Pub Date : 2026-01-02DOI: 10.1109/TSIPN.2025.3642225
Sivaram Krishnan;Jinho Choi;Jihong Park
A wide variety of real-world data, such as sea measurements, e.g., temperatures collected by distributed sensors and multiple unmanned aerial vehicles (UAV) trajectories, can be naturally represented as graphs, often exhibiting non-Euclidean structures. These graph representations may evolve over time, forming time-varying graphs. Effectively modeling and analyzing such dynamic graph data is critical for tasks like predicting graph evolution and reconstructing missing graph data. In this paper, we propose a framework based on the Koopman autoencoder (KAE) to handle time-varying graph data. Specifically, we assume the existence of a hidden non-linear dynamical system, where the state vector corresponds to the graph embedding of the time-varying graph signals. To capture the evolving graph structures, the graph data is first converted into a vector time series through graph embedding, representing the structural information in a finite-dimensional latent space. In this latent space, the KAE is applied to learn the underlying non-linear dynamics governing the temporal evolution of graph features, enabling both prediction and reconstruction tasks.
{"title":"Learning Time-Varying Graph Signals via Koopman","authors":"Sivaram Krishnan;Jinho Choi;Jihong Park","doi":"10.1109/TSIPN.2025.3642225","DOIUrl":"https://doi.org/10.1109/TSIPN.2025.3642225","url":null,"abstract":"A wide variety of real-world data, such as sea measurements, e.g., temperatures collected by distributed sensors and multiple unmanned aerial vehicles (UAV) trajectories, can be naturally represented as graphs, often exhibiting non-Euclidean structures. These graph representations may evolve over time, forming time-varying graphs. Effectively modeling and analyzing such dynamic graph data is critical for tasks like predicting graph evolution and reconstructing missing graph data. In this paper, we propose a framework based on the Koopman autoencoder (KAE) to handle time-varying graph data. Specifically, we assume the existence of a hidden non-linear dynamical system, where the state vector corresponds to the graph embedding of the time-varying graph signals. To capture the evolving graph structures, the graph data is first converted into a vector time series through graph embedding, representing the structural information in a finite-dimensional latent space. In this latent space, the KAE is applied to learn the underlying non-linear dynamics governing the temporal evolution of graph features, enabling both prediction and reconstruction tasks.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"12 ","pages":"16-30"},"PeriodicalIF":3.0,"publicationDate":"2026-01-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145904321","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2026-01-01DOI: 10.1109/TSIPN.2025.3650361
Chongyi Cui;Hong Sang;Ying Zhao;Peng Wang;Shuanghe Yu;Georgi M. Dimirovski
This paper investigates distributed $ell _{infty }$ filtering problem for discrete-time switched delayed systems (DTSDSs) in sensor networks (SNs) with dynamic event-triggered communication. Given that the presence of attacks can compromise the integrity and availability of data, as well as the critical role of switching signals in shaping the behavior of switched systems, a novel event-driven distributed filter is explored in scenarios where the switching signal of the controller experiences a denial-of-service (DoS) attack, characterized by bounded attack frequency and duration. It is significant to mention that the persistent and recurrent nature of attacks compromises the transmission of switching signals, resulting in significant asynchronous discrepancies between the switching of the filtering error system (FES) and the controller. To address the asynchronous behavior induced by DoS attacks, a piecewise time-dependent Lyapunov-Krasovskii functional (PTLKF) tailored to the characteristics of DTSDSs is proposed. Subsequently, sufficient conditions with reduced conservatism are formulated to ensure the exponential stability of the FES, while also guaranteeing an enhanced $ell _{infty }$ disturbance attenuation performance. Finally, two simulation examples are provided to exemplify the superiority and applicability of the proposed filtering technique.
