces system models, communication and sensing performance, and evaluation metrics. It then focuses on key OTFS-ISAC technologies, including waveform design and resource allocation, joint communication and sensing reception, network cooperation and security, and learning-based ISAC technologies. Finally, it discusses current technical challenges and outlines future research directions in the context of representative highly dynamic application scenarios.  
    Keywords:sixth-generation mobile communication system;integrated sensing and communication;orthogonal time frequency space;high-mobility communication  
    63
    |
    57
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-11-10

    Robust Secure Transmission in IRS-Assisted RSMA-ISAC Networks with Nonlinear Energy Harvesting AI Introduction

    Duan Siyi, Li Shidang, Li Chunguo, Zhao Xiaohu
    DOI:10.11959/j.issn.1000-436x.TXXB260604
    Abstract:For intelligent reflecting surface (IRS)-assisted rate-splitting multiple access (RSMA) integrated sensing and communication (ISAC) networks with statistical channel state information (CSI) errors, robust beamforming was proposed to maximize guaranteed sum rate under communication, sensing, energy, and security constraints. Energy-harvesting (EH) users were treated as potential eavesdroppers, with outage constraints imposed on decoding, leakage, and harvesting. Multi-branch EH was adopted for nonlinear rectifiers, and projected gradient ascent (PGA) was used offline to map minimum DC power requirements to RF input thresholds. An augmented channel model with direct and cascaded Gaussian errors was constructed to capture IRS-coupled channel-error statistics. Bernstein-type inequalities, successive convex approximation (SCA), rank-one penalties, and alternating optimization were employed for active–passive beamforming. At an outage threshold of 0.9, a guaranteed sum rate of 26.3 bit/(s·Hz) was achieved, 2.7% above conventional nonlinear EH, at 0.1, an 8.8% gain over IRS-free RSMA was retained under stringent reliability constraints.  
    Keywords:IRS;RSMA;ISAC;CSI;nonlinear energy harvesting;secure transmission  
    28
    |
    45
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-11-10

    BGP route hijacking detection based on AS business relationship confidence AI Introduction

    Shen Chen, Li Yuan, Cheng Zhiwei, Li Xiaoxia
    DOI:10.11959/j.issn.1000-436x.TXXB260551
    Abstract:To address the reliance of Border Gateway Protocol (BGP) route hijacking detection on deterministic autonomous system (AS) business relationships, which hardly accommodates legitimate multiple origins and complex interconnections, a detection method based on relationship uncertainty propagation and selective decision-making was proposed. Edge-level relationship probabilities were estimated from multi-source relationship results and propagated to path-policy risk, while announcement risk was scored on origin history, resource authorization status and historical novelty, with rejection rules applied to low-confidence decisions. Semi-synthetic experiments showed that the proposed method achieved an F1 score of 0.9248±0.0055 and a false positive rate of 0.0716, with recall of 0.9667 and 0.8966 for the two hijack types. When only 10% of observed peer ASes were retained, the false positive rate rose to 0.1975; average latency was 1.339 ms per announcement, and 11 of 12 public cases triggered alerts. Results indicate that relationship uncertainty modeling improves the interpretability and robustness of path anomaly decisions under limited observation coverage.  
    Keywords:BGP route hijacking;AS business relationship;trustworthy inference;multi-source evidence fusion;path anomaly detection  
    25
    |
    35
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-11-10

    Efficient Quantum Circuit Implementation of the ASCON-HASH Hash Function AI Introduction

    Zou Jian, Yu Jinlong, Sun Ying, Li Junkang
    DOI:10.11959/j.issn.1000-436x.TXXB260539
    Abstract:As a hash function in the NIST lightweight cryptography standard, ASCON-HASH has broad application prospects in scenarios such as the Internet of Things, embedded systems, and blockchain. In this paper, targeting existing quantum circuit implementations of ASCON-HASH, we propose a low-T-depth out-of-place implementation and a low-qubit in-place implementation, respectively. Compared with existing studies, the proposed out-of-place ASCON-HASH circuit reduces the number of qubits from 62,592 to 6,720 without increasing the circuit T-depth, while the proposed in-place implementation maintains a low T-depth without using additional ancillary qubits. The results show that the proposed quantum implementations of ASCON-HASH significantly outperform existing schemes in terms of qubit count, the T-depth-qubit product cost (T-DW-cost), and the T-depth-squared-qubit product cost (TD2-M-cost).  
    Keywords:quantum circuit;ASCON;hash function;qubit count optimization;T-depth optimization  
    27
    |
    35
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-11-10

