噪声协方差估计新算法在惯性导航系统中的仿真研究

研究一种在动态系统常值误差未知的情况下对线性时变随机系统误差协方差进行估计的新方法。该方法通过构造一个新的时间序列,其协方差由未知参数的线性组合组成,然后用递推最小二乘法来计算新序列的协方差,该方法不需要任何关于噪声的先验知识。从仿真结果来看达到了较好的效果。     

Particle flters for probability hypothesis density flter with the presence of unknown measurement noise covariance

In Bayesian multi-target fltering,knowledge of measurement noise variance is very important.Signifcant mismatches in noise parameters will result in biased estimates.In this paper,a new particle flter for a probability hypothesis density（PHD）flter handling unknown measurement noise variances is proposed.The approach is based on marginalizing the unknown parameters out of the posterior distribution by using variational Bayesian（VB）methods.Moreover,the sequential Monte Carlo method is used to approximate the posterior intensity considering non-linear and non-Gaussian conditions.Unlike other particle flters for this challenging class of PHD flters,the proposed method can adaptively learn the unknown and time-varying noise variances while fltering.Simulation results show that the proposed method improves estimation accuracy in terms of both the number of targets and their states.     

Exact multisensor dynamic bias estimation with local tracks

An exact solution is provided for the multiple sensor bias estimation problem based on local tracks. It is shown that the sensor bias estimates can be obtained dynamically using the outputs of the local (biased) state estimators. This is accomplished by manipulating the local state estimates such that they yield pseudomeasurements of the sensor biases with additive noises that are zero-mean, white, and with easily calculated covariances. These results allow evaluation of the Cramer-Rao lower bound (CRLB) on the covariance of the sensor bias estimates, i.e., a quantification of the available information about the sensor biases in any scenario. Monte Carlo simulations show that this method has significant improvement in performance with reduced rms errors of 70% compared with commonly used decoupled Kalman filter. Furthermore, the new method is shown to be statistically efficient, i.e., it meets the CRLB. The extension of the new technique for dynamically varying sensor biases is also presented.     

Estimating Noise Variances by Using Multiple Observers

Unknown variances of the noises that excite a time-invariant, linear dynamic system and/or in the observation of its output can be estimated by use of multiple observers. One observer is needed, in general, for each unknown variance. Each observer is time invariant and has different gains from the others. It is shown that each unknown variance is a linear combination of the variances of the residuals of the observers. The required estimates of the noise variances are obtained by using the measured variances of the residuals. The method presented in this paper is illustrated by an application to determining noise parameters in a ring laser gyro.     

Analysis of Modeling and Bias Errors in Discrete-Time State Estimation

This paper concerns the effects of modeling and bias errors in discrete-time state estimation. The newly derived algorithms include the effect of correlation between plant and measurement noise in the system. The effects of nonzero mean noise terms and bias errors are considered. With plant or measurement matrix errors, divergence can occur. The local or linear sensitivity approach to error analysis, where the sensitivity is defined as a partial derivative with respect to a variable parameter taken about the modeled value, will not show this divergence due to neglect of higher order terms. Approximate algorithms are presented which circumvent the problem inherent in the local sensitivity approach. These make use of a "conditional bias" concept which views system error as a bias, conditioned on knowledge of the state estimates. It is shown that the actual error in optimum estimation is orthogonal to the residue error for suboptimum estimation where the residue error is defined as the difference between the actual estimation error and the optimum estimation error. Two examples, one concerning an integrated navigation system, demonstrate the theoretical results.     