{"title":"Distributed Event-Driven $ {ell }_infty$ Filtering in Switched Delayed Systems Over Sensor Networks Against Switching Signal Attacks","authors":"Chongyi Cui;Hong Sang;Ying Zhao;Peng Wang;Shuanghe Yu;Georgi M. Dimirovski","doi":"10.1109/TSIPN.2025.3650361","DOIUrl":"https://doi.org/10.1109/TSIPN.2025.3650361","url":null,"abstract":"This paper investigates distributed <inline-formula><tex-math>$ell _{infty }$</tex-math></inline-formula> filtering problem for discrete-time switched delayed systems (DTSDSs) in sensor networks (SNs) with dynamic event-triggered communication. Given that the presence of attacks can compromise the integrity and availability of data, as well as the critical role of switching signals in shaping the behavior of switched systems, a novel event-driven distributed filter is explored in scenarios where the switching signal of the controller experiences a denial-of-service (DoS) attack, characterized by bounded attack frequency and duration. It is significant to mention that the persistent and recurrent nature of attacks compromises the transmission of switching signals, resulting in significant asynchronous discrepancies between the switching of the filtering error system (FES) and the controller. To address the asynchronous behavior induced by DoS attacks, a piecewise time-dependent Lyapunov-Krasovskii functional (PTLKF) tailored to the characteristics of DTSDSs is proposed. Subsequently, sufficient conditions with reduced conservatism are formulated to ensure the exponential stability of the FES, while also guaranteeing an enhanced <inline-formula><tex-math>$ell _{infty }$</tex-math></inline-formula> disturbance attenuation performance. Finally, two simulation examples are provided to exemplify the superiority and applicability of the proposed filtering technique.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"12 ","pages":"70-84"},"PeriodicalIF":3.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145982279","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2026-01-01DOI: 10.1109/TSIPN.2025.3650363
Sheng Zhang;Yishu Peng;Hongyu Han;Hing Cheung So
In this paper, we devise an enhanced diffusion LMS algorithm tailored for quantization-based communication in distributed networks. Departing from conventional diffusion approaches, the proposed algorithm, called EQ-DLMS, integrates four distinct steps: (i) weight update, (ii) quantization, (iii) modified weight combination, and (iv) moving average. Through mean-square error analysis, we show how the modified combination and moving average steps impact the steady-state error bound. Notably, without adjusting the quantizer precision, the steady-state error bound avoids the typical $O(mu ^{-1})$ dependence, where $mu$ represents the step-size. However, the EQ-DLMS introduces an additional term, $O(Vert mathbf {w}^{*}Vert ^{2})$, into the error bound, where $mathbf {w}^{*}$ denotes the optimal network parameter vector. To mitigate this, we then develop an improved version of the algorithm, termed DEQ-DLMS, which employs differential quantization while preserving the modified weight combination and moving average steps. Furthermore, we extend the EQ-DLMS update mechanism to address privacy concerns. This leads to the development of an enhanced privacy-aware diffusion LMS algorithm, accompanied by a mean-square stability analysis under non-zero mean protection noise. Finally, simulations are conducted to demonstrate the effectiveness of the proposed approaches and corroborate our theoretical derivations.
