Estimation using sampled data containing sequentially correlated noise.
Multistage linear dynamic systems with sequentially correlated noise, evaluating filtering, prediction and smoothing procedures
Bryson, A. E., Jr.↗
SEARCH · Search NASA
Results for “sequential filters”
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Lunar satellite techniques applicable to earth satellite geodesy
The techniques considered are related to gravity profiling from Doppler residuals, a solution for a surface mass distribution, and a dynamical approach for the estimation of random forces at each data point. The dynamical approach makes use of a new filtering method which uses sequential estimation theory to determine unknown and somewhat correlated forces which randomly perturb the orbit. It is concluded that lunar data reduction methods would be suitable for applications in connection with the GEOS-C/ATS-F mission.
Sjogren, W. L.↗
Guidance simulation and test support for differential GPS flight experiment
Three separate tasks which supported the test preparation, test operations, and post test analysis of the NASA Ames flight test evaluation of the differential Global Positioning System (GPS) are presented. Task 1 consisted of a navigation filter design, coding, and testing to optimally make use of GPS in a differential mode. The filter can be configured to accept inputs from external censors such as an accelerometer and a barometric or radar altimeter. The filter runs in real time onboard a NASA helicopter. It processes raw pseudo and delta range measurements from a single channel sequential GPS receiver. The Kalman filter software interfaces are described in detail, followed by a description of the filter algorithm, including the basic propagation and measurement update equations. The performance during flight tests is reviewed and discussed. Task 2 describes a refinement performed on the lateral and vertical steering algorithms developed on a previous contract. The refinements include modification of the internal logic to allow more diverse inflight initialization procedures, further data smoothing and compensation for system induced time delays. Task 3 describes the TAU Corp participation in the analysis of the real time Kalman navigation filter. The performance was compared to that of the Z-set filter in flight and to the laser tracker position data during post test analysis. This analysis allowed a more optimum selection of the parameters of the filter.
Geier, G. J.↗
Multi-Pass Sequential Mini-Batch Stochastic Gradient Descent Algorithms for Noise Covariance Estimation in Adaptive Kalman Filtering
Estimation of unknown noise covariances in a Kalman filter is a problem of significant practical interest in a wide array of applications. Although this problem has a long history, reliable algorithms for their estimation were scant, and necessary and sufficient conditions for identifiability of the covariances were in dispute until recently. Necessary and sufficient conditions for covariance estimation and a batch estimation algorithm were presented in our previous study. This paper presents stochastic gradient descent algorithms for noise covariance estimation in adaptive Kalman filters that are an order of magnitude faster than the batch method for similar or better root mean square error. More significantly, these algorithms are applicable to non-stationary systems where the noise covariances can occasionally jump up or down by an unknown magnitude. The computational efficiency of the new algorithms stems from adaptive thresholds for convergence, recursive fading memory estimation of the sample cross-correlations of the innovations, and accelerated stochastic gradient descent algorithms. The comparative evaluation of the proposed methods on a number of test cases demonstrates their computational efficiency and accuracy.
Adaptive Kalman filtering↗
Development and test of video systems for airborne surveillance of oil spills
Five video systems - potentially useful for airborne surveillance of oil spills - were developed, flight tested, and evaluated. The systems are: (1) conventional black and white TV, (2) conventional TV with false color, (3) differential TV, (4) prototype Lunar Surface TV, and (5) field sequential TV. Wavelength and polarization filtering were utilized in all systems. Greatly enhanced detection of oil spills, relative to that possible with the unaided eye, was achieved. The most practical video system is a conventional TV camera with silicon-diode-array image tube, filtered with a Corning 7-54 filter and a polarizer oriented with its principal axis in the horizontal direction. Best contrast between oil and water was achieved when winds and sea states were low. The minimum detectable oil film thickness was about 0.1 micrometer.
Millard, J. P.↗
Rotorcraft system identification techniques for handling qualities and stability and control evaluation
An integrated approach to rotorcraft system identification is described. This approach consists of sequential application of (1) data filtering to estimate states of the system and sensor errors, (2) model structure estimation to isolate significant model effects, and (3) parameter identification to quantify the coefficient of the model. An input design algorithm is described which can be used to design control inputs which maximize parameter estimation accuracy. Details of each aspect of the rotorcraft identification approach are given. Examples of both simulated and actual flight data processing are given to illustrate each phase of processing. The procedure is shown to provide means of calibrating sensor errors in flight data, quantifying high order state variable models from the flight data, and consequently computing related stability and control design models.
