Handling the challenge of missing measurements in nonlinear systems is a difficult problem in various scientific and engineering fields. Missing measurements, which can arise from technical faults during observation, diffusion channel shrinking, or the loss of specific metrics, can bring many challenges when estimating the state of nonlinear systems. To tackle this issue, this paper proposes a technique that utilizes a robust cubature Kalman filter (RCKF) by integrating Huber’s M-estimation theory with the standard conventional cubature Kalman filter (CKF). Although a CKF is often used for solving nonlinear filtering problems, its effectiveness might be limited due to a lack of knowledge regarding the nonlinear model of the state and noise-related statistical information. In contrast, the RCKF demonstrates an ability to mitigate performance degradation and discretization issues related to track curves by leveraging covariance matrix predictions for state estimation and output control amidst dynamic disruption errors—even when noise statistics deviate from prior assumptions. The performance of extended Kalman filters (EKFs), unscented Kalman filters (UKFs), CKFs, and RCKFs was compared and evaluated using two numerical examples involving the Univariate Non-stationary Growth Model (UNGM) and bearing-only tracking (BOT). The numerical experiments demonstrated that the RCKF outperformed the EKF, EnKF, and CKF in effectively handling anomaly errors. Specifically, in the UNGM example, the RCKF achieved a significantly lower ARMSE (4.83) and ANCI (3.27)—similar outcomes were observed in the BOT example.
11 October 2023 in IEEE Transactions on Signal Processing
In this paper, we study a novel marginal distributionally robust minimum mean-squared error (MDR-MMSE) estimation problem for a random state vector in a multisensory system consisting of several sensors linked with a fusion center by employing the minimax viewpoint, which involves finding the estimator with the best performance under the least favourable distribution within an uncertainty set. In contrast to previous studies, the uncertainty set includes a family of densities whose marginal densities are located in a neighborhood defined by placing a threshold on the Kullback-Leibler (KL) divergence between actual and nominal Gaussian marginal densities, which is suitable for the scenario of the multisensor system. Despite the infinitedimensional optimization nature of the MDR-MMSE estimation problem, we prove that there exists an affine estimator and a corresponding Gaussian distribution constituting its saddle point solution, which can be obtained by solving a finite-dimensional convex optimization problem. In addition, the “partially affine” and “partially Gaussian” properties are also analysed for all saddle point solutions. Moreover, we proved that the optimal estimator for a reduced MDR-MMSE estimation problem with all inactive sensors removed is also optimal for the original problem with all sensors remaining, thus reducing the communication burden of sending the observations from inactive sensors to the fusion center. Additionally, three distinct criteria for evaluating the relationships between the active sensors and the thresholds are established. Numerical experiments are also provided to verify the superiority of our proposed model.
18 March 2021 in IEEE Transactions on Signal Processing
This paper considers distributed estimation of an unknown deterministic parameter vector in a bandwidth constrained multisensor network with a fusion center (FC). Due to the stringent bandwidth requirements, each sensor compresses its observation as a low-dimensional vector via a linear transformation. Then, the FC linearly combines all received compressed data to estimate the deterministic parameter vector based on the best linear unbiased estimator. The problem of interest is to jointly design the dimension assignment (i.e., the compression dimension of each sensor) and the corresponding compression matrix when the total number of compression dimensions is given. Such a joint design problem is formulated as an optimization problem with rank and linear matrix equality constraints, which is shown to be NP-hard for the first time. In addition, penalty decomposition (PD), successive quadratic upper-bound minimization method of multipliers (SQUM-M) and SQUM-M-block coordinate descent (SQUM-M-BCD) are proposed to solve it approximately. Furthermore, we show that any accumulation point of the sequence generated by the PD satisfies the Karush-Kuhn-Tucker conditions of the equivalent formulation of the joint design problem; the SQUM-M admits the same convergence property as the PD under some conditions. Numerical experiments corroborate the merits of PD and SQUM-M-BCD as compared with existing strategies for the heterogeneous scenario, and further illustrate the effectiveness of the proposed algorithms for the correlated noise case.
11 December 2024 in IEEE Transactions on Signal Processing
In this paper, we propose a novel dual inexact nonsmooth Newton (DINN) method for solving a distributed optimization problem, which aims to minimize a sum of cost functions located among agents by communicating only with their neighboring agents over a network. Our method is based on the Lagrange dual of an appropriately formulated primal problem created by introducing local variables for each agent and enforcing a consensus constraint among these variables. Due to the decomposed structure of the dual problem, the DINN method guarantees a superlinear (or even quadratic) convergence rate for both the primal and dual iteration sequences, achieving the same convergence rate as its centralized counterpart. Furthermore, by exploiting the special structure of the dual generalized Hessian, we design a distributed iterative method based on Nesterov’s acceleration technique to approximate the dual Newton direction with suitable precision. Moreover, in contrast to existing second-order methods, the DINN method relaxes the requirement for the objective function to be twice continuously differentiable by using the linear Newton approximation of its gradient. This expands the potential applications of distributed Newton methods. Numerical experiments demonstrate that the DINN method outperforms the current state-of-the-art distributed optimization methods.
14 November 2022 in IEEE Transactions on Signal Processing
Support vector machines (SVMs) are successful supervised learning models that analyze data for classification and regression. Previous work has demonstrated the superiority of the SVMs in dealing with the high dimensional, low sample size problems. However, the increase of the sample size brings great challenges to accurately and efficiently solve the large-scale SVMs, especially for the nonlinear kernel SVMs, which may lead to huge computational costs and unaffordable storage burden. In this paper, we propose a highly efficient sparse semismooth Newton (SsN) based augmented Lagrangian (AL) method for solving a class of large-scale SVMs that can be formulated as a convex quadratic programming problem with a linear equality constraint and a simple box constraint. The asymptotic superlinear convergence rate of both the primal and the dual iteration sequences generated by the AL method is guaranteed due to the piecewise linear-quadratic structure of the problem. Furthermore, we reveal the close connection between the number of support vectors and the sparse structure of the generalized Jacobian for the inner subproblem of the AL method. By exploiting this hidden sparsity, the inner subproblem can be solved by the SsN method efficiently and accurately, which greatly reduces the storage burden and computational costs. In particular, for the nonlinear kernel SVMs, since the sparse structure may not manifest in the early iterations of the AL method, we solve a linear kernel SVM approximated by the random Fourier features method to produce a good initial point, and then transfer to solve the original problem. Numerical experiments demonstrate that the proposed algorithm outperforms the current state-of-the-art solvers for the large-scale SVMs.