Federated learning (FL) has been strongly promoted in wireless edge networks because it facilitates collaborative training of machine learning models while ensuring the privacy and security of individual user data. Among different FL frameworks, Decentralized Federated Learning (DFL) framework offers greater flexibility and broader applicability compared to traditional Centralized Federated Learning (CFL) framework. However, existing DFL framework fails to accommodate the heterogeneity of multi-dimensional resources (e.g., communication and computing resources), and further suffer from high communication overhead. These limitations result in reduced training efficiency. To address these issues, this paper proposes SoloDFL, a framework integrating heterogeneity-aware communication topology reconstruction and parameter filtering. In SoloDFL, each client communicates with only one neighboring client, constructing a one-on-one communication topology, and exchanges only filtered local model parameters during communication. By doing so, SoloDFL copes with the synchronization barriers caused by system heterogeneity and further reduces communication overhead. The convergence of SoloDFL is proved theoretically. Extensive experiments show that SoloDFL achieves up to a $3.9times $ acceleration and reduces communication costs by an average of 20.5% compared to the benchmark algorithms.
{"title":"Joint Communication Topology Reconstruction and Parameter Filtering for Heterogeneous Decentralized Federated Learning","authors":"Mingjun Duan;Xiaoning Zhang;Lang Fan;Yijing Liu;Jiayi Jiang;Haijun Zhang","doi":"10.1109/TCCN.2026.3662666","DOIUrl":"https://doi.org/10.1109/TCCN.2026.3662666","url":null,"abstract":"Federated learning (FL) has been strongly promoted in wireless edge networks because it facilitates collaborative training of machine learning models while ensuring the privacy and security of individual user data. Among different FL frameworks, Decentralized Federated Learning (DFL) framework offers greater flexibility and broader applicability compared to traditional Centralized Federated Learning (CFL) framework. However, existing DFL framework fails to accommodate the heterogeneity of multi-dimensional resources (e.g., communication and computing resources), and further suffer from high communication overhead. These limitations result in reduced training efficiency. To address these issues, this paper proposes SoloDFL, a framework integrating heterogeneity-aware communication topology reconstruction and parameter filtering. In SoloDFL, each client communicates with only one neighboring client, constructing a one-on-one communication topology, and exchanges only filtered local model parameters during communication. By doing so, SoloDFL copes with the synchronization barriers caused by system heterogeneity and further reduces communication overhead. The convergence of SoloDFL is proved theoretically. Extensive experiments show that SoloDFL achieves up to a <inline-formula> <tex-math>$3.9times $ </tex-math></inline-formula> acceleration and reduces communication costs by an average of 20.5% compared to the benchmark algorithms.","PeriodicalId":13069,"journal":{"name":"IEEE Transactions on Cognitive Communications and Networking","volume":"12 ","pages":"6293-6308"},"PeriodicalIF":7.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147299515","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Recently, Neural Radiance Fields (NeRF) have demonstrated excellent fidelity performance in 3D scene reconstruction. However, their significant data demands pose challenges for low-altitude platforms, particularly due to stringent bandwidth constraints. Semantic communication (SC) offers a solution by transmitting only the most task-relevant information, which would significantly reduce the required data throughput. Besides, the recent emergence of large AI models (LAMs) further strengthens SC by providing powerful, pre-trained semantic encoders that can extract and compress high-value features. Therefore, in this paper, we propose an end-to-end framework, LAM-SC-3DR, which integrates LAM-driven semantic extraction, semantic communication, and NeRF-based 3D reconstruction to optimize remote 3D scene recovery for low-altitude platforms. The framework is consists of three main modules: the Semantic Feature Extraction (SFE) module, which utilizes a pre-trained LAM to extract multi-level semantics (including object, appearance, and geometry) from 2D images; the Joint Semantic–Channel Coding (SCC) module, which integrates semantic compression with channel coding for reliable transmission over the noisy wireless links; and the 3D Scene Reconstruction (3DSR) module, which combines the received semantics to create photorealistic, semantically consistent volumetric renderings. Extensive evaluations demonstrate that LAM-SC-3DR can reduce transmission load by up to 96%, while maintaining 3D semantic reconstruction quality.
