IEEE Transactions on Signal Processing最新文献

{"title":"Novel Receiver Designs for Low-Resolution ADC-Based Uplink MU-MIMO Systems With Unknown Channel Covariance Information","authors":"Rubin Jose Peter;Chandra R. Murthy","doi":"10.1109/TSP.2026.3721208","DOIUrl":"https://doi.org/10.1109/TSP.2026.3721208","url":null,"abstract":"It is known that joint channel estimation and soft symbol decoding enhance the performance of multiple-input multiple-output (MIMO) communication systems with low-resolution analog-to-digital converters (ADCs). However, existing techniques require accurate knowledge of channel statistics to be effective. In this paper, we consider an uplink multi-user MIMO system and develop novel algorithms for joint estimation of channel coefficients and statistics along with soft symbol decoding. Specifically, we develop a sequential processing technique, based on variational Bayesian inference, for jointly estimating the covariance matrices of all users’ channels, the corresponding channel matrices, and soft symbols. We also develop an alternative block processing scheme for the same problem. Corresponding algorithms for sequential and block processing of unquantized data can be obtained as simplifications of our algorithms. On the analytical side, we derive a marginalized Bayesian Cramér-Rao Lower Bound (MB-CRLB) for covariance matrix estimation and elucidate the effect of the number of snapshots and SNR on the estimation error. Finally, we empirically evaluate the new algorithms and demonstrate their superior performance in scenarios where the channel statistics are unknown. The results also show that the mean squared error in channel covariance matrix estimation via the proposed algorithm closely follows the behavior of the MB-CRLB as a function of the SNR and number of frames.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"74 ","pages":"3724-3739"},"PeriodicalIF":5.5,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148925065","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}

{"title":"Semi-Supervised Model-Free Bayesian State Estimation from Compressed Measurements","authors":"Anubhab Ghosh;Yonina C. Eldar;Saikat Chatterjee","doi":"10.1109/TSP.2026.3691273","DOIUrl":"https://doi.org/10.1109/TSP.2026.3691273","url":null,"abstract":"We consider data-driven Bayesian state estimation from compressed measurements (BSCM) of a model-free process. The dimension of the temporal measurement vector is lower than that of the temporal state vector to be estimated, leading to an under-determined inverse problem. The underlying dynamical model of the state’s evolution is unknown for a ‘model-free process.’ Hence, it is difficult to use traditional model-driven methods, for example, Kalman and particle filters. Instead, we consider data-driven methods. We experimentally show that two existing unsupervised learning-based data-driven methods fail to address the BSCM problem in a model-free process. The methods are – data-driven nonlinear state estimation (DANSE) and deep Markov model (DMM). While DANSE provides good predictive/forecasting performance to model the temporal measurement data as a time series, its unsupervised learning lacks suitable regularization for tackling the BSCM task. We then propose a semi-supervised learning approach and develop a semi-supervised learning-based DANSE method, referred to as SemiDANSE. In SemiDANSE, we use a large amount of unlabelled data along with a limited amount of labelled data, i.e., pairwise measurement-and-state data, which provides the desired regularization. Using benchmark chaotic dynamical systems, we empirically show that the data-driven SemiDANSE provides competitive state estimation performance for BSCM using a handful of different measurement systems, against a hybrid method called KalmanNet and two model-driven methods (extended Kalman filter and unscented Kalman filter) that know the dynamical models exactly.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"74 ","pages":"2005-2019"},"PeriodicalIF":5.8,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148171592","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}