{"title":"New Diffusion Least-Mean-Squares Algorithms With Quantization and Privacy Awareness","authors":"Sheng Zhang;Yishu Peng;Hongyu Han;Hing Cheung So","doi":"10.1109/TSIPN.2025.3650363","DOIUrl":"https://doi.org/10.1109/TSIPN.2025.3650363","url":null,"abstract":"In this paper, we devise an enhanced diffusion LMS algorithm tailored for quantization-based communication in distributed networks. Departing from conventional diffusion approaches, the proposed algorithm, called EQ-DLMS, integrates four distinct steps: (i) weight update, (ii) quantization, (iii) modified weight combination, and (iv) moving average. Through mean-square error analysis, we show how the modified combination and moving average steps impact the steady-state error bound. Notably, without adjusting the quantizer precision, the steady-state error bound avoids the typical <inline-formula><tex-math>$O(mu ^{-1})$</tex-math></inline-formula> dependence, where <inline-formula><tex-math>$mu$</tex-math></inline-formula> represents the step-size. However, the EQ-DLMS introduces an additional term, <inline-formula><tex-math>$O(Vert mathbf {w}^{*}Vert ^{2})$</tex-math></inline-formula>, into the error bound, where <inline-formula><tex-math>$mathbf {w}^{*}$</tex-math></inline-formula> denotes the optimal network parameter vector. To mitigate this, we then develop an improved version of the algorithm, termed DEQ-DLMS, which employs differential quantization while preserving the modified weight combination and moving average steps. Furthermore, we extend the EQ-DLMS update mechanism to address privacy concerns. This leads to the development of an enhanced privacy-aware diffusion LMS algorithm, accompanied by a mean-square stability analysis under non-zero mean protection noise. Finally, simulations are conducted to demonstrate the effectiveness of the proposed approaches and corroborate our theoretical derivations.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"12 ","pages":"56-69"},"PeriodicalIF":3.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145982314","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2025-12-24DOI: 10.1109/TSIPN.2025.3648285
Weihua Chen;Zonglin Xie;Feng Liu;Ruipeng Gao
Transmission scheduling plays a critical role in energy conservation in wireless sensor networks (WSNs), particularly in mobile health (mHealth) systems that rely on multiple distributed sensing modalities. Although recent studies have proposed approaches to balance transmission efficiency and timeliness such as periodic sleep scheduling to reduce power consumption, these strategies often result in data loss which can severely degrade real-time diagnostic accuracy. To address this issue, this paper integrates transmission scheduling with data imputation. We propose an energy-efficient software-hardware co-designed framework for mHealth systems, and investigate a Wasserstein Generative Adversarial Imputation Network (WGAIN) to recover missing data. Specifically, the WGAIN model captures heterogeneous inter-sensor correlations, temporal dependencies, and missing - data patterns through a divide-and-conquer learning strategy. Furthermore, we incorporate a dropout-based uncertainty approximation method into the imputation framework and demonstrate its theoretical equivalence to Gaussian processes under variational inference. In addition, a reinforcement learning -based algorithm is developed to dynamically schedule transmissions across heterogeneous sensing modules, with the objective of minimizing overall uncertainty at a target service time. Extensive experiments conducted on the MIT-BIH dataset, together with evaluations on a real-world system prototype, have demonstrated that our approach consistently outperforms existing methods.
{"title":"Energy-Efficient Transmission Scheduling With Uncertainty-Aware Data Imputation for Mhealth","authors":"Weihua Chen;Zonglin Xie;Feng Liu;Ruipeng Gao","doi":"10.1109/TSIPN.2025.3648285","DOIUrl":"https://doi.org/10.1109/TSIPN.2025.3648285","url":null,"abstract":"Transmission scheduling plays a critical role in energy conservation in wireless sensor networks (WSNs), particularly in mobile health (mHealth) systems that rely on multiple distributed sensing modalities. Although recent studies have proposed approaches to balance transmission efficiency and timeliness such as periodic sleep scheduling to reduce power consumption, these strategies often result in data loss which can severely degrade real-time diagnostic accuracy. To address this issue, this paper integrates transmission scheduling with data imputation. We propose an energy-efficient software-hardware co-designed framework for mHealth systems, and investigate a Wasserstein Generative Adversarial Imputation Network (WGAIN) to recover missing data. Specifically, the WGAIN model captures heterogeneous inter-sensor correlations, temporal dependencies, and missing - data patterns through a divide-and-conquer learning strategy. Furthermore, we incorporate a dropout-based uncertainty approximation method into the imputation framework and demonstrate its theoretical equivalence to Gaussian processes under variational inference. In addition, a reinforcement learning -based algorithm is developed to dynamically schedule transmissions across heterogeneous sensing modules, with the objective of minimizing overall uncertainty at a target service time. Extensive experiments conducted on the MIT-BIH dataset, together with evaluations on a real-world system prototype, have demonstrated that our approach consistently outperforms existing methods.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"12 ","pages":"1-15"},"PeriodicalIF":3.0,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145904346","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper is concerned with filtering problem related to multi-rate systems in sensor networks suffering from denial of service attacks under binary encoding scheme. Binary encoding scheme is used for scheduling the transmission of innovation between sensor nodes due to limited bandwidth. Random bit error is considered in order to reflect the existence of binary bit flipping during actual channel transmission. A switching-model approach is adopted to convert the multi-rate systems into the single-rate ones. Stochastic nonlinearity is characterized by statistical properties to enhance generality. Different occurrence probabilities of attacks in distinct channels are characterized by virtue of a group of random variables following the Bernoulli distribution. The objective of the addressed filtering issue is to design a distributed filter such that the filtering error dynamics is stochastically finite-time bounded and satisfies $H_{infty }$ performance requirement. Sufficient conditions guaranteeing the satisfaction of specified filtering performance are established with the assistance of matrix inequalities and stochastic analysis techniques. The gain parameters of the distributed filter are determined by solving certain linear matrix inequalities. Simulation outcomes validate the efficacy of the proposed filtering method.