{"title":"From Local to Global: Semantic Communication-Driven Remote 3D Scene Reconstruction Using Low-Altitude Platforms","authors":"Tianle Mai;Haipeng Yao;Gepeng Zhu;Chenlang Jin;Xiangjun Xin","doi":"10.1109/TCCN.2026.3662333","DOIUrl":"10.1109/TCCN.2026.3662333","url":null,"abstract":"Recently, Neural Radiance Fields (NeRF) have demonstrated excellent fidelity performance in 3D scene reconstruction. However, their significant data demands pose challenges for low-altitude platforms, particularly due to stringent bandwidth constraints. Semantic communication (SC) offers a solution by transmitting only the most task-relevant information, which would significantly reduce the required data throughput. Besides, the recent emergence of large AI models (LAMs) further strengthens SC by providing powerful, pre-trained semantic encoders that can extract and compress high-value features. Therefore, in this paper, we propose an end-to-end framework, LAM-SC-3DR, which integrates LAM-driven semantic extraction, semantic communication, and NeRF-based 3D reconstruction to optimize remote 3D scene recovery for low-altitude platforms. The framework is consists of three main modules: the Semantic Feature Extraction (SFE) module, which utilizes a pre-trained LAM to extract multi-level semantics (including object, appearance, and geometry) from 2D images; the Joint Semantic–Channel Coding (SCC) module, which integrates semantic compression with channel coding for reliable transmission over the noisy wireless links; and the 3D Scene Reconstruction (3DSR) module, which combines the received semantics to create photorealistic, semantically consistent volumetric renderings. Extensive evaluations demonstrate that LAM-SC-3DR can reduce transmission load by up to 96%, while maintaining 3D semantic reconstruction quality.","PeriodicalId":13069,"journal":{"name":"IEEE Transactions on Cognitive Communications and Networking","volume":"12 ","pages":"6207-6220"},"PeriodicalIF":7.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146146084","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-09DOI: 10.1109/TCCN.2026.3662678
Fuliang Li;Bocheng Liang;Haozhi Lang;Jiajie Zhang;Jiaxing Shen;Chengxi Gao;Xingwei Wang
Manual network configuration tools are constrained by their reliance on extensive domain expertise and rigid, single-purpose designs, limiting their adaptability to diverse scenarios and complex applications. This paper introduces PreConfig, a novel language model-based framework for automating network configuration tasks. By framing tasks such as configuration generation, translation, analysis, and completion as text-to-text transformations, PreConfig unifies these processes under a single versatile model. Leveraging advancements in natural language processing, PreConfig eliminates the need for extensive manual re-engineering by automatically learning domain-specific patterns through continued training on a specialized network configuration corpus. To address the lack of domain knowledge in general language models, we construct a comprehensive dataset from vendor manuals and community forums and fine-tune a programming language model for robust performance across various tasks. Additionally, we propose ConfigBLEU, a novel evaluation metric that incorporates syntax-aware features to assess the accuracy of generated configurations. Experimental results demonstrate that PreConfig significantly outperforms existing tools and general-purpose language models in both syntactic accuracy and semantic correctness across diverse network configuration tasks. This work establishes a unified and adaptable approach for advancing network configuration automation.