{"title":"Accelerated Decentralized Constraint-Coupled Optimization: A Dual${}^{2}$ Approach","authors":"Jingwang Li;Vincent Lau","doi":"10.1109/TSP.2026.3695195","DOIUrl":"https://doi.org/10.1109/TSP.2026.3695195","url":null,"abstract":"In this paper, we focus on a class of decentralized constraint-coupled optimization problem: <inline-formula><tex-math>$mathbf{min}_{boldsymbol x_{boldsymbol i}inmathbb{R}^{boldsymbol d_{boldsymbol i}},boldsymbol{iinmathcal{I};yinmathbb{R}^{boldsymbol p}}}$</tex-math></inline-formula> <inline-formula><tex-math>$sum_{boldsymbol i = 1}^{boldsymbol n}left(boldsymbol f_{boldsymbol i}(boldsymbol x_{boldsymbol i})+boldsymbol g_{boldsymbol i}(boldsymbol x_{boldsymbol i})right)+boldsymbol h(boldsymbol y) mathbf{s.t.} sum_{boldsymbol i = 1}^{boldsymbol n}boldsymbol A_{boldsymbol i}boldsymbol x_{boldsymbol i}=boldsymbol{y}$</tex-math></inline-formula>, over an undirected and connected network of <inline-formula><tex-math>$boldsymbol{n}$</tex-math></inline-formula> agents. Here, <inline-formula><tex-math>$boldsymbol{f_{i}}$</tex-math></inline-formula>, <inline-formula><tex-math>$boldsymbol{g_{i}}$</tex-math></inline-formula>, and <inline-formula><tex-math>$boldsymbol{A_{i}}$</tex-math></inline-formula> represent private information of agent <inline-formula><tex-math>$boldsymbol{i}inmathcal{I}={mathbf{1,cdots,}boldsymbol{n}}$</tex-math></inline-formula>, while <inline-formula><tex-math>$boldsymbol{h}$</tex-math></inline-formula> is public for all agents. Building on a novel dual<inline-formula><tex-math>${}^{mathbf 2}$</tex-math></inline-formula> approach, we develop two accelerated algorithms to solve this problem: the inexact Dual<inline-formula><tex-math>${}^{mathbf 2}$</tex-math></inline-formula> Accelerated (iD2A) gradient method and the Multi-consensus inexact Dual<inline-formula><tex-math>${}^{mathbf 2}$</tex-math></inline-formula> Accelerated (MiD2A) gradient method. We demonstrate that both iD2A and MiD2A can guarantee asymptotic convergence under a milder condition on <inline-formula><tex-math>$boldsymbol{h}$</tex-math></inline-formula> compared to existing algorithms. Furthermore, under additional assumptions, we establish linear convergence rates and derive significantly lower communication and computational complexity bounds than those of existing algorithms. Several numerical experiments validate our theoretical analysis and demonstrate the practical superiority of the proposed algorithms.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"74 ","pages":"2208-2223"},"PeriodicalIF":5.8,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148236159","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}

{"title":"Neural-Network-Augmented Tobit Kalman Filter With Adaptive Covariance for Censored State Estimation","authors":"Sarocha Jetawatthana;Thanakorn Khamvilai","doi":"10.1109/TSP.2026.3695448","DOIUrl":"https://doi.org/10.1109/TSP.2026.3695448","url":null,"abstract":"To estimate states from censored or saturated measurements, the Tobit Kalman Filter (TKF) was developed. However, model errors or mismatches can impair the performance of the TKF, resulting in lower estimation accuracy. In order to increase the TKF’s accuracy and robustness, this study proposes three significant improvements. First, the correlation between the states and the expectation of censoring indicator variables, which predict the occurrence of saturation, is explicitly incorporated into the filtering framework. Second, a neural network, trained online, is augmented to the process model, which allows the filter to dynamically adjust to model mismatches. Third, the theoretical convergence analysis and simulations, based on accepted benchmarks in the literature, are used to assess our proposed Neural-Network-Augmented Tobit Kalman Filter (NN-TKF). Simulation results demonstrate that the NN-TKF outperforms the standard TKF and other estimation methods.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"74 ","pages":"2304-2315"},"PeriodicalIF":5.8,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148236255","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}

