Pub Date : 2026-04-01Epub Date: 2025-11-05DOI: 10.1016/j.sigpro.2025.110380
Tao Cui , Peng Dong , Zhongliang Jing , Kai Shen , Wujun Chen , Baitao Tang
To overcome the limitations of traditional distributed average fusion under measurement uncertainty, this paper proposes a distributed weighted average consensus algorithm. Based on the estimated states of each platform, the Wasserstein distance is utilized to adaptively determine real-time fusion weights. A matrix formulation of the weighted consensus iterations is developed within the alternating direction method of multipliers (ADMM) framework, followed by an analysis of the algorithm’s convergence properties. Simulation results demonstrate that the proposed algorithm significantly improves fusion accuracy in complex environments.
{"title":"Distributed weighted average consensus fusion based on ADMM under measurement uncertainty","authors":"Tao Cui , Peng Dong , Zhongliang Jing , Kai Shen , Wujun Chen , Baitao Tang","doi":"10.1016/j.sigpro.2025.110380","DOIUrl":"10.1016/j.sigpro.2025.110380","url":null,"abstract":"<div><div>To overcome the limitations of traditional distributed average fusion under measurement uncertainty, this paper proposes a distributed weighted average consensus algorithm. Based on the estimated states of each platform, the Wasserstein distance is utilized to adaptively determine real-time fusion weights. A matrix formulation of the weighted consensus iterations is developed within the alternating direction method of multipliers (ADMM) framework, followed by an analysis of the algorithm’s convergence properties. Simulation results demonstrate that the proposed algorithm significantly improves fusion accuracy in complex environments.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"241 ","pages":"Article 110380"},"PeriodicalIF":3.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528936","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2025-11-08DOI: 10.1016/j.sigpro.2025.110381
Haifeng Li, Xinxin Geng
The Lorentzian-based iterative hard thresholding (LIHT) algorithm demonstrates significant superiority over conventional sparse reconstruction techniques in impulsive noise environments. Previous analyses of the algorithm’s guaranteed recovery performance have primarily focused on the restricted isometry property (RIP) of the sensing matrix. In this work, we establish a tighter RIP-based convergence bound for LIHT, refining the required condition to , where represents the restricted isometry constant of order . Furthermore, for scenarios with partial support knowledge (LIHT-PKS), we propose an enhanced bound requiring only , where denotes the number of known support elements. Finally, we present the impact of parameter on the performance of LIHT.
{"title":"Improved RIP-based bounds performance guarantee for sparse signal recovery via Lorentzian iterative hard thresholding","authors":"Haifeng Li, Xinxin Geng","doi":"10.1016/j.sigpro.2025.110381","DOIUrl":"10.1016/j.sigpro.2025.110381","url":null,"abstract":"<div><div>The Lorentzian-based iterative hard thresholding (LIHT) algorithm demonstrates significant superiority over conventional sparse reconstruction techniques in impulsive noise environments. Previous analyses of the algorithm’s guaranteed recovery performance have primarily focused on the restricted isometry property (RIP) of the sensing matrix. In this work, we establish a tighter RIP-based convergence bound for LIHT, refining the required condition to <span><math><mrow><msub><mrow><mi>δ</mi></mrow><mrow><mn>3</mn><mi>s</mi></mrow></msub><mo><</mo><mfrac><mrow><msqrt><mrow><mn>5</mn></mrow></msqrt><mo>−</mo><mn>1</mn></mrow><mrow><mn>2</mn></mrow></mfrac></mrow></math></span>, where <span><math><msub><mrow><mi>δ</mi></mrow><mrow><mn>3</mn><mi>s</mi></mrow></msub></math></span> represents the restricted isometry constant of order <span><math><mrow><mn>3</mn><mi>s</mi></mrow></math></span>. Furthermore, for scenarios with partial support knowledge (LIHT-PKS), we propose an enhanced bound requiring only <span><math><mrow><msub><mrow><mi>δ</mi></mrow><mrow><mn>3</mn><mi>s</mi><mo>−</mo><mn>2</mn><mi>k</mi></mrow></msub><mo><</mo><mfrac><mrow><msqrt><mrow><mn>5</mn></mrow></msqrt><mo>−</mo><mn>1</mn></mrow><mrow><mn>2</mn></mrow></mfrac></mrow></math></span>, where <span><math><mi>k</mi></math></span> denotes the number of known support elements. Finally, we present the impact of parameter <span><math><mi>γ</mi></math></span> on the performance of LIHT.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"241 ","pages":"Article 110381"},"PeriodicalIF":3.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528939","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2025-11-11DOI: 10.1016/j.sigpro.2025.110406