    Multimodal stochastic modeling for slant-path clutter loss below 10 GHz AI Introduction

    LI Jianxin, GUAN Ke, AN Hao, DING Xianhua, LIU Zehan, ZHANG Lei, QIAN Zhaojun
    DOI:10.11959/j.issn.1000-436x.TXXB260575
    Abstract:A multimodal stochastic model was proposed to predict the slant-path clutter loss of non-terrestrial networks (NTNs) below 10 GHz. A Gaussian distribution was adopted for line-of-sight (LOS) conditions and a generalized Gaussian distribution for non-line-of-sight (NLOS) ones, which were fused with probability weights determined by elevation angle, ground station height and scene type. The model parameters were fitted from ray-tracing simulation data and independently validated against measurements collected in different areas of Beijing and Xi’an, with Kolmogorov-Smirnov (KS) distances of 0.15-0.45 and root-mean-square errors (RMSEs) of 0.07-0.25. In addition, the slant-path clutter-loss model in Section 3.3 of ITU-R P.2108-1 was extended to frequencies below 10 GHz for comparison. The results indicate that directly applying its existing parameterization does not fully capture the measured clutter-loss distributions in the considered lower-frequency bands, whereas the proposed model may provide useful insights for the development and standardization of slant-path clutter-loss models below 10 GHz.  
    Keywords:slant-path clutter loss;non-terrestrial network;ray tracing;generalized Gaussian distribution;maximum entropy principle;6G;air-to-ground link;ITU-R P.2108-1  
    21
    |
    40
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-11-10

    Multi-prototype learning for wireless signal open-set joint recognition AI Introduction

    LIAO Leiyao, GUO Hao, JIANG Yunda
    DOI:10.11959/j.issn.1000-436x.TXXB260610
    Abstract:To address the challenges of recognizing unknown signals and characterizing intra-class multimodal distributions in open electromagnetic environments, a multi-prototype learning method for open-set wireless signal joint recognition was pro-posed in this paper. A joint learning framework was constructed, in which a shared backbone network was developed to learn collaborative representations, and a multi-prototype learning strategy was designed to adaptively learn the complex multimodal distributions of communication signals, thereby improving the accuracy of dual-branch joint recognition. In the proposed model, Weibull distributions were fitted to the sub-prototypes based on extreme value theory, and rejection thresh-olds for known and unknown classes were adaptively calculated to enable accurate rejection of unknown samples. Experimental results on RadComAWGN dataset demonstrate that our method outperforms the OpenMax baseline with the accuracy of known classes (AKC), unknown detection rate (UDR), and macro-averaged F1-score (Macro-F1) on the modulation task by 7.44%, 5.74%, and 6.08%, respectively. Moreover, compared to the single-branch performance of the proposed method, these improvements are 5.04%, 4.82%, and 3.25%, confirming the effectiveness of the joint identification approach.  
    Keywords:open-set recognition;wireless signal recognition;dual-task learning;multi-prototype representation;extreme value theory  
    28
    |
    52
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-11-10

    Identity Based Anamorphic Encryption under Various Regulatory Strategies AI Introduction

    Zhao Yi, Li Siqi, Ming Yang, Zhang Qifu
    DOI:10.11959/j.issn.1000-436x.TXXB260543
    Abstract:Modern traditional encryption relies on two premises: user autonomy in message selection and private key confidentiality. When regulators obtain private keys, privacy communication is threatened. Anamorphic encryption enables secure covert communication, and identity based anamorphic encryption has become a research focus with the wide use of identity based encryption. This paper designs identity based anamorphic encryption channels with access control for two regulatory scenarios. For weakly regulated settings allowing identity based encryption, a homogeneous framework leverages structural similarity to achieve covert extraction via ciphertext randomness, balancing consistency and transferability. For highly constrained scenarios only permitting public key encryption, a dual-layer heterogeneous framework ensures compliance and security even if regular keys are leaked.  
    Keywords:Anamorphic encryption;public key encryption;Certificateless public key encryption;Identity based encryption;Identity based anamorphic encryption  
    61
    |
    51
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-26