{"title":"Efficient Decomposition of Multistage Composite Length FFT for Complex and Real Signals","authors":"Sin-Wei Chiu;Keshab K. Parhi","doi":"10.1109/TSP.2026.3664785","DOIUrl":"10.1109/TSP.2026.3664785","url":null,"abstract":"Fast Fourier Transform (FFT) algorithms play a fundamental role in modern digital signal processing (DSP). A composite length FFT decomposes a longer-length FFT into multiple shorter-length FFTs, enabling a bottom-up construction approach. If the input signal is complex-valued (real-valued), the DFT computation is referred to as CFFT (RFFT). If the input is real-valued, redundancies and symmetries in the RFFT can be exploited to design more efficient algorithms. In this work, we extend the concept of two-stage composite FFTs to multistage composite FFTs. In a multistage FFT, the decomposition and parenthesization of sub-blocks significantly affect the structure’s efficiency. Although the set of sub-blocks remains the same, different parenthesizations lead to varying numbers of twiddle factor multiplications in the interconnect stages. We adopt radix-2 FFTs as the fundamental building blocks and apply dynamic programming to determine the optimal decomposition and parenthesization for arbitrary power-of-two FFT lengths. Our work presents optimal strategies for both complex-valued and real-valued FFTs. Compared to the radix-2 FFT, our multistage FFT reduces the number of complex multiplications by approximately 20% for complex-valued inputs and 30% for real-valued inputs. These results demonstrate that the multistage FFT approach not only simplifies the design process but also improves computational efficiency by significantly reducing the number of complex multiplications. Furthermore, the composite multistage structure enables an efficient bottom-up design methodology, requiring only two types of sub-blocks for CFFT and four types for RFFT implementations. Designers can follow the proposed straightforward design strategies, which guarantee the construction of optimal CFFT and RFFT structures.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"74 ","pages":"892-903"},"PeriodicalIF":5.8,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146205430","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}

{"title":"Tractable Approximation of Labeled Multi-Object Posterior Densities","authors":"Thi Hong Thai Nguyen;Ba-Ngu Vo;Ba Tuong Vo","doi":"10.1109/TSP.2026.3702946","DOIUrl":"https://doi.org/10.1109/TSP.2026.3702946","url":null,"abstract":"Multi-object estimation in state-space models (SSMs) wherein the system state is represented as a finite set has attracted significant interest in recent years. In Bayesian inference, the posterior density captures all information on the system trajectory since it considers the history of the states. In most multi-object SSM applications, closed-form multi-object posteriors are not available for non-standard multi-object models. Thus, functional approximation is necessary because these posteriors are very high-dimensional. This work provides a tractable multi-scan Generalized Labeled Multi-Bernoulli (GLMB) approximation that matches the trajectory cardinality distribution of the labeled multi-object posterior density. The proposed approximation is also proven to minimize the Kullback-Leibler divergence over a special class of multi-scan GLMB model. Additionally, we develop a tractable algorithm for computing the approximate multi-object posteriors over finite windows. Numerical multi-object tracking experiments, using a simulated social force model with uninformative observations, and a real-world social force pedestrian dataset are presented to validate the proposed approximation method.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"74 ","pages":"2803-2819"},"PeriodicalIF":5.5,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148484377","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}

{"title":"Filtering Markov Jump Systems With Partially Known Dynamics: A Model-Based Deep Learning Approach","authors":"George Stamatelis;George C. Alexandropoulos","doi":"10.1109/TSP.2026.3687278","DOIUrl":"10.1109/TSP.2026.3687278","url":null,"abstract":"This paper presents the Markov Jump Filtering Network (MJFNet), a novel model-based deep learning framework for real-time state-state estimation in Markov jump systems with unknown noise statistics and mode transition dynamics. A hybrid architecture comprising two Recurrent Neural Networks (RNNs) is proposed: one for mode prediction and another for filtering that is based on a mode-augmented version of the recently presented KalmanNet architecture. The proposed RNNs are trained jointly using an alternating least squares strategy that enables mutual adaptation without supervision of the latent modes. Extensive numerical experiments on linear and nonlinear systems, including target tracking, pendulum angle tracking, Lorenz attractor dynamics, and a real-life dataset demonstrate that the proposed MJFNet framework outperforms classical model-based filters (e.g., interacting multiple models and particle filters) as well as model-free deep learning baselines, particularly in non-stationary and high-noise regimes. It is also showcased that MJFNet achieves a small yet meaningful improvement over the KalmanNet framework, which becomes much more pronounced in complicated systems or long trajectories. Finally, the method’s performance is empirically validated to be consistent and reliable, exhibiting low sensitivity to initial conditions, hyperparameter selection, as well as to incorrect model knowledge.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"74 ","pages":"1897-1912"},"PeriodicalIF":5.5,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147739685","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}