Hall, W. E., Jr.↗
Relative performance of algorithms for autonomous satellite orbit determination
Limited word size in contemporary microprocessors causes numerical problems in autonomous satellite navigation applications. Numerical error introduced in navigation computations performed on small wordlength machines can cause divergence of sequential estimation algorithms. To insure filter reliability, square root algorithms have been adopted in many applications. The optimal navigation algorithm requires a careful match of the estimation algorithm, dynamic model, and numerical integrator. In this investigation, the relationship of several square root filters and numerical integration methods is evaluated to determine their relative performance for satellite navigation applications. The numerical simulations are conducted using the Phase I GPS constellation to determine the orbit of a LANDSAT-D type satellite. The primary comparison is based on computation time and relative estimation accuracy.
Tapley, B. D.↗
TDRSS-user orbit determination using batch least-squares and sequential methods
The Goddard Space Flight Center (GSFC) Flight Dynamics Division (FDD) commissioned Applied Technology Associates, Incorporated, to develop the Real-Time Orbit Determination/Enhanced (RTOD/E) system on a Disk Operating System (DOS)-based personal computer (PC) as a prototype system for sequential orbit determination of spacecraft. This paper presents the results of a study to compare the orbit determination accuracy for a Tracking and Data Relay Satellite System (TDRSS) user spacecraft, Landsat-4, obtained using RTOD/E, operating on a PC, with the accuracy of an established batch least-squares system, the Goddard Trajectory Determination System (GTDS), and operating on a mainframe computer. The results of Landsat-4 orbit determination will provide useful experience for the Earth Observing System (EOS) series of satellites. The Landsat-4 ephemerides were estimated for the January 17-23, 1991, timeframe, during which intensive TDRSS tracking data for Landsat-4 were available. Independent assessments were made of the consistencies (overlap comparisons for the batch case and covariances and the first measurement residuals for the sequential case) of solutions produced by the batch and sequential methods. The forward-filtered RTOD/E orbit solutions were compared with the definitive GTDS orbit solutions for Landsat-4; the solution differences were less than 40 meters after the filter had reached steady state.
Oza, D. H.↗
Evaluation of Landsat-4 orbit determination accuracy using batch least-squares and sequential methods
The Goddard Space Flight Center (GSFC) Flight Dynamics Division (FDD) commissioned Applied Technology Associates, Incorporated, to develop the Real-Time Orbit Determination/Enhanced (RTOD/E) system on a Disk Operating System (DOS)-based personal computer (PC) as a prototype system for sequential orbit determination of spacecraft. This paper presents the results of a study to compare the orbit determination accuracy for a Tracking and Data Relay Satellite (TDRS) System (TDRSS) user spacecraft, Landsat-4, obtained using RTOD/E, operating on a PC, with the accuracy of an established batch least-squares system, the Goddard Trajectory Determination System (GTDS), operating on a mainframe computer. The results of Landsat-4 orbit determination will provide useful experience for the Earth Observing System (EOS) series of satellites. The Landsat-4 ephemerides were estimated for the May 18-24, 1992, timeframe, during which intensive TDRSS tracking data for Landsat-4 were available. During this period, there were two separate orbit-adjust maneuvers on one of the TDRSS spacecraft (TDRS-East) and one small orbit-adjust maneuver for Landsat-4. Independent assessments were made of the consistencies (overlap comparisons for the batch case and covariances and the first measurement residuals for the sequential case) of solutions produced by the batch and sequential methods. The forward-filtered RTOD/E orbit solutions were compared with the definitive GTDS orbit solutions for Landsat-4; the solution differences were generally less than 30 meters after the filter had reached steady state.
Oza, D. H.↗
Automated daily processing of more than 1000 ground-based GPS receivers for studying intense ionospheric storms
To take advantage of the vast amount of GPS data, researchers use a number of techniques to estimate satellite and receiver interfrequency biases and the total electron content (TEC) of the ionosphere. Most techniques estimate vertical ionospheric structure and, simultaneously, hardware-related biases treated as nuisance parameters. These methods often are limited to 200 GPS receivers and use a sequential least squares or Kalman filter approach. The biases are later removed from the measurements to obtain unbiased TEC. In our approach to calibrating GPS receiver and transmitter interfrequency biases we take advantage of all available GPS receivers using a new processing algorithm based on the Global Ionospheric Mapping (GIM) software developed at the Jet Propulsion Laboratory. This new capability is designed to estimate receiver biases for all stations. We solve for the instrumental biases by modeling the ionospheric delay and removing it from the observation equation using precomputed GIM maps. The precomputed GIM maps rely on 200 globally distributed GPS receivers to establish the ''background'' used to model the ionosphere at the remaining 800 GPS sites.