Adaptive Two-Stage Extended Kalman Filter for a Fault-Tolerant INS-GPS Loosely Coupled System

The well-known conventional Kalman filter requires an accurate system model and exact stochastic information. But in a number of situations, the system model has an unknown bias, which may degrade the performance of the Kalman filter or may cause the filter to diverge. The effect of the unknown bias may be more pronounced on the extended Kalman filter (EKF), which is a nonlinear filter. The two-stage extended Kalman filter (TEKF) with respect to this problem has been receiving considerable attention for a long time. Recently, the optimal two-stage Kalman filter (TKF) for linear stochastic systems with a constant bias or a random bias has been proposed by several researchers. A TEKF can also be similarly derived as the optimal TKF. In the case of a random bias, the TEKF assumes that the information of a random bi?s is known. But the information of a random bias is unknown or partially known in general. To solve this problem, this paper proposes an adaptive two-stage extended Kalman filter (ATEKF) using an adaptive fading EKF. To verify the performance of the proposed ATEKF, the ATEKF is applied to the INS-GPS (inertial navigation system-Global Positioning System) loosely coupled system with an unknown fault bias. The proposed ATEKF tracked/estimated the unknown bias effectively although the information about the random bias was unknown.     

A practical filter error method for aerodynamic parameter estimation of aircraft in turbulence

It is common for aircraft to encounter atmospheric turbulence in flight tests. Turbulence is usually modeled as stochastic process noise in the flight dynamics equations. In this paper, parameter estimation of nonlinear dynamic system with both process and measurement noise was studied, and a practical filter error method was proposed. The linearized Kalman filter of first-order approximation was used for state estimation, in which the filter gain, along with the system parameters and the initial states, constituted the parameter vector to be estimated. The unknown parameters and measurement noise covariance were estimated alternately by a relaxation iteration method, and the sensitivities of observations to unknown parameters were calculated by finite difference approximation. Some practical aspects of the method application were discussed. The proposed filter error method was validated by the flight simulation data of a research aircraft. Then, the method was applied to the flight tests of a subscale aircraft, and the aerodynamic stability and control derivatives were estimated. All the estimation results were compared with the results of the output error method to demonstrate the effectiveness of the approach. It is shown that the filter error method is superior to the output error method for flight tests in atmospheric turbulence.     

A Recursive Method for Signal Resolution

A recursive method is given for resolving signals overlapping in time. Assume that the signal waveform is known and several signals are received. The signals (of unknown number) may overlap with one another and the amount of time delay of each individual signal is unknown. The signals are corrupted with additive white Gaussian noise. The problem is to estimate the number, the amplitudes, and the time delays of the overlapping signals. Assume that at a certain instant tk-1 estimates have been made on the number of signals arriving in the time interval (O, tk-1) and the amplitudes and time delays of these signals. Using these estimates, we test at tk the hypothesis H1 that a new signal arrives at tk against the null hypothesis Ho that no new signal arrives. The decision gives the number of signals arriving in the time interval (0, tk); the parameters are then re-estimated. The overlapping signals are detected and resolved, and the estimates are improved at each stage. The system is analyzed in detail, and computer-simulated results are presented.     

Efficient solution and performance analysis of 3-D positionestimation by trilateration

A new exact, explicit, and computationally efficient solution for three-dimensional (3-D) position estimation based on range measurements from three stations is proposed. The simple polynomial-type form of the new algorithm facilitates the performance analysis. Formulae are provided for both the variance and the bias of the position estimates. The systematic error is a joint effect of both the measurement noise and the system nonlinearity and its magnitude cannot be ignored if highly accurate localization is required. Performance evaluation results are presented for various conditions     

Noise subspace techniques in non-gaussian noise using cumulants

We consider noise subspace methods for narrowband direction-of-arrival or harmonic retrieval in colored linear non-gaussian noise of unknown covariance and unknown distribution. The non-gaussian noise covariance is estimated via higher order cumulants and combined with correlation information to solve a generalized eigenvalue problem. The estimated eigenvectors are used in a variety of noise subspace methods such as multiple signal classification (MUSIC), MVDR and eigenvector. The noise covariance estimates are obtained in the presence of the harmonic signals, obviating the need for noise-only training records. The covariance estimates may be obtained nonparametrically via cumulant projections, or parametrically using autoregressive moving average (ARMA) models. An information theoretic criterion using higher order cumulants is presented which may be used to simultaneously estimate the ARMA model order and parameters. Third- and fourth-order cumulants are employed for asymmetric and symmetric probability density function (pdf) cases, respectively. Simulation results show considerable improvement over conventional methods with no prewhitening. The effects of prewhitening are particularly evident in the dominant eigenvalues, as revealed by singular value decomposition (SVD) analysis     