{"title":"PreConfig: A Unified Language Model Framework for Network Configuration Automation","authors":"Fuliang Li;Bocheng Liang;Haozhi Lang;Jiajie Zhang;Jiaxing Shen;Chengxi Gao;Xingwei Wang","doi":"10.1109/TCCN.2026.3662678","DOIUrl":"https://doi.org/10.1109/TCCN.2026.3662678","url":null,"abstract":"Manual network configuration tools are constrained by their reliance on extensive domain expertise and rigid, single-purpose designs, limiting their adaptability to diverse scenarios and complex applications. This paper introduces PreConfig, a novel language model-based framework for automating network configuration tasks. By framing tasks such as configuration generation, translation, analysis, and completion as text-to-text transformations, PreConfig unifies these processes under a single versatile model. Leveraging advancements in natural language processing, PreConfig eliminates the need for extensive manual re-engineering by automatically learning domain-specific patterns through continued training on a specialized network configuration corpus. To address the lack of domain knowledge in general language models, we construct a comprehensive dataset from vendor manuals and community forums and fine-tune a programming language model for robust performance across various tasks. Additionally, we propose ConfigBLEU, a novel evaluation metric that incorporates syntax-aware features to assess the accuracy of generated configurations. Experimental results demonstrate that PreConfig significantly outperforms existing tools and general-purpose language models in both syntactic accuracy and semantic correctness across diverse network configuration tasks. This work establishes a unified and adaptable approach for advancing network configuration automation.","PeriodicalId":13069,"journal":{"name":"IEEE Transactions on Cognitive Communications and Networking","volume":"12 ","pages":"6320-6330"},"PeriodicalIF":7.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147299512","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Mobile edge computing (MEC) meets the requirements of various delay-sensitive applications by providing high-speed computing services to a large number of user vehicles simultaneously. Nevertheless, the inherent open feature of wireless channels and the constraints of limited spectrum resources present significant challenges to achieving both secure offloading and high offloading rate simultaneously. Millimeter wave (mmWave) can provide user vehicles with abundant spectrum resources, but its short wavelength causes high path loss. In this paper, we utilize hybrid beamforming and non-orthogonal multiple access (NOMA) technologies to improve the offloading rate of user vehicles and to interfere with eavesdroppers, thus improving the security of the offloading process in mmWave vehicular edge computing (VEC) networks. We first use the K-means algorithm to cluster user vehicles. Then, we minimize the system delay by jointly optimizing the analog beamforming matrix, the user vehicle transmit power and the allocation ratio of the MEC server computation resource while ensuring the security of the offloading process. The above optimization problem is formulated as a Markov decision process (MDP) and a twin Delayed Deep Deterministic Policy Gradient (TD3)-Dueling Double Deep Q Network (D3QN) based multi-agent secure computation offloading scheme is proposed to solve the MDP problem. Simulation results demonstrate that the TD3-D3QN based multi-agent scheme is able to adapt to highly dynamic VEC networks when guaranteeing the security of the offloading process and low system delay.
{"title":"NOMA and Hybrid Beamforming Aided Secure Computation Offloading for mmWave VEC Networks With Multi-Agent DRL","authors":"Ying Ju;Zhiwei Cao;Mingdong Li;Lei Liu;Qingqi Pei;Mianxiong Dong;Shahid Mumtaz;Mohsen Guizani","doi":"10.1109/TCCN.2026.3662303","DOIUrl":"10.1109/TCCN.2026.3662303","url":null,"abstract":"Mobile edge computing (MEC) meets the requirements of various delay-sensitive applications by providing high-speed computing services to a large number of user vehicles simultaneously. Nevertheless, the inherent open feature of wireless channels and the constraints of limited spectrum resources present significant challenges to achieving both secure offloading and high offloading rate simultaneously. Millimeter wave (mmWave) can provide user vehicles with abundant spectrum resources, but its short wavelength causes high path loss. In this paper, we utilize hybrid beamforming and non-orthogonal multiple access (NOMA) technologies to improve the offloading rate of user vehicles and to interfere with eavesdroppers, thus improving the security of the offloading process in mmWave vehicular edge computing (VEC) networks. We first use the K-means algorithm to cluster user vehicles. Then, we minimize the system delay by jointly optimizing the analog beamforming matrix, the user vehicle transmit power and the allocation ratio of the MEC server computation resource while ensuring the security of the offloading process. The above optimization problem is formulated as a Markov decision process (MDP) and a twin Delayed Deep Deterministic Policy Gradient (TD3)-Dueling Double Deep Q Network (D3QN) based multi-agent secure computation offloading scheme is proposed to solve the MDP problem. Simulation results demonstrate that the TD3-D3QN based multi-agent scheme is able to adapt to highly dynamic VEC networks when guaranteeing the security of the offloading process and low system delay.","PeriodicalId":13069,"journal":{"name":"IEEE Transactions on Cognitive Communications and Networking","volume":"12 ","pages":"6089-6103"},"PeriodicalIF":7.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146146086","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/TCCN.2026.3661502