    Bilevel Reinforcement Learning Dynamic Incentive Mechanism for Trustworthy Data Sharing AI Introduction

    Duan Meijiao, Huang Ridong, Li Hui, Wang Xi
    DOI:10.11959/j.issn.1000-436x.TXXB260553
    Abstract:The operation of data-driven platforms relies on users' continuous contribution of high-quality data, while data uploading is often accompanied by privacy risks and individual costs. Existing incentive mechanisms are mostly based on static rules, rendering them inadequate for dynamic environments and user heterogeneity, thereby leading to low incentive efficiency and an imbalance between platform revenue and incentive costs. To address these challenges, a bilevel reinforcement learning-based dynamic incentive mechanism for trustworthy data sharing was proposed. A dynamic game-theoretic model incorporating privacy utility, incentive payoffs, and approximate Shapley value-based data pricing was constructed. A bilevel reinforcement learning framework was designed, in which the optimal incentive policy was learned by the upper-level platform agent, while data uploading behaviors were determined by lower-level users according to individual preferences. Differentiated data sharing was encouraged through a two-dimensional incentive vector consisting of marginal incentives and fixed incentives. Decentralized incentive settlement was implemented using blockchain and smart contracts to ensure the trustworthiness of the mechanism. Simulation experiments show that, compared with the baseline schemes, the proposed mechanism reduces the total training incentive cost to approximately 1.9% of the average baseline cost, and improves the productivity metric by approximately 21%. The results demonstrate the effectiveness and adaptability of the proposed mechanism under heterogeneous user responses.  
    Keywords:bilevel reinforcement learning;data sharing;dynamic incentive mechanism;privacy cost;Shapley value;blockchain  
    69
    |
    84
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-26

    Multimodal Deepfake Detection Based on Audio-Visual Emotional Consistency AI Introduction

    WANG Lihua, YAN Weidan, WANG Xinyu, WU Zhongjun, LI Songbin, ZHANG Dengyin
    DOI:10.11959/j.issn.1000-436x.TXXB260571
    Abstract:To address the weakening audio-visual artifacts and the difficulty in capturing emotional inconsistencies in high-quality deepfake videos, this paper proposes AVEmo, a multimodal deepfake detection method based on audio-visual emotional consistency, which explores cross-modal forgery cues from both emotional content and dynamic variation. AVEmo extracts facial expression and speech emotion features, and models emotional content consistency and temporal synchronization through cross-modal attention. A multi-task learning strategy is further adopted to jointly optimize authenticity classification for the audio, visual, and audiovisual fusion branches, thereby enhancing the model’s ability to identify cross-modal emotional anomalies. Cross-dataset evaluation on five selected generation subsets of HiFi-AVDF yields an average AP of 76.8% and an average AUC of 78.9%. The average AUC exceeds those of AVFF and HAVIC by 10.6% and 2.1%, respectively. These results suggest that AVEmo can generalize across datasets under the evaluated conditions. The code is available at: https://anonymous.4open.science/r/AVEmo-00A2.  
    Keywords:Deepfake detection;Audio-visual emotional consistency;multimodal fusion;multi-task learning  
    58
    |
    68
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-26

    Verifiable CP-ABE scheme supporting attribute-level user revocation in cloud storage AI Introduction

    LI Shuquan, SHE Kun, LI Ruojie
    DOI:10.11959/j.issn.1000-436x.TXXB260632
    Abstract:To revoke a single attribute of a user without requiring key updates from the other users sharing that attribute, a verifiable ciphertext-policy attribute-based encryption (CP-ABE) scheme was constructed from the minimum cover set of a key encryption key (KEK) binary tree and user-bound path elements. A data service manager performed revocation on the ciphertext side by updating only the affected ciphertext rows and node components; the remaining users kept their keys unchanged, and collusion between revoked and non-revoked users was prevented. All bilinear pairings were outsourced, and the user verified the transformed result. Selective message confidentiality of the scheme was proved in the standard model under the decisional q-parallel bilinear Diffie-Hellman exponent (BDHE) assumption. Experimental results show that the user-side decryption cost is constant and independent of the access policy size.  
    Keywords:ciphertext-policy attribute-based encryption;cloud storage security;attribute-level user revocation;verifiable outsourced decryption;standard model  
    89
    |
    76
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-26