{"title":"Finite-Time Distributed Filtering for Multi-Rate Nonlinear Systems Suffering From Denial-of-Service Attacks: A Binary Encoding Scheme","authors":"Weijian Ren;Mengdi Chang;Fengcai Huo;Chaohai Kang;Lu Ren","doi":"10.1109/TSIPN.2025.3648309","DOIUrl":"https://doi.org/10.1109/TSIPN.2025.3648309","url":null,"abstract":"This paper is concerned with filtering problem related to multi-rate systems in sensor networks suffering from denial of service attacks under binary encoding scheme. Binary encoding scheme is used for scheduling the transmission of innovation between sensor nodes due to limited bandwidth. Random bit error is considered in order to reflect the existence of binary bit flipping during actual channel transmission. A switching-model approach is adopted to convert the multi-rate systems into the single-rate ones. Stochastic nonlinearity is characterized by statistical properties to enhance generality. Different occurrence probabilities of attacks in distinct channels are characterized by virtue of a group of random variables following the Bernoulli distribution. The objective of the addressed filtering issue is to design a distributed filter such that the filtering error dynamics is stochastically finite-time bounded and satisfies <inline-formula><tex-math>$H_{infty }$</tex-math></inline-formula> performance requirement. Sufficient conditions guaranteeing the satisfaction of specified filtering performance are established with the assistance of matrix inequalities and stochastic analysis techniques. The gain parameters of the distributed filter are determined by solving certain linear matrix inequalities. Simulation outcomes validate the efficacy of the proposed filtering method.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"12 ","pages":"42-55"},"PeriodicalIF":3.0,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145982259","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2025-12-24DOI: 10.1109/TSIPN.2025.3648322
Yi Hua;Zhangfa Wu;Hongping Gan
Streaming graph signal (GS) estimation is common in various network systems. Several graph filter algorithms have been proposed for streaming GS estimation, but they still fail to reach optimal levels. To achieve optimal performance in both estimation accuracy and convergence rate, this paper adopts the recursive least squares (RLS) method in processing GS. When the RLS algorithm is directly combined with GS, its recursive mechanism causes the estimation performance to experience severe degradation. To address this issue, a graph RLS with non-cooperation algorithm and a distributed graph diffusion RLS (DRLS) algorithm, both following the fully recursive structure of the standard RLS, are proposed first. By analyzing these two algorithms, it is found that streaming GS and graph topology are complex and variable, so the previous recursive mechanism is not suitable. Therefore, a dynamic adaptive recursive mechanism is designed, and based on this, a distributed graph improved DRLS (IDRLS) algorithm is proposed. Convergence analysis confirms that the proposed algorithm achieves mean stability and mean-square convergence at a linear rate. Furthermore, we thoroughly examine the causes of performance degradation and demonstrate the superiority of the distributed graph IDRLS algorithm. Finally, experiments, conducted on two different graphs with different levels of sparsity and real-world dataset, verify that the proposed graph IDRLS algorithm can achieve the superior estimation performance and convergence rate and be more effective than the related graph algorithms.