Yaoqi Yang;Ming He;Wei Han;Xin Lu;Chenping Hou
Low altitude economy (LAE) can significantly promote the air-to-ground communication development by providing various data collection, transmission and computing services. However, due to the openness trait of wireless channel, LAE also faces timeliness and security concerns on sensing data’s stale and leakage at ground base station (BS) end. This can result in delayed information delivery, compromised data integrity, and increased vulnerability to malicious attacks. In this regard, to address the above concerns, this paper first establishes the blockchain-based UAV-enabled mobile crowdsensing (MCS) model. Then, after determining the impact of the blockchain on data timeliness, the mathematical expressions of the data freshness metric are derived in closed form. On this basis, a data freshness minimization problem with security premise is formulated, where the UAV’s transmission power, computing rate, and BS’s power allocation ratio are jointly optimized. Furthermore, one deep reinforcement learning-aided multi-objective optimization (MOP) algorithm is proposed to solve the formulated problem. At last, under various parameter settings, numerical results have evaluated the effectiveness of the proposals.
{"title":"Data Freshness Performance Analysis and Optimization in Timely and Secure Low Altitude Economics","authors":"Yaoqi Yang;Ming He;Wei Han;Xin Lu;Chenping Hou","doi":"10.1109/TCCN.2026.3661502","DOIUrl":"10.1109/TCCN.2026.3661502","url":null,"abstract":"Low altitude economy (LAE) can significantly promote the air-to-ground communication development by providing various data collection, transmission and computing services. However, due to the openness trait of wireless channel, LAE also faces timeliness and security concerns on sensing data’s stale and leakage at ground base station (BS) end. This can result in delayed information delivery, compromised data integrity, and increased vulnerability to malicious attacks. In this regard, to address the above concerns, this paper first establishes the blockchain-based UAV-enabled mobile crowdsensing (MCS) model. Then, after determining the impact of the blockchain on data timeliness, the mathematical expressions of the data freshness metric are derived in closed form. On this basis, a data freshness minimization problem with security premise is formulated, where the UAV’s transmission power, computing rate, and BS’s power allocation ratio are jointly optimized. Furthermore, one deep reinforcement learning-aided multi-objective optimization (MOP) algorithm is proposed to solve the formulated problem. At last, under various parameter settings, numerical results have evaluated the effectiveness of the proposals.","PeriodicalId":13069,"journal":{"name":"IEEE Transactions on Cognitive Communications and Networking","volume":"12 ","pages":"6016-6030"},"PeriodicalIF":7.0,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134232","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/TCCN.2026.3661500
Yandie Yang;Sicheng Zhang;Kuixian Li;Yun Lin
Automatic Modulation Classification (AMC) plays a crucial role in spectrum monitoring and communication security. Open Set Recognition (OSR) methods have been widely applied to AMC tasks to address the challenge of recognizing known modulation types and identifying unknown modulation signals in open-world scenarios. However, existing research primarily emphasize the accuracy of open set recognition while overlooking potential security threats posed by adversarial attacks. To address this gap, we investigate the security vulnerabilities of AMC methods under adversarial attacks in open electromagnetic environments from the perspective of artificial intelligence security. We propose two types of multi-oriented open set adversarial attacks, including Label-oriented Open Set Adversarial Attacks (OSLoA) and Feature-oriented Open Set Adversarial Attack (OSFoA). Based on the discrimination mechanism of the OSR model, we propose the OSLoA method. This method increases the confidence of misclassification for unknown signals, which causes them to be recognized as known classes. Additionally, we introduce the innovative OSFoA method. It reduces the distance between the class activation features of signals from unknown classes and those of the known classes into which unknown signals are most likely to be misclassified. As a result, the unknown classes are pushed closer to the known classes in the feature space, further enhancing the attack effectiveness. Notably, during the computation of class activation features, only those features that make positive contributions to the prediction output are retained. Comprehensive experiments were conducted on both public and real-world datasets. The results demonstrate that the proposed OSLoA and OSFoA methods achieve excellent performance and further reveal the vulnerability of AMC methods to open adversarial security threats.