{"title":"Robust Transceiver Design for Wideband MIMO Radar Under Space-Frequency Scheme","authors":"Junhui Qian;Jiheng Liu;Yujie Zou;Xiaobing Song;Jing Yang;Guobing Qian","doi":"10.1109/TSP.2026.3732566","DOIUrl":"10.1109/TSP.2026.3732566","url":null,"abstract":"Aiming at the problem of angle uncertainty in wideband Multiple-Input Multiple-Output (MIMO) radar system, this paper investigates the robust design of space-frequency joint design for radar transceiver. Firstly, a system model considering angle disturbance is established, and the average Signal-to-Interference-Noise Ratio (SINR) in statistical sense is introduced as a performance indicator to enhance the robustness of target detection. Subsequently, two types of joint optimization strategies are developed: one is a joint optimization method with SINR and Integrated Mainlobe to Sidelobe Ratio (IMSR) as weighted objectives; the other is the SINR maximization problem within a <inline-formula><tex-math>$boldsymbol{varepsilon}$</tex-math></inline-formula> constraint framework. On this basis, combining the Maximization-Minimization (MM) framework, the Dinkelbach framework with the adaptive Alternating Direction Penalty Method (ADPM), the corresponding multi-objective iterative optimization procedures are proposed to comprehensively optimize the mainlobe energy concentration and sidelobe leakage suppression performance, thereby improving the robustness of the algorithm and the flexibility of parameter adjustment. The simulation results verify the superiority of the proposed method in convergence speed and detection performance in multi-target, strong interference and angle uncertainty scenarios.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"74 ","pages":"4356-4371"},"PeriodicalIF":5.5,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915531","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}

{"title":"Dynamic Decomposition Diffusion Model for Probabilistic Multivariate Time Series Forecasting","authors":"Yuxin Li;Ao Wang;Xinyue Hu;Bo Chen;Wenchao Chen;Penghui Wang;Hongwei Liu;Mingyuan Zhou","doi":"10.1109/TSP.2026.3733192","DOIUrl":"10.1109/TSP.2026.3733192","url":null,"abstract":"Accurate probabilistic forecasting of multivariate time series (MTS) is crucial for many real-world applications but remains challenging due to the inherent complexity and non-stationarity of these signals. To address these challenges, we propose the Dynamic Decomposition Diffusion Model (D3M), which integrates advanced signal decomposition techniques with diffusion-based generative models. D3M dynamically decomposes MTS into narrowband intrinsic mode functions (IMFs) and employs a patching channel-independence transformer to efficiently manage the increased data dimensionality. A conditional diffusion model estimates the probabilistic distributions of the decomposed components, with the final MTS distribution obtained as the joint distribution of all components. These modules are seamlessly unified within a Bayesian joint optimization framework, improving forecasting accuracy and uncertainty quantification. Experimental results on six real-world datasets confirm that D3M consistently outperforms state-of-the-art methods in both accuracy and robustness.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"74 ","pages":"4339-4355"},"PeriodicalIF":5.5,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148946696","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}

{"title":"Learned Universal Entropy Coding With Tensor Networks","authors":"Xiaoxuan Fan;Wenrui Dai;Shaohui Li;Wen Fei;Chenglin Li;Junni Zou;Hongkai Xiong","doi":"10.1109/TSP.2026.3733034","DOIUrl":"https://doi.org/10.1109/TSP.2026.3733034","url":null,"abstract":"Existing methods for universal entropy coding suffer from the dilemma between versatility to general data sources and ability in exploiting complex structured statistics. They usually rely on presumed structural and statistical assumptions to decompose joint probability distribution for tractable entropy modeling. To address this problem, in this paper, we propose the first learned universal entropy coding method that leverages tensor networks (TNs) to learn joint probability distribution of diverse data sources without presuming distribution priors or context structures. Specifically, we resort to the tensor network formulation based on matrix product states (MPS) to learn the tractable low-order tensor approximation for high-order tensor representation of joint probability distribution. Entropy encoding and decoding are achieved in a sequential fashion by constructing cumulative distribution functions with marginal probabilities of sources on top of the probabilistic modeling. We demonstrate in theory that the proposed entropy model converges to stationary distribution, guarantees upper and lower bounds of entropy rates, and exponentially decays in truncation error of entropy rate with truncation order and correlation length. Furthermore, TN-based practical coding framework is designed with unified Gray-code bit-plane decomposition to mitigate numerical overflow. Experimental results demonstrate that the proposed method consistently outperforms existing universal approaches and is comparable to task-specific approaches in the tasks of lossless graph and image compression and learned lossy image compression.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"74 ","pages":"4402-4418"},"PeriodicalIF":5.5,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149130960","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}