Search

This technical note focuses on efficiently identifying the unknown or time-varying random latency probability (RLP) of the measurements in the networked multi-sensor system by resorting to expectation maximization (EM) framework. Firstly, a novel scheme is proposed for equivalently decomposing the complete data log-likelihood function into a summation form parameterized by RLP. Secondly, the rapid computation of the expectation in E-step is achieved by skillfully introducing Jessen's inequality to avoid the state augmentation of the traditional method. Thirdly, the analytical identification result of RLP is obtained in M-step by constructing Lagrange operator to maximize the expectation with the parameter constraint. Naturally, such analytical result is so simple that it can be quickly carried out, which is demonstrated by quantitative computation complexity analysis. Finally, an example motivated by the maneuvering target tracking application is presented to show the superiority of the new method. [ABSTRACT FROM AUTHOR]

This paper is concerned with the Gaussian approximation (GA) smoothing estimation for the nonlinear system with the colored measurement noise modeled as an autoregressive process. Firstly, based on measurement differencing scheme, designing the GA smoothers with the colored measurement noise is transformed into deriving the GA ones with delayed state. Secondly, the novel fixed- interval, fixed-point and fixed-lag GA smoothers are proposed via the recursive operation of analytical computation and nonlinear integrals, as the general and unifying frameworks: they are applicable for both linear and nonlinear systems; by setting the noise correlation parameter as zero, they can automatically reduce to the standard GA smoothers with uncorrelated white noises; many implementations of the GA frameworks can be developed through utilizing different numerical technologies for computing such nonlinear integrals, e.g., the cubature rule based cubature Kalman smoothers (CKSs) with the colored measurement noise. Finally, the superior performance in estimation accuracy and computation efficiency of the proposed smoothing methods is demonstrated with a multi-sensor target tracking example. [ABSTRACT FROM PUBLISHER]

This technical note is concerned with the nonlinear state smoothing problem in the case that the measurements are randomly delayed by one sampling time. Under Gaussian domain, two general Gaussian approximation (GA) and Gaussian mixture approximation (GMA) smoothers are proposed in minimum mean square error (MMSE) sense. The smoothing implementation is transformed into computing some special posterior covariances, which triggers the development of the new unscented Kalman smoother (UKS) by applying unscented transformation (UT). Simulation results demonstrate the superior performance of the proposed GA-UKS and GMA-UKS algorithms. [ABSTRACT FROM PUBLISHER]