自动调制分类（AMC）在频谱监控和通信安全中起着至关重要的作用。开放集识别（OSR）方法已广泛应用于AMC任务中，以解决在开放世界场景中识别已知调制类型和识别未知调制信号的挑战。然而，现有的研究主要强调开放集识别的准确性，而忽略了对抗性攻击带来的潜在安全威胁。为了解决这一问题，我们从人工智能安全的角度研究了开放电磁环境下AMC方法在对抗性攻击下的安全漏洞。我们提出了两种多面向开放集对抗攻击，即面向标签的开放集对抗攻击（Label-oriented open set adversarial Attack, OSLoA）和面向特征的开放集对抗攻击（Feature-oriented open set adversarial Attack, OSFoA）。基于OSR模型的判别机制，提出了OSLoA方法。该方法提高了对未知信号误分类的置信度，使其被识别为已知类。此外，我们还介绍了创新的OSFoA方法。它减小了未知类信号的类激活特征与已知类信号的类激活特征之间的距离，而已知类信号最有可能被错误分类。从而使未知类在特征空间中向已知类靠拢，进一步提高了攻击的有效性。值得注意的是，在计算类激活特征时，只保留那些对预测输出有积极贡献的特征。在公共和现实世界的数据集上进行了全面的实验。结果表明，提出的OSLoA和OSFoA方法取得了优异的性能，并进一步揭示了AMC方法在打开对抗性安全威胁时的脆弱性。
{"title":"Multi-Oriented Open Set Adversarial Attacks to Automatic Modulation Classification","authors":"Yandie Yang;Sicheng Zhang;Kuixian Li;Yun Lin","doi":"10.1109/TCCN.2026.3661500","DOIUrl":"10.1109/TCCN.2026.3661500","url":null,"abstract":"Automatic Modulation Classification (AMC) plays a crucial role in spectrum monitoring and communication security. Open Set Recognition (OSR) methods have been widely applied to AMC tasks to address the challenge of recognizing known modulation types and identifying unknown modulation signals in open-world scenarios. However, existing research primarily emphasize the accuracy of open set recognition while overlooking potential security threats posed by adversarial attacks. To address this gap, we investigate the security vulnerabilities of AMC methods under adversarial attacks in open electromagnetic environments from the perspective of artificial intelligence security. We propose two types of multi-oriented open set adversarial attacks, including Label-oriented Open Set Adversarial Attacks (OSLoA) and Feature-oriented Open Set Adversarial Attack (OSFoA). Based on the discrimination mechanism of the OSR model, we propose the OSLoA method. This method increases the confidence of misclassification for unknown signals, which causes them to be recognized as known classes. Additionally, we introduce the innovative OSFoA method. It reduces the distance between the class activation features of signals from unknown classes and those of the known classes into which unknown signals are most likely to be misclassified. As a result, the unknown classes are pushed closer to the known classes in the feature space, further enhancing the attack effectiveness. Notably, during the computation of class activation features, only those features that make positive contributions to the prediction output are retained. Comprehensive experiments were conducted on both public and real-world datasets. The results demonstrate that the proposed OSLoA and OSFoA methods achieve excellent performance and further reveal the vulnerability of AMC methods to open adversarial security threats.","PeriodicalId":13069,"journal":{"name":"IEEE Transactions on Cognitive Communications and Networking","volume":"12 ","pages":"6281-6292"},"PeriodicalIF":7.0,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134237","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/TCCN.2026.3661515
Tao Huang;Gengyuan Lu;Mingan Luan;Zheng Chang;Ying-Chang Liang
The rapid evolution of unmanned aerial vehicle (UAV) technology has given rise to the new concept of the “low-altitude economy (LAE)” and promoted innovation in various application fields. However, the efficiency of low-altitude services is still significantly constrained by the energy consumption and payload limitations of UAVs. Accordingly, in this paper, we propose a hybrid UAV and unmanned ground vehicle (UGV) LAE framework to address the limitation in UAV-based LAE networks. Specifically, in virtue of the advantages of UGVs in advanced communication, large batteries, and computation modules, the hybrid mode can provide an additional degree of freedom for environmental information collection in LAE networks. Nevertheless, for the considered hybrid framework, which involves the collaborative optimization of numerous agents across multiple dimensions, traditional