total electron content (TEC)↗
Accurate Magnetometer/Gyroscope Attitudes Using a Filter with Correlated Sensor Noise
Magnetometers and gyroscopes have been shown to provide very accurate attitudes for a variety of spacecraft. These results have been obtained, however, using a batch-least-squares algorithm and long periods of data. For use in onboard applications, attitudes are best determined using sequential estimators such as the Kalman filter. When a filter is used to determine attitudes using magnetometer and gyroscope data for input, the resulting accuracy is limited by both the sensor accuracies and errors inherent in the Earth magnetic field model. The Kalman filter accounts for the random component by modeling the magnetometer and gyroscope errors as white noise processes. However, even when these tuning parameters are physically realistic, the rate biases (included in the state vector) have been found to show systematic oscillations. These are attributed to the field model errors. If the gyroscope noise is sufficiently small, the tuned filter 'memory' will be long compared to the orbital period. In this case, the variations in the rate bias induced by field model errors are substantially reduced. Mistuning the filter to have a short memory time leads to strongly oscillating rate biases and increased attitude errors. To reduce the effect of the magnetic field model errors, these errors are estimated within the filter and used to correct the reference model. An exponentially-correlated noise model is used to represent the filter estimate of the systematic error. Results from several test cases using in-flight data from the Compton Gamma Ray Observatory are presented. These tests emphasize magnetometer errors, but the method is generally applicable to any sensor subject to a combination of random and systematic noise.
Sedlak, J.↗
On optimal nonlinear estimation. I - Continuous observation.
A generalization of Bucy's (1965) representation theorem is obtained under very weak hypotheses. The generalized theorem is shown to play the same role in the case of general optimal estimation for an arbitrary random process as does the Bucy theorem in the case of optimal filtering for a diffusion process. At least for the models considered, the possibility is pointed out to reduce all sequential estimation problems to the problem of filtering. Hence, filtering theory is seen to represent the core of estimation theory, and is believed to define the direction in which future research should be focused.
Lo, J. T.↗
Estimation and filtering for high-precision GPS positioning applications
Estimation of GPS satellite orbits and other parameters for high-accuracy geodetic and tracking applications is carried out with a multi-satellite batch sequential pseudo-epoch state process noise filter. Both square-root information filtering and UD factorized covariance filtering and smoothing formulations are implemented in the software to enhance numerical stability. The filtering algorithms have been arranged to take advantage of sparse matrices and other characteristics of the GPS measurement scenarios. The filter includes unique error evaluation capabilities to assess effects from mismodeling. Process noise plays a key role in the orbit determination for stochastic behavior of transmitter/receiver clocks, atmospheric-induced delay fluctuations, and unmodeled spacecraft accelerations. Present GPS orbit accuracy is better than 1 m, with ground baselines determined simultaneously to relative accuracy of 1.5 parts in 10 to the 8th over several thousand km distance.
Lichten, Stephen M.↗
Comparing Consider-Covariance Analysis with Sigma-Point Consider Filter and Linear-Theory Consider Filter Formulations
Recent literature in applied estimation theory reflects growing interest in the sigma-point (also called unscented ) formulation for optimal sequential state estimation, often describing performance comparisons with extended Kalman filters as applied to specific dynamical problems [c.f. 1, 2, 3]. Favorable attributes of sigma-point filters are described as including a lower expected error for nonlinear even non-differentiable dynamical systems, and a straightforward formulation not requiring derivation or implementation of any partial derivative Jacobian matrices. These attributes are particularly attractive, e.g. in terms of enabling simplified code architecture and streamlined testing, in the formulation of estimators for nonlinear spaceflight mechanics systems, such as filter software onboard deep-space robotic spacecraft. As presented in [4], the Sigma-Point Consider Filter (SPCF) algorithm extends the sigma-point filter algorithm to the problem of consider covariance analysis. Considering parameters in a dynamical system, while estimating its state, provides an upper bound on the estimated state covariance, which is viewed as a conservative approach to designing estimators for problems of general guidance, navigation and control. This is because, whether a parameter in the system model is observable or not, error in the knowledge of the value of a non-estimated parameter will increase the actual uncertainty of the estimated state of the system beyond the level formally indicated by the covariance of an estimator that neglects errors or uncertainty in that parameter. The equations for SPCF covariance evolution are obtained in a fashion similar to the derivation approach taken with standard (i.e. linearized or extended) consider parameterized Kalman filters (c.f. [5]). While in [4] the SPCF and linear-theory consider filter (LTCF) were applied to an illustrative linear dynamics/linear measurement problem, in the present work examines the SPCF as applied to nonlinear sequential consider covariance analysis, i.e. in the presence of nonlinear dynamics and nonlinear measurements. A simple SPCF for orbit determination, exemplifying an algorithm hosted in the guidance, navigation and control (GN&C) computer processor of a hypothetical robotic spacecraft, was implemented, and compared with an identically-parameterized (standard) extended, consider-parameterized Kalman filter. The onboard filtering scenario examined is a hypothetical spacecraft orbit about a small natural body with imperfectly-known mass. The formulations, relative complexities, and performances of the filters are compared and discussed.