Yuxuan Liu, Yichen Bao, Han Lu, Quanxue Gao
Fuzzy K-Means is a classic clustering method that performs fuzzy partitioning of data by iteratively updating the cluster centers and the membership degrees of each data point. This makes it particularly suitable for handling data with unclear boundaries. However, the algorithm is highly sensitive to the choice of initial cluster centroids, which can affect the stability of the clustering results. To address this issue, we propose a robust fuzzy K-Means clustering algorithm(FKMVC) that eliminate the reliance on cluster centroids, obtaining membership metrices solely through distance matrix computation. Specifically, we reexpress fuzzy K-Means from the perspective of manifold, construct the manifold structure by labels, and then perform clustering update labels on the manifold structure, so that the labels can be obtained without centroid estimation, and the consistency of manifolds and labels is maintained. In addition, our proposed model supports various types of distance matrices to accommodate complex linearly inseparable data. The results from extensive experiments across multiple databases substantiate the superiority of our proposed model.
{"title":"Fuzzy K-means clustering without cluster centroids","authors":"Yuxuan Liu, Yichen Bao, Han Lu, Quanxue Gao","doi":"10.1016/j.sigpro.2025.110406","DOIUrl":"10.1016/j.sigpro.2025.110406","url":null,"abstract":"<div><div>Fuzzy <em>K</em>-Means is a classic clustering method that performs fuzzy partitioning of data by iteratively updating the cluster centers and the membership degrees of each data point. This makes it particularly suitable for handling data with unclear boundaries. However, the algorithm is highly sensitive to the choice of initial cluster centroids, which can affect the stability of the clustering results. To address this issue, we propose a robust fuzzy <em>K</em>-Means clustering algorithm(FKMVC) that eliminate the reliance on cluster centroids, obtaining membership metrices solely through distance matrix computation. Specifically, we reexpress fuzzy <em>K</em>-Means from the perspective of manifold, construct the manifold structure by labels, and then perform clustering update labels on the manifold structure, so that the labels can be obtained without centroid estimation, and the consistency of manifolds and labels is maintained. In addition, our proposed model supports various types of distance matrices to accommodate complex linearly inseparable data. The results from extensive experiments across multiple databases substantiate the superiority of our proposed model.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"241 ","pages":"Article 110406"},"PeriodicalIF":3.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145529419","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2025-10-30DOI: 10.1016/j.sigpro.2025.110377
Jielong Lu , Zhenkai Zhang , Boon-Chong Seet , Baiheng Wang
Dual-functional radar-communication (DFRC) has emerged as an effective solution in recent years to address spectrum scarcity in maritime environments, enabling efficient integrated communication and sensing. To mitigate path loss over complex sea surfaces, the intelligent reflecting surface (IRS) is introduced into DFRC systems, enhancing signal quality by providing an additional propagation path. To address the impact of sea wave fluctuations on the communication channel of maritime vessels, an alternating optimization (AO) algorithm based on semidefinite relaxation and fractional programming (SDR-FP) is proposed. First, the non-ideal channel state information (CSI) is modeled using a bounded channel uncertainty model via the S-procedure. Second, under constraints on radar detection performance and transmit power, the problem is formulated to maximize the communication sum-rate. Next, the proposed AO algorithm decomposes the original high-dimensional problem into two low-complexity subproblems. Finally, a minimization algorithm is applied to reformulate the non-convex subproblem into a tractable quadratically constrained quadratic program (QCQP). Simulation results demonstrate that the proposed method significantly enhances the communication sum-rate while achieving faster convergence compared to benchmarks.