Estimation of 3D angular motion using gyroscopes and linear accelerometers

The authors present a novel, real-time angular motion estimation technique using a linear Gaussian estimator, and the outputs of linear accelerometers and gyroscopes, to assess the actual angular velocity of a rigid body in three-dimensional (3D) space. The method obtains the covariances of the random actual 3D angular velocity, the angular velocity measurement, and the measurement noise from the time averages of the outputs of an array of nine linear accelerometers and the outputs of three orthogonal gyroscopes. These statistics are used by the estimator to calculate the angular velocity of the rigid body in 3D space. The multisensor technique performance is evaluated through a computer simulation. Results indicate that the method leads to more accurate angular velocity values than are obtained conventionally.<>     

Adaptive distributed filtering in multicoordinated systems

The direct estimation of optimal steady-state gain in the single filtering process introduced by B. Carew et al. (1973) is extended to multicoordinated systems, and the distributed optimal steady-state gains are directly estimated for adaptive distributed filtering. The correlation method using distributed innovation processes is used. The algorithm assumes little prior information about the unknown covariances and adaptively changes the weights to best integrate the distributed estimates obtained in local filtering processes. The term best is used in the sense that the result of the adaptive distributed filtering is as close to that of the optimal distributed filtering as possible     

Kalman Filter Compensation for a Special Class of Systems

Filter compensation techniques for several special but practical cases are discussed. A general set of bias and covariance equations for linear filters with modeling errors is first summarized. A method for relating the modeling errors to the selection of the covariance of "process noise" for model error compensation is suggested. A performance ordering for cases when the true system becomes a subsystem of the model used for the filter construction is given. A bias correcting filter is derived for the case when the filter is matched only to a subsystem of the actual system.     

混合线性/非线性模型的准高斯Rao-Blackwellized粒子滤波法

 针对混合线性/非线性模型,提出一种新的递推估计滤波算法,称为准高斯Rao-Blackwellized粒子滤波器（Q-GRBPF）。算法采用Rao-Blackwellized思想,将线性状态与非线性状态进行分离,对非线性状态运用准高斯粒子滤波（Q-GPF）算法进行估计,并将其后验分布近似为单个高斯分布,再利用非线性状态的估计值对线性状态进行卡尔曼滤波（KF）估计。将Q-GRBPF应用于目标跟踪的仿真结果表明,与Rao-Blackwellized粒子滤波器（RBPF）相比,Q-GRBPF在保证估计精度的前提下有效降低了计算复杂度,计算时间约为RBPF的58%;与Q-GPF相比,x坐标与y坐标的估计精度分别提升了45％和30％,而计算时间也节省了约30％。     

一种基于冗余测量的自适应卡尔曼滤波算法

针对目前自适应滤波算法的不足,在测量系统量测噪声方差未知的情况下,设计了一种基于冗余测量的自适应卡尔曼滤波(RMAKF)算法。通过对系统冗余测量值的一阶、二阶差分序列进行有效的统计分析,可以准确估计系统量测噪声统计特性,进而在滤波过程中自适应调节噪声方差阵R,提高滤波精度。以全球定位系统/惯性导航系统(GPS/INS)松组合导航系统为对象进行了仿真实验,结果表明该算法在测量系统噪声特性未知或发生改变时,可对其进行准确估计,在采用低精度惯性器件情况下,滤波结果较其他主要自适应卡尔曼滤波算法有较明显的改进。     

A Stratified Statistical Filter with Nonlinear Application

This paper introduces a statistical filter (or, more strictly, a filtering algorithm) which has intended application in the area of nonlinear systems. Within this context, the filter enables one to investigate the convergence effects produced by varying the initial estimates associated with the respective state variables, together with the various system parameters. The present algorithm is not intended to replace the more powerful optimal statistical filters used in linear theory, but rather to provide a simulation tool which can readily be applied to a given nonlinear system. The application considered in this paper bears a similarity to a tracking problem which might be encountered by an optical device, where angular information is the primary observable quanity. In this particular application, angular observations are available, and statistical estimates are desired for a position variable, together with an unknown parameter. The application is introduced primarily for the purpose of demonstrating the behavior of the filter when applied to a relatively simple nonlinear system.     