Lisano, Michael E.↗
Current-State Constrained Filter Bank for Wald Testing of Spacecraft Conjunctions
We propose a filter bank consisting of an ordinary current-state extended Kalman filter, and two similar but constrained filters: one is constrained by a null hypothesis that the miss distance between two conjuncting spacecraft is inside their combined hard body radius at the predicted time of closest approach, and one is constrained by an alternative complementary hypothesis. The unconstrained filter is the basis of an initial screening for close approaches of interest. Once the initial screening detects a possibly risky conjunction, the unconstrained filter also governs measurement editing for all three filters, and predicts the time of closest approach. The constrained filters operate only when conjunctions of interest occur. The computed likelihoods of the innovations of the two constrained filters form a ratio for a Wald sequential probability ratio test. The Wald test guides risk mitigation maneuver decisions based on explicit false alarm and missed detection criteria. Since only current-state Kalman filtering is required to compute the innovations for the likelihood ratio, the present approach does not require the mapping of probability density forward to the time of closest approach. Instead, the hard-body constraint manifold is mapped to the filter update time by applying a sigma-point transformation to a projection function. Although many projectors are available, we choose one based on Lambert-style differential correction of the current-state velocity. We have tested our method using a scenario based on the Magnetospheric Multi-Scale mission, scheduled for launch in late 2014. This mission involves formation flight in highly elliptical orbits of four spinning spacecraft equipped with antennas extending 120 meters tip-to-tip. Eccentricities range from 0.82 to 0.91, and close approaches generally occur in the vicinity of perigee, where rapid changes in geometry may occur. Testing the method using two 12,000-case Monte Carlo simulations, we found the method achieved a missed detection rate of 0.1%, and a false alarm rate of 2%.
Carpenter, J. Russell↗
Using Filter Methods to Guide Convergence for ADMM, with Applications to Nonnegative Matrix Factorization Problems
Nonconvex, nonlinear optimization problems arise naturally in parameter fitting and machine learning. While augmented Lagrangian methods have demonstrated robust convergence for classes of these problems, their convergence for block updates has been relatively unexplored outside of the context of the alternating direction method of multipliers (ADMM). ADMM has seen extensive use in these applications, but may exhibit uncertain convergence behavior in many practical nonconvex settings, and struggles with general nonlinear constraints. In contrast, filter methods have proved effective in enforcing convergence for sequential quadratic programming methods and interior point methods with feasibility criteria. We develop an ADMM-filter method for highly nonlinear and nonconvex problems. Here, we show convergence under mild assumptions for several types of coordinate descent schemes, and demonstrate our algorithm on nonnegative matrix factorization and completion problems in imaging and chemical spectrum analysis.
Nonconvex optimization↗
A model of the human observer in failure detection tasks
A model for the human observer in failure detection tasks is proposed which consists of two stages: a linear estimator and a decision mechanism. The estimator is a Kalman filter, and the decision mechanism, which is based on Wald's sequential analysis, leads to a decision function which is the integration of the filter residuals. The final result is a simple detection system which depends on only three parameters, and the sensitivity of the model to these parameters is analyzed. The results of an experiment designed to test the validity of the model are reported. The question of open and closed decision intervals as well as the generalization of the model to more complicated cases is discussed.
Gai, E. G.↗
Pipeline active filter utilizing a booth type multiplier
Multiplier units of the modified Booth decoder and carry-save adder/full adder combination are used to implement a pipeline active filter wherein pixel data is processed sequentially, and each pixel need only be accessed once and multiplied by a predetermined number of weights simultaneously, one multiplier unit for each weight. Each multiplier unit uses only one row of carry-save adders, and the results are shifted to less significant multiplier positions and one row of full adders to add the carry to the sum in order to provide the correct binary number for the product Wp. The full adder is also used to add this product Wp to the sum of products .SIGMA.Wp from preceding multiply units. If m.times.m multiplier units are pipelined, the system would be capable of processing a kernel array of m.times.m weighting factors.
Nathan, Robert↗