methods are inefficient and difficult to implement effectively. Hence, we present a machine learning (ML)-based method to handle this issue. More in detail, we first design an iterative self-organizing data analysis techniques strategy to form sensing nodes (SNs) clusters, where the cluster heads (CHs) collect data from SNs in advance in order to reduce the transmission delay. Then, a multi-UGV Hamiltonian trajectory planning algorithm is involved to design the trajectory of the UGVs. After the data has been transmitted from the CHs to UGVs, the UAVs are used to collect data from the UGVs and transmit the data to the remote base station. Correspondingly, we propose a multi-agent deep reinforcement learning algorithm using multi-agent deep deterministic policy gradient to design the trajectory of the UAVs. The simulation results demonstrate the effectiveness and advantages of our proposed hybrid UAV-UGV strategy and the ML-based algorithm.
{"title":"Trajectory Optimization for Data Collection in Hybrid UAV and UGV Low-Altitude Economy Network","authors":"Tao Huang;Gengyuan Lu;Mingan Luan;Zheng Chang;Ying-Chang Liang","doi":"10.1109/TCCN.2026.3661515","DOIUrl":"10.1109/TCCN.2026.3661515","url":null,"abstract":"The rapid evolution of unmanned aerial vehicle (UAV) technology has given rise to the new concept of the “low-altitude economy (LAE)” and promoted innovation in various application fields. However, the efficiency of low-altitude services is still significantly constrained by the energy consumption and payload limitations of UAVs. Accordingly, in this paper, we propose a hybrid UAV and unmanned ground vehicle (UGV) LAE framework to address the limitation in UAV-based LAE networks. Specifically, in virtue of the advantages of UGVs in advanced communication, large batteries, and computation modules, the hybrid mode can provide an additional degree of freedom for environmental information collection in LAE networks. Nevertheless, for the considered hybrid framework, which involves the collaborative optimization of numerous agents across multiple dimensions, traditional methods are inefficient and difficult to implement effectively. Hence, we present a machine learning (ML)-based method to handle this issue. More in detail, we first design an iterative self-organizing data analysis techniques strategy to form sensing nodes (SNs) clusters, where the cluster heads (CHs) collect data from SNs in advance in order to reduce the transmission delay. Then, a multi-UGV Hamiltonian trajectory planning algorithm is involved to design the trajectory of the UGVs. After the data has been transmitted from the CHs to UGVs, the UAVs are used to collect data from the UGVs and transmit the data to the remote base station. Correspondingly, we propose a multi-agent deep reinforcement learning algorithm using multi-agent deep deterministic policy gradient to design the trajectory of the UAVs. The simulation results demonstrate the effectiveness and advantages of our proposed hybrid UAV-UGV strategy and the ML-based algorithm.","PeriodicalId":13069,"journal":{"name":"IEEE Transactions on Cognitive Communications and Networking","volume":"12 ","pages":"6031-6044"},"PeriodicalIF":7.0,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134236","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/tccn.2026.3661494
Kenneth L Witham, Nishanth Marer Prabhu, Marius Necsoiu, Chad Spooner, Gunar Schirner
{"title":"TACM-MR: Topographically-Augmented Channel Model Multi-Receiver Dataset","authors":"Kenneth L Witham, Nishanth Marer Prabhu, Marius Necsoiu, Chad Spooner, Gunar Schirner","doi":"10.1109/tccn.2026.3661494","DOIUrl":"https://doi.org/10.1109/tccn.2026.3661494","url":null,"abstract":"","PeriodicalId":13069,"journal":{"name":"IEEE Transactions on Cognitive Communications and Networking","volume":"91 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134235","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/TCCN.2026.3661503
