Data processing: measuring – calibrating – or testing – Measurement system – Measured signal processing
Reexamination Certificate
2002-04-26
2004-12-07
Barlow, John (Department: 2863)
Data processing: measuring, calibrating, or testing
Measurement system
Measured signal processing
C702S196000
Reexamination Certificate
active
06829568
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates generally to a signal processing method and apparatus for fusing a plurality of signals corresponding to estimates of the state of an object, system, or process. The method and apparatus is specialized or programmed for (1) receiving estimates, each of which can be expressed in the form of a state vector and an error covariance matrix (at least one estimate of which can be decomposed into a sum of an independent error covariance matrix and a potentially correlated error covariance matrix), and (2) transmitting a resulting signal corresponding to an estimate, which can be expressed in the form of a state vector and an error matrix, in order to evoke a physical response from a system receiving the signal.
2. Discussion of Background and Prior Art
2.1 The Data Fusion Signal Processing Problem:
The data fusion problem is essentially one of producing a single estimate of the state of an object, system, or process that combines information from multiple estimates related to the state of that object, system, or process. A typical physical example of a data fusion system is a tracking filter that maintains an estimate of the position, speed, and bearing of an autonomous vehicle and must fuse that estimate with estimate signals obtained from physical sensors, which measure quantities related to the state of the vehicle, in order to produce a single improved estimate signal that can be used by a controlling device to precisely steer the vehicle.
A signal herein is defined as any measurable quantity that is related to the changing of the physical state of a process, system, or substance. A signal includes, but is not limited to, radiation produced by a natural or man made process, electrical fluctuations produced by a natural or a man made process, distinctive materials or chemicals produced by a natural or man made process, distinctive structures or configurations of materials produced by a natural or man made process, or distinctive patterns of radiation or electrical activity produced by a natural or a man made process.
Most generally, a signal representing a measurement of any physical system inherently has some degree of random error associated therewith. Thus, the model of any physical system, if it is to accurately account for that random error, must include a way to estimate the expected values and uncertainties in the values of the physical system that occur due to the random error. Some methods assume that the error values have known, relatively small, bounded magnitudes. Unfortunately, such bounds are typically unavailable in practice, so bounded error approaches are not widely applied. More generally, only estimates of the expected error values can be made, as will be discussed subsequently.
The measurement of a signal is provided by a measuring device. A measuring device as defined herein may be, but is not limited to, any physical device that interacts with a physical system and provides information that can be converted into an estimate comprised of a nominal estimate of the state of the system and an estimate of the error associated with that nominal state estimate. A measuring device as defined herein includes any device that emits a signal and measures the change of the signal upon its return, a device that measures a signal that is naturally produced by a physical process, or any device that measures a signal that is produced by a man made process.
2.2 Data Fusion For Signals Containing Mean and Covariance Information:
In one of the most common formulations of the data fusion problem, each estimate is represented as, or can be converted to, a pair comprising a state vector (often referred to as the mean) a and an error matrix A, denoted {a, A}. The state vector a is a (column) vector composed of m elements in which element a
i
corresponds to a variable, such as size or temperature, describing the state of a system of interest. The error matrix A is a matrix having m rows and m columns in which the element A
ij
, for any choice of i and j between 1 and m, is related to the expected value of the product of the errors associated with the values stored in elements a
i
and a
j
. If the value of element A
ij
is precisely the expected value of the product of the errors associated with the values stored in elements a
i
and a
j
, for any choice of i and j between 1 and m, then A is referred to as the error covariance of the estimated state vector a.
The error matrix A is often referred to as a covariance matrix according to a general definition of a covariance matrix as being a symmetric matrix with nonnegative eigenvalues, but A is not in general the true error covariance T associated with the state estimate a because T is usually unknown. The standard practice is to choose A large enough that it can be assumed to be of the form A=T+E, where E is an unknown covariance matrix representing the overestimated components of A. An overestimated covariance matrix is said to be “conservative” because it suggests that the state estimate is less accurate than it actually is. This is preferred to an underestimated covariance that suggests that the state estimate is more accurate than it actually is. For example, an underestimated covariance could lead a chemical plant operator to believe that the state of a chemical reaction is comfortably within safe operating bounds when in fact the magnitude of the errors in the state estimate are sufficiently large that the true state could easily be outside of the bounds.
The terms mean and covariance sometimes will be used hereafter as abbreviations for state vector and error matrix, respectively, in a manner consistent with colloquial usage in the fields of estimation, filtering, and data fusion.
The data fusion problem for mean and covariance estimates is exemplified by the case of two conservative estimates {a, A} and {b,B}, generated from different sensors, of the state of an autonomous vehicle. In order to steer the vehicle it is necessary to fuse the given estimates into a single estimate to be used by the steering mechanism. A trivial (though apparently never described in the literature) approach for fusing the two estimates would be to produce a fused estimate that consists only of the estimate having the “smaller” covariance matrix, in terms of some measure such as the trace or determinant, and disregard any information contained in the other estimate. In order to avoid having to disregard information from one of the estimates, it is necessary to have a means for generating a fused estimate that combines information from both given estimates and is still guaranteed to be conservative.
2.3 The Kalman Filter Update Equation:
The method that is most commonly used for fusing mean and covariance estimates is the Kalman update equation, which is the fundamental component of the well known Kalman filter. The Kalman update equation is known to generate an optimal conservative fused estimate {c, C} from given estimates {a, A} and {b, B} as long as the errors associated with a and b are not correlated. If the errors are correlated, however, then the Kalman estimate is not guaranteed to be conservative. An enhancement to the Kalman update equation has been developed for the case when the exact degree of correlation (defined by a cross covariance matrix) is known, but it cannot be applied in the general case when the degree of correlation is unknown.
A common approach for addressing the problem of fusing correlated estimates is to use the Kalman update equation, knowing that it may yield an underestimated covariance matrix, and then enlarge the resulting covariance matrix by some ad hoc heuristic method so that it can be assumed conservative. In many problem areas, such as in large decentralized data fusion networks, the problem is most acute because no amount of heuristic tweaking can avoid the limitations of the Kalman update algorithm. This is of enormous consequence because of the general trend t
Julier Simon Justin
Uhlmann Jeffrey Kent
No associations
LandOfFree
Method and apparatus for fusing signals with partially known... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for fusing signals with partially known..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for fusing signals with partially known... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3281808