{"title":"Improved Diffusion Recursive Least Squares for Graph Signal Estimation on Distributed Network","authors":"Yi Hua;Zhangfa Wu;Hongping Gan","doi":"10.1109/TSIPN.2025.3648322","DOIUrl":"https://doi.org/10.1109/TSIPN.2025.3648322","url":null,"abstract":"Streaming graph signal (GS) estimation is common in various network systems. Several graph filter algorithms have been proposed for streaming GS estimation, but they still fail to reach optimal levels. To achieve optimal performance in both estimation accuracy and convergence rate, this paper adopts the recursive least squares (RLS) method in processing GS. When the RLS algorithm is directly combined with GS, its recursive mechanism causes the estimation performance to experience severe degradation. To address this issue, a graph RLS with non-cooperation algorithm and a distributed graph diffusion RLS (DRLS) algorithm, both following the fully recursive structure of the standard RLS, are proposed first. By analyzing these two algorithms, it is found that streaming GS and graph topology are complex and variable, so the previous recursive mechanism is not suitable. Therefore, a dynamic adaptive recursive mechanism is designed, and based on this, a distributed graph improved DRLS (IDRLS) algorithm is proposed. Convergence analysis confirms that the proposed algorithm achieves mean stability and mean-square convergence at a linear rate. Furthermore, we thoroughly examine the causes of performance degradation and demonstrate the superiority of the distributed graph IDRLS algorithm. Finally, experiments, conducted on two different graphs with different levels of sparsity and real-world dataset, verify that the proposed graph IDRLS algorithm can achieve the superior estimation performance and convergence rate and be more effective than the related graph algorithms.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"12 ","pages":"31-41"},"PeriodicalIF":3.0,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145904280","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2025-12-23DOI: 10.1109/TSIPN.2025.3642234
Chang Liu;Zeyi Liu;Hongjing Liang;Md Altab Hossin
In this paper, a full-agent connectivity-preserving control strategy for multi-agent systems under multi-frequency deception attacks is introduced, avoiding the singularity problem caused by the initial position of the agent. This policy eliminates the need to screen agents during the initial phase and ensures the continuous presence of all agents within the communication boundaries at all times. In addition, this study considers the communication dynamics between leader and followers affected by boundaries and proposes a connectivity-preserving strategy that takes into account the full agent population. To effectively characterize the characteristics of multi-frequency attacks in practice, a time function incorporating both frequency and period information of deception attacks has been developed. This function serves to encapsulate the intentions of multi-frequency deception attackers. An extended state observer is employed to monitor and mitigate the instability caused by deception attacks. The final control framework ensures the convergence of tracking errors and the stability of the state signals.
{"title":"Full-Agent Connectivity-Preserving Secure Strategy Under Multi-Frequency Deception Attacks","authors":"Chang Liu;Zeyi Liu;Hongjing Liang;Md Altab Hossin","doi":"10.1109/TSIPN.2025.3642234","DOIUrl":"https://doi.org/10.1109/TSIPN.2025.3642234","url":null,"abstract":"In this paper, a full-agent connectivity-preserving control strategy for multi-agent systems under multi-frequency deception attacks is introduced, avoiding the singularity problem caused by the initial position of the agent. This policy eliminates the need to screen agents during the initial phase and ensures the continuous presence of all agents within the communication boundaries at all times. In addition, this study considers the communication dynamics between leader and followers affected by boundaries and proposes a connectivity-preserving strategy that takes into account the full agent population. To effectively characterize the characteristics of multi-frequency attacks in practice, a time function incorporating both frequency and period information of deception attacks has been developed. This function serves to encapsulate the intentions of multi-frequency deception attackers. An extended state observer is employed to monitor and mitigate the instability caused by deception attacks. The final control framework ensures the convergence of tracking errors and the stability of the state signals.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"11 ","pages":"1605-1618"},"PeriodicalIF":3.0,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145830842","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2025-12-23DOI: 10.1109/TSIPN.2025.3647210
Huali Zhu;Hua Xu;Yunhao Shi;Wanyi Gu;Xin Jia
Automatic Modulation recognition (AMR) is essential for intelligent communication receivers, with broad applications in civilian and military contexts. Deep Learning (DL) enhances recognition accuracy with high-quality, well-labeled datasets, but struggles with poorly labeled datasets or incomplete signals. To address this, we propose a semi-supervised learning approach using a $p$-Laplacian Graph Convolutional Network (GpCN) for AMR, which enhances the feature extraction capabilities by using of $p$-order convolution kernels of GCN. It is built upon the simple signal graph based on Transformer mask mechanism, which prioritize sampling points by Transformer's weight distribution. And a semi-supervised loss function reconstructed by Transformer feature reconstruction. This approach consistently yields a recognition rate of 50% with just 1% of labels on RML2016.10a dataset, outperforming the fully supervised recognition rates of existing methods. Similarly, applying the TMGpCN to a more complex dataset RML2018.01a (SNR = [−10, 10]), still achieves good performance under low-label conditions. With only 1% labeled data, the recognition accuracy for 24 types of signals reached 42.61%, which is only 8.86% lower than full supervision.