{"title":"IRS-assisted communication performance optimization method for shipborne DFRC system","authors":"Jielong Lu , Zhenkai Zhang , Boon-Chong Seet , Baiheng Wang","doi":"10.1016/j.sigpro.2025.110377","DOIUrl":"10.1016/j.sigpro.2025.110377","url":null,"abstract":"<div><div>Dual-functional radar-communication (DFRC) has emerged as an effective solution in recent years to address spectrum scarcity in maritime environments, enabling efficient integrated communication and sensing. To mitigate path loss over complex sea surfaces, the intelligent reflecting surface (IRS) is introduced into DFRC systems, enhancing signal quality by providing an additional propagation path. To address the impact of sea wave fluctuations on the communication channel of maritime vessels, an alternating optimization (AO) algorithm based on semidefinite relaxation and fractional programming (SDR-FP) is proposed. First, the non-ideal channel state information (CSI) is modeled using a bounded channel uncertainty model via the S-procedure. Second, under constraints on radar detection performance and transmit power, the problem is formulated to maximize the communication sum-rate. Next, the proposed AO algorithm decomposes the original high-dimensional problem into two low-complexity subproblems. Finally, a minimization algorithm is applied to reformulate the non-convex subproblem into a tractable quadratically constrained quadratic program (QCQP). Simulation results demonstrate that the proposed method significantly enhances the communication sum-rate while achieving faster convergence compared to benchmarks.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"241 ","pages":"Article 110377"},"PeriodicalIF":3.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528911","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2025-11-08DOI: 10.1016/j.sigpro.2025.110400
Wei Zhang, Pei Zeng, Bo Ou
Video reversible data hiding (V-RDH) is widely applied in various fields to protect the security and integrity of data. In this paper, a new V-RDH method for high efficiency video coding (HEVC) is proposed by using histogram shifting (HS) and matrix embedding. Unlike previous HS-based algorithms that exhibit arbitrariness in selection for peak and zero bins, we propose a new strategy to trade off the capacity versus the distortion drift. The invertible matrix embedding is designed to improve the embedding efficiency. Our method does not need any side information for reversibility, and the distortion drift can be eliminated by the modifications without causing error propagation. Experimental results demonstrate that compared with the existing well-performing methods, the proposed method can achieve a better visual quality of the marked video with the satisfactory embedding capacity.
{"title":"Video reversible data hiding using histogram shifting and matrix embedding for HEVC","authors":"Wei Zhang, Pei Zeng, Bo Ou","doi":"10.1016/j.sigpro.2025.110400","DOIUrl":"10.1016/j.sigpro.2025.110400","url":null,"abstract":"<div><div>Video reversible data hiding (V-RDH) is widely applied in various fields to protect the security and integrity of data. In this paper, a new V-RDH method for high efficiency video coding (HEVC) is proposed by using histogram shifting (HS) and matrix embedding. Unlike previous HS-based algorithms that exhibit arbitrariness in selection for peak and zero bins, we propose a new strategy to trade off the capacity versus the distortion drift. The invertible matrix embedding is designed to improve the embedding efficiency. Our method does not need any side information for reversibility, and the distortion drift can be eliminated by the modifications without causing error propagation. Experimental results demonstrate that compared with the existing well-performing methods, the proposed method can achieve a better visual quality of the marked video with the satisfactory embedding capacity.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"241 ","pages":"Article 110400"},"PeriodicalIF":3.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528909","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2025-11-11DOI: 10.1016/j.sigpro.2025.110405
Qiang Wang , Lu Lu , Tao Yu , Guangya Zhu
This paper proposes a novel adaptive filtering algorithm, termed the bias-compensated censored regression Euclidean direction search (BC-CR-EDS) algorithm, to address the joint challenges of censored outputs and noisy inputs in censored regression (CR) models. In the CR model, the output data outside the specified range are censored and not measured exactly. The traditional adaptive filtering algorithms may not work effectively in such a model. Furthermore, a CR model with noisy input can lead to a biased estimation of the algorithm. In this scenario, the bias-compensated Heckman (BC-Heckman) algorithm was developed, but its convergence rate and steady-state performance may deteriorate. To surmount this problem, a novel BC-CR-EDS algorithm is proposed. Benefiting from the numerical stability of the EDS algorithm and the unbiasedness principle criterion, the BC-CR-EDS algorithm can achieve improved steady-state performance and tracking performance. As an additional contribution, an online method to estimate the variance of the input data is developed for the BC-CR-EDS algorithm. In addition, the steady-state performance of mean and mean-square for the BC-CR-EDS algorithm is analyzed. Simulation results demonstrate that the BC-CR-EDS algorithm achieves approximately a 10 dB improvement in steady-state performance compared to the existing algorithms for system identification and acoustic echo cancellation.