    Physical Layer Key Generation Method Based on Joint Temporal Prediction Network Learning AI Introduction

    Ye Ning, Yang Linfei, Zheng Meng, Ding Jingfang
    DOI:10.11959/j.issn.1000-436x.TXXB260555
    Abstract:Physical layer key generation technique extracts shared keys from observed signals of legitimate communication parties by exploiting the randomness, reciprocity and spatial decorrelation of wireless channels. Most existing key generation methods rely only on instantaneous sampling values, failing to effectively extract temporal dynamic features and adapt to slow-varying channel scenarios. In addition, the observation discrepancy of channel states at both ends caused by non-ideal reciprocal channels leads to an increased key disagreement rate after quantization. To address these issues, a physical layer key generation method based on a joint temporal prediction network architecture is proposed. First, a prediction network based on a gated recurrent unit encoder was constructed and trained with the channel states of both the transmitter and receiver at future moments as the prediction target, yielding a base model capable of characterizing the common channel evolution law of both parties. On this basis, the feature discrepancy between the two parties was further introduced as the loss function to fine-tune the network, guiding the features of both sides toward consistency. Finally, to reduce the key disagreement caused by quantization, the quantization error was incorporated as a new loss function for a second round of fine-tuning, so that the network takes subsequent quantization into account during feature extraction, thereby lowering the key disagreement rate. Simulation results show that over the entire SNR range, the proposed method reduces the key disagreement rate by more than 58% compared with amplitude/phase quantization-based physical layer key generation methods, and by more than 23% compared with existing deep learning-based physical layer key generation methods.‌  
    Keywords:Physical layer secret key;Time-series prediction network;Channel feature extraction;physical layer security  
    71
    |
    53
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-26

    Spatiotemporal Intelligent Internet of Things: Concept, Architecture, and Implementation Paths AI Introduction

    GUO Yongan, WANG Mi, ZHANG Shougang, ZHU Shaonan, LUO Bingqing, YANG Lijun, NI Yiyang, LIU Tao
    DOI:10.11959/j.issn.1000-436x.TXXB260513
    Abstract:To address the structural contradiction between physical spatiotemporal order and digital computational intelligence, this paper proposes Spatiotemporal Intelligent Internet of Things (ST-IoT), which builds a spatiotemporal benchmark with space-based services, integrates communication-sensing-computing-intelligence as its functional core, uses massive IoT terminals as sensing endpoints, and generates intelligence via spatiotemporal intelligence, world models, and digital twins. It proposes a 5S five-layer architecture—Spatiotemporal Foundation Layer, Synoptic Sensing Layer, Synergy Network Layer, Semantic Computing Layer, Smart Service Layer—supporting progression from sensing to cognition to action; explores application paths in the low-altitude economy, smart transportation, and public safety; and analyzes technological, application, and security-trustworthiness challenges, aiming to shift IoT from passive response to proactive foresight and provide infrastructure support for new quality productive forces.  
    Keywords:Spatiotemporal Intelligence;IoT;integration of communication, sensing, computing, and intelligence;digital twin;World model  
    211
    |
    133
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-26

    Frequency-aware discriminative feature learning for cross-modality person re-identification AI Introduction