{"title":"TMGpCN: Semi-Supervised GpCN Based on Transformer Mask for Automatic Modulation Recognition","authors":"Huali Zhu;Hua Xu;Yunhao Shi;Wanyi Gu;Xin Jia","doi":"10.1109/TSIPN.2025.3647210","DOIUrl":"https://doi.org/10.1109/TSIPN.2025.3647210","url":null,"abstract":"Automatic Modulation recognition (AMR) is essential for intelligent communication receivers, with broad applications in civilian and military contexts. Deep Learning (DL) enhances recognition accuracy with high-quality, well-labeled datasets, but struggles with poorly labeled datasets or incomplete signals. To address this, we propose a semi-supervised learning approach using a <inline-formula><tex-math>$p$</tex-math></inline-formula>-Laplacian Graph Convolutional Network (GpCN) for AMR, which enhances the feature extraction capabilities by using of <inline-formula><tex-math>$p$</tex-math></inline-formula>-order convolution kernels of GCN. It is built upon the simple signal graph based on Transformer mask mechanism, which prioritize sampling points by Transformer's weight distribution. And a semi-supervised loss function reconstructed by Transformer feature reconstruction. This approach consistently yields a recognition rate of 50% with just 1% of labels on RML2016.10a dataset, outperforming the fully supervised recognition rates of existing methods. Similarly, applying the TMGpCN to a more complex dataset RML2018.01a (SNR = [−10, 10]), still achieves good performance under low-label conditions. With only 1% labeled data, the recognition accuracy for 24 types of signals reached 42.61%, which is only 8.86% lower than full supervision.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"12 ","pages":"85-97"},"PeriodicalIF":3.0,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145982302","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
From mass media theory, the message flow and opinion evolution in real social networks adhere to a two-step communication process: opinion leaders first receive the messages from the message sources, and then convey their opinions to the normal agents. During this process, opinion leaders often exhibit confirmation bias by updating their opinions based on messages close to their opinions, reflecting their limited willingness to accept dissonant messages. However, there are limited prior works jointly considering the two-step communication process involving message sources, opinion leaders, and normal agents, as well as the confirmation bias of opinion leaders. In this paper, we propose a unified framework called MOLENA to analyze how messages flow and opinions evolve in the two-step communication process. In the MOLENA framework, we introduce a mathematically tractable message preference model to quantitatively describe the confirmation bias. We obtain the approximate analytical solutions of the opinion leaders' and normal agents' steady-state opinions, and theoretically analyze the influence of system parameters on their steady-state opinions. Furthermore, we examine the influence of messages on opinion leaders' steady-state opinions and study how opinion leaders influence normal agents' steady-state opinions in the two-step communication process. Finally, we validate our theoretical analysis through numerical experiments and verify the correctness of the MOLENA framework via social experiments. This study is critical to understanding how messages flow and how opinions form and evolve in real social networks, and to designing effective mechanisms to guide agents' opinions.