{"title":"Unbiased censored regression Euclidean direction search algorithm","authors":"Qiang Wang , Lu Lu , Tao Yu , Guangya Zhu","doi":"10.1016/j.sigpro.2025.110405","DOIUrl":"10.1016/j.sigpro.2025.110405","url":null,"abstract":"<div><div>This paper proposes a novel adaptive filtering algorithm, termed the bias-compensated censored regression Euclidean direction search (BC-CR-EDS) algorithm, to address the joint challenges of censored outputs and noisy inputs in censored regression (CR) models. In the CR model, the output data outside the specified range are censored and not measured exactly. The traditional adaptive filtering algorithms may not work effectively in such a model. Furthermore, a CR model with noisy input can lead to a biased estimation of the algorithm. In this scenario, the bias-compensated Heckman (BC-Heckman) algorithm was developed, but its convergence rate and steady-state performance may deteriorate. To surmount this problem, a novel BC-CR-EDS algorithm is proposed. Benefiting from the numerical stability of the EDS algorithm and the unbiasedness principle criterion, the BC-CR-EDS algorithm can achieve improved steady-state performance and tracking performance. As an additional contribution, an online method to estimate the variance of the input data is developed for the BC-CR-EDS algorithm. In addition, the steady-state performance of mean and mean-square for the BC-CR-EDS algorithm is analyzed. Simulation results demonstrate that the BC-CR-EDS algorithm achieves approximately a 10 dB improvement in steady-state performance compared to the existing algorithms for system identification and acoustic echo cancellation.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"241 ","pages":"Article 110405"},"PeriodicalIF":3.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528907","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2025-11-05DOI: 10.1016/j.sigpro.2025.110378
Zhenxia Xue , Yang Yang , Shouhe Lin , Jun Ma
Twin Extreme Learning Machine (TELM) is an outstanding algorithm that has been widely applied various fields due to its excellent performance, which includes a simple structure, few parameters and robust generalization ability. However, TELM employs the L2-norm metric and hinge loss function, which can amplify the impact of noise. In this paper, we replace the L2-norm metric with the capped L-norm metric. Moreover, we propose a novel hook loss function which can effectively handle the impact of noise by adjusting the parameters and . Furthermore, we explore several significant properties of our loss function, such as asymmetry, robustness, nonconvexity, and Fisher consistency. By integrating the hook loss function and capped L-norm into TELM, this paper presents an Adaptive Robust Twin Extreme Learning Machine (ARTELM) which not only inherits the advantages of TELM but also reduces the impact of outliers. To fully exploit the information contained in unlabeled samples, this paper extends ARTELM to a semi-supervised framework and proposes the Laplacian ARTELM (Lap-ARTELM) model. Additionally, two efficient algorithms are proposed and their computational complexity and convergence analysis are provided. Finally, experimental results on multiple datasets demonstrate that the proposed algorithms are competitive compared to several existing state-of-art methods.