    CHENG Deqiang, NIU Kaikai, HAO Dashuai, SONG Tianshu, KOU Qiqi
    DOI:10.11959/j.issn.1000-436x.TXXB260554
    Abstract:To address the lack of joint supervision of frequency-band modulation weights based on cross-modal consistency and identity separability in existing frequency-domain methods for visible-infrared person re-identification (VI-ReID), this paper proposes a frequency-aware discriminative feature learning method for cross-modality person re-identification. First, a Phase-Guided Spectral Modulation (PGSM) module is designed to construct a shared reference amplitude from visible and infrared amplitude statistics. It adaptively adjusts amplitude responses under the guidance of phase structure and aggregates low-frequency structural information with high-frequency details, reducing inter-modality amplitude discrepancies while preserving identity-related structure and details. Second, an Adaptive Frequency-Band Characteristic Modeling (AFCM) module is developed to jointly exploit amplitude and phase descriptors to estimate the relative weights of modality-specific, modality-shared, and identity-discriminative mappings within each frequency band. Spectral responses are then adjusted according to the resulting weight distributions, enabling differentiated modeling across frequency bands. Finally, a Frequency-Band Characteristic-Guided Loss (FBGL) is introduced to construct soft characteristic-weight targets based on the cross-modal consistency and identity separability of each frequency band. These targets supervise characteristic-weight estimation, guiding the network to learn weight distributions for cross-modality identity matching. Comparisons with mainstream methods on three public datasets demonstrate that the proposed method achieves a R-1 accuracy of 78.39% and a mAP of 74.36% under the all-search setting on SYSU-MM01, validating its effectiveness.  
    Keywords:person re-identification;cross-modality;frequency-aware learning;frequency-domain modeling  
    51
    |
    51
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-26

    Multimodal Fake News Detection Driven by Event Structure–Content Collaboration AI Introduction

    DOI:10.11959/j.issn.1000-436x.TXXB260563
    Abstract:In event-graph-based multimodal fake news detection, argument omission and predicate absence can lead to inconsistent organization of event elements, while content information, including headline semantics, visual scenes, and external knowledge, remains underutilized. To address these issues, this paper proposes Dual-Perspective Collaboration for Multimodal Fake News Detection (DPC). The proposed method jointly models event structures and textual–visual content through three components: 1) organizing event elements adaptively according to predicate presence and mapping them into a unified relation modeling space; 2) modeling event relations using multi-view graph attention and employing relative modality gating to regulate the contributions of textual and visual event representations, thereby enhancing structural consistency discrimination; and 3) enriching headline semantic representations with encyclopedic knowledge and integrating textual semantics with object-level visual information through cross-modal attention to supplement content-level discriminative evidence. Experiments on three public datasets, Twitter, Weibo, and PHEME, show that DPC achieves accuracy scores ranging from 0.912 to 0.938 and consistently outperforms all compared variants in macro-F1 on predicate-free structure subsets. The results demonstrate that DPC can improve text–image consistency discrimination under conditions of incomplete event information.  
    Keywords:fake news detection;multimodal;multi-view graph attention network;event graph;relative modality gating  
    48
    |
    70
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-26

    Joint Design for Multi-UAV Cooperative Secure Fluid-Antenna ISAC AI Introduction

    Li Siwen, Ren Shuangyin, Deng Boyu, Wang Jingchao
    DOI:10.11959/j.issn.1000-436x.TXXB260492
    Abstract:Low-altitude unmanned aerial vehicle (UAV) networks require coordinated communication coverage, target localization, and secure transmis-sion. However, onboard fluid antennas introduce strong coupling among UAV positions, antenna positions, and communication and sensing beamformers, posing a high-dimensional nonconvex optimization challenge. This paper investigates a multi-UAV cooperative secure downlink system with fluid-antenna-enabled integrated sensing and communication. Local directional information is mapped to a common position co-ordinate system, and UAV positions, antenna positions, communication beamformers, and dedicated beamformers serving both sensing and artificial-noise functions are jointly optimized to maximize the secrecy sum rate subject to a constraint on the Cramér–Rao bounds for multi-target cooperative localization. The proposed distributed product-manifold Riemannian augmented Lagrangian method reconstructs local block gradients through two-stage aggregation of low-dimensional statistics and enforces fixed radiated-power constraints on complex spheres. Simulation results show that the method improves the secrecy sum rate while satisfying the sensing constraint and reduces coordination traffic compared with full-variable gradient tracking, achieving an effective tradeoff between secrecy performance and coordination overhead.  
    Keywords:low-altitude economy;fluid antenna system;integrated sensing and communication;physical-layer security;Riemannian manifold optimization  
    50
    |
    95
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-26

    Two-slot joint noncoherent constellation design for massive SIMO systems over time-correlated fading channels AI Introduction