{"title":"MOLENA: Analyzing the Two-Step Message Flow and Opinion Evolution Process in Social Networks","authors":"Huisheng Wang;Yuejiang Li;Yiqing Lin;H. Vicky Zhao","doi":"10.1109/TSIPN.2025.3646301","DOIUrl":"https://doi.org/10.1109/TSIPN.2025.3646301","url":null,"abstract":"From mass media theory, the message flow and opinion evolution in real social networks adhere to a two-step communication process: opinion leaders first receive the messages from the message sources, and then convey their opinions to the normal agents. During this process, opinion leaders often exhibit confirmation bias by updating their opinions based on messages close to their opinions, reflecting their limited willingness to accept dissonant messages. However, there are limited prior works jointly considering the two-step communication process involving message sources, opinion leaders, and normal agents, as well as the confirmation bias of opinion leaders. In this paper, we propose a unified framework called MOLENA to analyze how messages flow and opinions evolve in the two-step communication process. In the MOLENA framework, we introduce a mathematically tractable message preference model to quantitatively describe the confirmation bias. We obtain the approximate analytical solutions of the opinion leaders' and normal agents' steady-state opinions, and theoretically analyze the influence of system parameters on their steady-state opinions. Furthermore, we examine the influence of messages on opinion leaders' steady-state opinions and study how opinion leaders influence normal agents' steady-state opinions in the two-step communication process. Finally, we validate our theoretical analysis through numerical experiments and verify the correctness of the MOLENA framework via social experiments. This study is critical to understanding how messages flow and how opinions form and evolve in real social networks, and to designing effective mechanisms to guide agents' opinions.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"12 ","pages":"212-227"},"PeriodicalIF":3.0,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175743","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Statistical models of inter-point distances are pivotal for analyzing and optimizing wireless communication networks and other spatial systems, such as vehicular swarms and distributed sensing networks. However, the analytical intractability of exact distance distributions often hinders closed-form performance evaluations and obscures parameter-performance relationships. To address these challenges, this paper introduces a low-complexity polynomial substitute for inter-point distance distributions and a systematic framework for parameter mapping. The framework employs two complementary mapping schemes, Relative Entropy Minimization (REM) which promotes fidelity to the original distribution in the Kullback–Leibler sense, and Mean Square Error Minimization (MSEM) which minimizes the mean squared error between the two distributions. These mappings yield parameter correspondences between the original and substitute distributions, enabling efficient and accurate approximations. The substitutes are validated on representative spatial models, preserving fidelity to the original distributions while using a low-complexity polynomial representation. This advancement facilitates closed-form evaluations and optimizations in random networks, enhancing the analytical toolkit for stochastic geometry and control theory.
{"title":"Parameter Mapping of Distribution Substitution for Inter-Point Distances in Random Networks","authors":"Shuping Dang;Jia Ye;Shuaishuai Guo;Raed Shubair;Marwa Chafii","doi":"10.1109/TSIPN.2025.3642229","DOIUrl":"https://doi.org/10.1109/TSIPN.2025.3642229","url":null,"abstract":"Statistical models of inter-point distances are pivotal for analyzing and optimizing wireless communication networks and other spatial systems, such as vehicular swarms and distributed sensing networks. However, the analytical intractability of exact distance distributions often hinders closed-form performance evaluations and obscures parameter-performance relationships. To address these challenges, this paper introduces a low-complexity polynomial substitute for inter-point distance distributions and a systematic framework for parameter mapping. The framework employs two complementary mapping schemes, Relative Entropy Minimization (REM) which promotes fidelity to the original distribution in the Kullback–Leibler sense, and Mean Square Error Minimization (MSEM) which minimizes the mean squared error between the two distributions. These mappings yield parameter correspondences between the original and substitute distributions, enabling efficient and accurate approximations. The substitutes are validated on representative spatial models, preserving fidelity to the original distributions while using a low-complexity polynomial representation. This advancement facilitates closed-form evaluations and optimizations in random networks, enhancing the analytical toolkit for stochastic geometry and control theory.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"11 ","pages":"1619-1633"},"PeriodicalIF":3.0,"publicationDate":"2025-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886582","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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：