{"title":"An adaptive robust twin extreme learning machine with hook loss function and its semi-supervised framework","authors":"Zhenxia Xue , Yang Yang , Shouhe Lin , Jun Ma","doi":"10.1016/j.sigpro.2025.110378","DOIUrl":"10.1016/j.sigpro.2025.110378","url":null,"abstract":"<div><div>Twin Extreme Learning Machine (TELM) is an outstanding algorithm that has been widely applied various fields due to its excellent performance, which includes a simple structure, few parameters and robust generalization ability. However, TELM employs the L<sub>2</sub>-norm metric and hinge loss function, which can amplify the impact of noise. In this paper, we replace the L<sub>2</sub>-norm metric with the capped L<span><math><msub><mrow></mrow><mrow><mn>2</mn><mo>,</mo><mi>p</mi></mrow></msub></math></span>-norm metric. Moreover, we propose a novel hook loss function which can effectively handle the impact of noise by adjusting the parameters <span><math><mi>a</mi></math></span> and <span><math><mi>c</mi></math></span>. Furthermore, we explore several significant properties of our loss function, such as asymmetry, robustness, nonconvexity, and Fisher consistency. By integrating the hook loss function and capped L<span><math><msub><mrow></mrow><mrow><mn>2</mn><mo>,</mo><mi>p</mi></mrow></msub></math></span>-norm into TELM, this paper presents an Adaptive Robust Twin Extreme Learning Machine (ARTELM) which not only inherits the advantages of TELM but also reduces the impact of outliers. To fully exploit the information contained in unlabeled samples, this paper extends ARTELM to a semi-supervised framework and proposes the Laplacian ARTELM (Lap-ARTELM) model. Additionally, two efficient algorithms are proposed and their computational complexity and convergence analysis are provided. Finally, experimental results on multiple datasets demonstrate that the proposed algorithms are competitive compared to several existing state-of-art methods.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"241 ","pages":"Article 110378"},"PeriodicalIF":3.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528914","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2025-11-12DOI: 10.1016/j.sigpro.2025.110408
Bo Zhu , Haoxuan Li , Tao Geng , Wenqiang Duan , Boxin Ren
Accurate wall material recognition is essential for robotic operations in extreme environments such as mining tunnels, disaster sites, and search-and-rescue missions, where conventional sensors like cameras and LiDAR often fail due to darkness, dust, smoke, or obstructions. Ultrasonic sensing offers a robust alternative, but its echo signals exhibit complex spatial–temporal patterns that are difficult to model with traditional methods. This study proposes AE-CS-TCN (Attention-Enhanced Cross-Scale Temporal Convolutional Network), a deep learning architecture for non-contact wall material recognition using raw ultrasonic echoes. The model integrates spatial attention, dilated temporal convolutions, cross-scale fusion, and cross-attention to effectively capture and align multi-resolution features. Experiments on both the public LMT dataset and a self-built dataset show that AE-CS-TCN achieves 96% average accuracy, outperforming conventional and deep learning baselines while maintaining strong robustness to noise and distance variations.
{"title":"A non-contact material recognition method using ultrasonic echo signals and deep learning","authors":"Bo Zhu , Haoxuan Li , Tao Geng , Wenqiang Duan , Boxin Ren","doi":"10.1016/j.sigpro.2025.110408","DOIUrl":"10.1016/j.sigpro.2025.110408","url":null,"abstract":"<div><div>Accurate wall material recognition is essential for robotic operations in extreme environments such as mining tunnels, disaster sites, and search-and-rescue missions, where conventional sensors like cameras and LiDAR often fail due to darkness, dust, smoke, or obstructions. Ultrasonic sensing offers a robust alternative, but its echo signals exhibit complex spatial–temporal patterns that are difficult to model with traditional methods. This study proposes AE-CS-TCN (Attention-Enhanced Cross-Scale Temporal Convolutional Network), a deep learning architecture for non-contact wall material recognition using raw ultrasonic echoes. The model integrates spatial attention, dilated temporal convolutions, cross-scale fusion, and cross-attention to effectively capture and align multi-resolution features. Experiments on both the public LMT dataset and a self-built dataset show that AE-CS-TCN achieves 96% average accuracy, outperforming conventional and deep learning baselines while maintaining strong robustness to noise and distance variations.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"241 ","pages":"Article 110408"},"PeriodicalIF":3.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528940","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2025-11-07DOI: 10.1016/j.sigpro.2025.110387
Ehsan Lari , Reza Arablouei , Vinay Chakravarthi Gogineni , Stefan Werner
Federated learning (FL) leverages client–server communications to train global models on decentralized data. However, communication noise or errors can impair model accuracy. To address this challenge, we propose a novel FL algorithm that enhances robustness against communication noise while also reducing communication load. We derive the proposed algorithm by solving the weighted least-squares (WLS) regression problem, framed as a distributed convex optimization problem over a federated network with random client scheduling, via the alternating direction method of multipliers (ADMM). To counteract the detrimental effects of cumulative communication noise, we introduce a key modification by eliminating the dual variable and implementing a new local model update at each participating client. This subtle yet effective change results in using a single noisy global model update at each client instead of two, improving robustness against additive communication noise. Furthermore, we incorporate another modification enabling clients to continue local updates even when not selected by the server, leading to substantial performance improvements. Our theoretical analysis confirms the convergence of the proposed algorithm in both mean and mean-square senses, even when the server communicates with a random subset of clients over noisy links. Numerical results validate the effectiveness of our algorithm and corroborate theoretical findings.