    Li Shuangzhi, Li Luqi, Gao Xiangchuan, Guo Xin
    DOI:10.11959/j.issn.1000-436x.TXXB260560
    Abstract:To address the constellation mismatch problem caused by dynamic variations of channel statistical properties in noncoherent transmission over massive single-input multiple-output (SIMO) systems under high-mobility scenarios, this paper proposes a two-slot joint constellation design method for time-correlated channels. By exploiting the non-Euclidean geometric characteristics of the received signal covariance matrix, the method adopts Riemannian distance to construct a constellation discrimination criterion for characterizing the separability of statistical properties corresponding to different constellation points under time-correlated channels. On this basis, a three-dimensional decomposable constellation structure with amplitude-angle-phase dimensions is designed. A layered optimization strategy is adopted to jointly solve geometric parameters and bit allocation, enabling constellation parameters to be adjusted with the variation of channel correlation. Meanwhile, a supporting cascaded one-dimensional parameter estimation and detection framework is put forward to realize signal detection while controlling decoding complexity. Theoretical analysis and simulation results demonstrate that the proposed scheme achieves lower symbol error rates than the benchmark schemes across a wide range of channel correlation coefficients and numbers of receive antennas, providing a feasible solution for noncoherent transmission over time-correlated channels.  
    Keywords:massive SIMO systems;noncoherent transmission;time-correlated channels;two-slot joint constellation design;Riemannian distance  
    65
    |
    70
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-26

    Joint Optimization Design of Low-PAPR Integrated Sensing and Communication Waveform Based on PRS-IM Fusion AI Introduction

    LIAO Xi, AI Yinhua, WANG Yang, ZHOU Jihua, ZHAO Tao
    DOI:10.11959/j.issn.1000-436x.TXXB260424
    Abstract:To address the high PAPR and RSSL caused by random modulation in integrated sensing and communication, this paper proposes a PRS-IM-based low-PAPR waveform design method. Integrating PRS, IM, and QAM, a PRS-IM-QAM waveform structure is constructed. Under reliable communication power constraints, a joint PSLR and PAPR optimization model is established. For the nonconvex minimax fractional PSLR optimization problem, the Charnes-Cooper transform converts it into a second-order cone programming convex problem. For PAPR optimization, a TLADMM algorithm with inner-loop iterations reduces computational cost. Results show that PRS-IM-QAM reduces RSSL from about -35 dB to -45 dB and PAPR to about 4.5 dB. Its communication performance is slightly worse than 16QAM. Compared with the PRS benchmark, it sacrifices only about 0.75 m range resolution while significantly improving weak-target detection by approximately 20 dB.  
    Keywords:ISAC;waveform design;index modulation;Charnes–Cooper transform;ADMM;PAPR  
    58
    |
    78
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-14

    Blockchain anomaly detection technology based on sample space reconstruction AI Introduction

    Liu Aodi, Yu Jiaxin, Du Xuehui, Wang Na, Cao Lifeng
    DOI:10.11959/j.issn.1000-436x.TXXB260548
    Abstract:To solve the challenge that blockchain network anomaly detection struggles to cope with the dynamic evolution of attacks and normal behaviors, this paper proposes a blockchain anomaly detection technology based on sample space reconstruction. The method can identify the statistical distribution shift in the sample space and avoid catastrophic forgetting during model updates. The core idea of the technique is as follows. First, a temporal autoencoder is used to unsupervised learn normal traffic and construct the sample space representation of normal behaviors. Second, isotonic regression is introduced to probabilistically calibrate anomaly scores, making detection results from different time windows comparable. KL divergence and permutation tests are adopted to statistically infer traffic distributions, identify sample space shifts caused by normal behaviors, and distinguish normal sample variations from anomalous attacks. Finally, once a shift is detected, the model is updated adaptively. While retaining the memory of normal patterns, new normal samples are incorporated into the sample space. Experimental results show that the proposed blockchain anomaly detection method achieves an average AUC of 0.89. It improves performance by approximately 40% compared with static models, and the false positive rate is reduced by more than 70%.  
    Keywords:blockchain;traffic detection;sample space reconstruction;temporal autoencoder;unsupervised learning  
    68
    |
    74
    |
    0
    citations on Dimensions.
    citations on Dimensions.
    <HTML>
    <L-PDF><WORD><XML><Meta-XML>
    <Citation><Bulk Citation>
    Updated:2026-10-14