Pengchao Han;Zhenshuai Yin;Chang Liu;Yi Fang
Federated learning (FL) has emerged as a key enabler of collaborative intelligence in wireless edge networks by allowing distributed clients to train models without sharing raw data. However, FL faces significant challenges, including high communication costs, a rigid framework, and privacy concerns, as it shares local model parameters with a homogeneous architecture across clients. Federated knowledge distillation (FedKD) addresses these issues by allowing collaboration among clients with black-box heterogeneous model architectures. However, FedKD often suffers from limited model performance due to the restricted scope of shared knowledge, i.e., model outputs. In this paper, we propose a novel Federated feature distillation (FFD) framework with a guaranteed convergence bound to enhance FedKD by enabling knowledge distillation among clients’ intermediate features. However, transmitting intermediate features and performing feature distillation introduce additional communication and computation overheads, which is difficult to optimize due to the dynamic and stochastic characteristics of model training. To address this, we embrace agentic artificial intelligence (AI) paradigm and propose an actor-critic reinforcement learning algorithm that adaptively selects and weights features for distillation. Extensive experiments across various datasets demonstrate that our proposed algorithm achieves superior model accuracy while significantly reducing resource costs compared to baseline approaches, highlighting the potential of integrating agentic AI with FL to enable efficient and intelligent collaboration in wireless edge networks.
{"title":"Agentic AI-Driven Federated Feature Distillation for Adaptive Resource–Performance Tradeoffs in Wireless Edge Networks","authors":"Pengchao Han;Zhenshuai Yin;Chang Liu;Yi Fang","doi":"10.1109/TCCN.2026.3661503","DOIUrl":"10.1109/TCCN.2026.3661503","url":null,"abstract":"Federated learning (FL) has emerged as a key enabler of collaborative intelligence in wireless edge networks by allowing distributed clients to train models without sharing raw data. However, FL faces significant challenges, including high communication costs, a rigid framework, and privacy concerns, as it shares local model parameters with a homogeneous architecture across clients. Federated knowledge distillation (FedKD) addresses these issues by allowing collaboration among clients with black-box heterogeneous model architectures. However, FedKD often suffers from limited model performance due to the restricted scope of shared knowledge, i.e., model outputs. In this paper, we propose a novel Federated feature distillation (FFD) framework with a guaranteed convergence bound to enhance FedKD by enabling knowledge distillation among clients’ intermediate features. However, transmitting intermediate features and performing feature distillation introduce additional communication and computation overheads, which is difficult to optimize due to the dynamic and stochastic characteristics of model training. To address this, we embrace agentic artificial intelligence (AI) paradigm and propose an actor-critic reinforcement learning algorithm that adaptively selects and weights features for distillation. Extensive experiments across various datasets demonstrate that our proposed algorithm achieves superior model accuracy while significantly reducing resource costs compared to baseline approaches, highlighting the potential of integrating agentic AI with FL to enable efficient and intelligent collaboration in wireless edge networks.","PeriodicalId":13069,"journal":{"name":"IEEE Transactions on Cognitive Communications and Networking","volume":"12 ","pages":"6076-6088"},"PeriodicalIF":7.0,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134230","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-06DOI: 10.1109/TCCN.2026.3661499