{"title":"Noise-robust and resource-efficient ADMM-based federated learning for WLS regression","authors":"Ehsan Lari , Reza Arablouei , Vinay Chakravarthi Gogineni , Stefan Werner","doi":"10.1016/j.sigpro.2025.110387","DOIUrl":"10.1016/j.sigpro.2025.110387","url":null,"abstract":"<div><div>Federated learning (FL) leverages client–server communications to train global models on decentralized data. However, communication noise or errors can impair model accuracy. To address this challenge, we propose a novel FL algorithm that enhances robustness against communication noise while also reducing communication load. We derive the proposed algorithm by solving the weighted least-squares (WLS) regression problem, framed as a distributed convex optimization problem over a federated network with random client scheduling, via the alternating direction method of multipliers (ADMM). To counteract the detrimental effects of cumulative communication noise, we introduce a key modification by eliminating the dual variable and implementing a new local model update at each participating client. This subtle yet effective change results in using a single noisy global model update at each client instead of two, improving robustness against additive communication noise. Furthermore, we incorporate another modification enabling clients to continue local updates even when not selected by the server, leading to substantial performance improvements. Our theoretical analysis confirms the convergence of the proposed algorithm in both mean and mean-square senses, even when the server communicates with a random subset of clients over noisy links. Numerical results validate the effectiveness of our algorithm and corroborate theoretical findings.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"241 ","pages":"Article 110387"},"PeriodicalIF":3.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528913","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-04-01Epub Date: 2025-11-11DOI: 10.1016/j.sigpro.2025.110404
Shibo Jin , Lujuan Dang , Badong Chen
The linear minimum mean square error (LMMSE) framework is widely used for state estimation in dynamic systems owing to the robust performance of this approach under Gaussian assumptions. However, in nonlinear estimation problems, the performance of LMMSE is relatively inferior compared to other nonlinear algorithms. The optimized conversion-sample filter (OCF) was developed within the LMMSE framework. This filter obtains final estimates through optimized uncorrelated conversion (UC), while enhancing nonlinear system processing capabilities through integration of the deterministic sampling (DS) method and constrained Rayleigh quotient optimization techniques. Although OCF demonstrates improved performance in nonlinear systems, its effectiveness remains compromised under impulsive noise conditions. To overcome this issue, we propose an optimized conversion-sample filter based on the maximum correntropy criterion (MCOCF). MCOCF integrates the maximum correntropy criterion (MCC), thereby enhancing DS and the constrained Rayleigh quotient to a certain extent. Simulation results indicate that the MCOCF not only improves performance in environments with impulse noise but also significantly enhances the ability to process nonlinear systems.
{"title":"Optimized conversion-sample filter under maximum correntropy criterion","authors":"Shibo Jin , Lujuan Dang , Badong Chen","doi":"10.1016/j.sigpro.2025.110404","DOIUrl":"10.1016/j.sigpro.2025.110404","url":null,"abstract":"<div><div>The linear minimum mean square error (LMMSE) framework is widely used for state estimation in dynamic systems owing to the robust performance of this approach under Gaussian assumptions. However, in nonlinear estimation problems, the performance of LMMSE is relatively inferior compared to other nonlinear algorithms. The optimized conversion-sample filter (OCF) was developed within the LMMSE framework. This filter obtains final estimates through optimized uncorrelated conversion (UC), while enhancing nonlinear system processing capabilities through integration of the deterministic sampling (DS) method and constrained Rayleigh quotient optimization techniques. Although OCF demonstrates improved performance in nonlinear systems, its effectiveness remains compromised under impulsive noise conditions. To overcome this issue, we propose an optimized conversion-sample filter based on the maximum correntropy criterion (MCOCF). MCOCF integrates the maximum correntropy criterion (MCC), thereby enhancing DS and the constrained Rayleigh quotient to a certain extent. Simulation results indicate that the MCOCF not only improves performance in environments with impulse noise but also significantly enhances the ability to process nonlinear systems.</div></div>","PeriodicalId":49523,"journal":{"name":"Signal Processing","volume":"241 ","pages":"Article 110404"},"PeriodicalIF":3.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145529036","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号