Observer-type Kalman innovation filter for uncertain linear systems

An observer-type of Kalman innovation filtering algorithm to find a practically implementable "best" Kalman filter, and such an algorithm based on the evolutionary programming (EP) optima-search technique, are proposed, for linear discrete-time systems with time-invariant unknown-but-hounded plant and noise uncertainties. The worst-case parameter set from the stochastic uncertain system represented by the interval form with respect to the implemented "best" filter is also found in this work for demonstrating the effectiveness of the proposed filtering scheme. The new EP-based algorithm utilizes the global optima-searching capability of EP to find the optimal Kalman filter and state estimates at every iteration, which include both the best possible worst case Interval and the optimal nominal trajectory of the Kalman filtering estimates of the system state vectors. Simulation results are included to show that the new algorithm yields more accurate estimates and is less conservative as compared with other related robust filtering schemes     

Reduced state estimators for consistent tracking of maneuvering targets

Linear Kalman filters, using fewer states than required to completely specify target maneuvers, are commonly used to track maneuvering targets. Such reduced state Kalman filters have also been used as component filters of interacting multiple model (IMM) estimators. These reduced state Kalman filters rely on white plant noise to compensate for not knowing the maneuver - they are not necessarily optimal reduced state estimators nor are they necessarily consistent. To be consistent, the state estimation and innovation covariances must include the actual errors during a maneuver. Blair and Bar-Shalom have shown an example where a linear Kalman filter used as an inconsistent reduced state estimator paradoxically yields worse errors with multisensor tracking than with single sensor tracking. We provide examples showing multiple facets of Kalman filter and IMM inconsistency when tracking maneuvering targets with single and multiple sensors. An optimal reduced state estimator derived in previous work resolves the consistency issues of linear Kalman filters and IMM estimators.     

Jump detection/estimation using stepwise regression

The detection and estimation of jumps with unknown time and magnitude in specific states of a dynamic process is addressed. Unlike most of the techniques described in the literature, the method can handle multiple jumps within the data window. This allows the use of longer data spans with consequently improved jump estimation. Jumps are treated as bias states, and the innovations and innovation sensitivities from a jump-free filter are used as data for regression. Forward stepwise regression provides the means to systematically search all the jump possibilities. Removal of other bias states from the filter and inclusion in the regression improves the performance of the method. A realistic inertial navigation example with multiple jumps is given to demonstrate the advantages of the technique. The method works best offline using the entire data span, but the performance of the online moving window version is only slightly degraded     

Multisensor multitarget bias estimation for general asynchronous sensors

A novel solution is provided for the bias estimation problem in multiple asynchronous sensors using common targets of opportunity. The decoupling between the target state estimation and the sensor bias estimation is achieved without ignoring or approximating the crosscovariance between the state estimate and the bias estimate. The target data reported by the sensors are usually not time-coincident or synchronous due to the different data rates. Since the bias estimation requires time-coincident target data from different sensors, a novel scheme is used to transform the measurements from the different times of the sensors into pseudomeasurements of the sensor biases with additive noises that are zero-mean, white, and with easily calculated covariances. These results allow bias estimation as well as the evaluation of the Cramer-Rao lower bound (CRLB) on the covariance of the bias estimate, i.e., the quantification of the available information about the biases in any scenario. Monte Carlo simulation results show that the new method is statistically efficient, i.e., it meets the CRLB. The use of this technique for scale and sensor location biases in addition to the usual additive biases is also presented.