Qi Qi;Yan Lei;Yixin Li;Hongyi Li;Qi Zhou
The rapid evolution of Intelligent Connected Vehicles (ICVs) has transformed automobiles into data-rich edge nodes, enabled by progress in artificial intelligence (AI), federated learning (FL), and 5G-enabled communication. While FL facilitates privacy-preserving model training by retaining data locally, its practical deployment in ICVs faces two key challenges: vulnerability to malicious parameter updates and degraded convergence under non-independent and identically distributed (non-IID) data. To address these issues, this paper proposes a secure and efficient cluster-based blockchain scheme for Byzantine resilience and model synchronization of FL. The scheme introduces a mainchain for periodic global model aggregation and multiple subchains that coordinate localized training. These subchains dynamically group vehicles into clusters based on the similarity of their model parameters, which is quantified using the Wasserstein distance. This clustering approach effectively handles non-IID data by ensuring that similar nodes train together. Each subchain employs a hybrid consensus protocol, combining stake-weighted validator election with the Byzantine fault-tolerant HotStuff BFT consensus, to robustly validate local models. Malicious updates are filtered using a combination of distance-based aggregation and gradient thresholding. Furthermore, a dynamic reputation mechanism incentivizes reliable participation through token staking and behavior-based rewards. Extensive experiments on MNIST and CIFAR-10 datasets demonstrate our scheme’s superiority over traditional FL methods, particularly in mitigating the impacts of non-IID data and Byzantine attacks.
{"title":"CPFL: A Cluster-Based Parallel Blockchain Scheme for Secure and Efficient Federated Learning in Intelligent Connected Vehicles","authors":"Qi Qi;Yan Lei;Yixin Li;Hongyi Li;Qi Zhou","doi":"10.1109/TCCN.2026.3661499","DOIUrl":"10.1109/TCCN.2026.3661499","url":null,"abstract":"The rapid evolution of Intelligent Connected Vehicles (ICVs) has transformed automobiles into data-rich edge nodes, enabled by progress in artificial intelligence (AI), federated learning (FL), and 5G-enabled communication. While FL facilitates privacy-preserving model training by retaining data locally, its practical deployment in ICVs faces two key challenges: vulnerability to malicious parameter updates and degraded convergence under non-independent and identically distributed (non-IID) data. To address these issues, this paper proposes a secure and efficient cluster-based blockchain scheme for Byzantine resilience and model synchronization of FL. The scheme introduces a mainchain for periodic global model aggregation and multiple subchains that coordinate localized training. These subchains dynamically group vehicles into clusters based on the similarity of their model parameters, which is quantified using the Wasserstein distance. This clustering approach effectively handles non-IID data by ensuring that similar nodes train together. Each subchain employs a hybrid consensus protocol, combining stake-weighted validator election with the Byzantine fault-tolerant HotStuff BFT consensus, to robustly validate local models. Malicious updates are filtered using a combination of distance-based aggregation and gradient thresholding. Furthermore, a dynamic reputation mechanism incentivizes reliable participation through token staking and behavior-based rewards. Extensive experiments on MNIST and CIFAR-10 datasets demonstrate our scheme’s superiority over traditional FL methods, particularly in mitigating the impacts of non-IID data and Byzantine attacks.","PeriodicalId":13069,"journal":{"name":"IEEE Transactions on Cognitive Communications and Networking","volume":"12 ","pages":"5966-5982"},"PeriodicalIF":7.0,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146134234","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：