Data processing: measuring – calibrating – or testing – Measurement system – Measured signal processing
Reexamination Certificate
1998-11-13
2002-06-04
Hilten, John S. (Department: 2854)
Data processing: measuring, calibrating, or testing
Measurement system
Measured signal processing
C702S022000, C702S030000
Reexamination Certificate
active
06401057
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to signal and pattern data detection and classification and, more particularly to data detection and classification using estimated nonlinear correlation parameters and dynamical correlation parameters that reflect possible deterministic properties of the observed data.
BACKGROUND OF THE INVENTION
Existing signal data detection and classification techniques generally use linear models derived from an integro-differential operator such as an ordinary differential equation or a partial differential equation. A set of model parameters are estimated using an optimization technique that minimizes a cost function, e.g. the least squares optimization technique. The model parameters can be used to replicate the signal and classification is applied to the replicated model.
Such signal data detection and classification techniques based on integro-differential operators are computationally intensive and have difficulty in environments having high levels of background noise or interfering signals. Existing techniques also typically fail to take advantage of physical information possibly derived from a signal generating system.
Accordingly, there exists a need for signal and pattern data detection and classification techniques that are computationally efficient and that can provide performance in environments subject to relatively high levels of background noise or interfering signals. The present invention satisfies these needs.
SUMMARY OF THE INVENTION
The present invention is a data detection and classification system for revealing aspects of information or observed data signals that reflect deterministic properties of the data signals. The deterministic properties of an observed data signal may be efficiently estimated, according to the invention, using correlation parameters based on nonlinear dynamical principles. The system is particularly advantageous for detecting and classifying observed data signals provided by complicated nonlinear dynamical systems and processes, which data signals may be spectrally broadband and very difficult to detect using standard signal processing and transform techniques.
The invention is embodied in a method, and related apparatus, for detecting and classifying signals in which a data signal is acquired from a dynamical system, normalized, and used to calculate at least one of a nonlinear or dynamical correlation coefficient. The correlation coefficient may result from a correlation between the normalized data signal and a derivative of the normalized data signal, or from a correlation between the normalized data signal and an exponent of the normalized signal wherein the exponent may be an integer of 2 or greater. Further, the data signal may be normalized to zero mean and unit variance.
Alternatively, the invention may be embodied in an apparatus, and related method, for processing an input signal wherein the apparatus has a first differentiator, a delay circuit and a first correlator. The first differentiator receives the input signal and generates a first derivative signal which is based on a derivative of the input signal. The delay circuit delays the input signal by a predetermined time period to generate a delayed signal and the first correlator correlates the delayed signal with the first derivative signal to generate a first correlated signal. The apparatus may further or alternatively include a first function generator and a second correlator. The first function generator may receive the input signal and generate a first processed signal based on a first predetermined function and the second correlator correlates the delayed signal with the first processed signal to generate a second correlated signal. Alternatively, the first function generator may generate the first processed signal using the first derivative signal. Further, the first predetermined function may be a square of the input signal received by the first function generator. Also, the correlated signals may be further processed to detect deterministic properties in the input signal.
The invention may be also embodied in a method, and related apparatus, for processing an analog input signal in which the analog input signal is digitized to generate a digital input signal and then normalized to a normalized digital signal. Next, correlation signals are calculated based on the normalized signal to generate a correlation matrix and a derivative of the correlation matrix is calculated to generate a derivative correlation coefficient matrix. Estimating coefficients are then calculated based on the correlation matrix and the derivative correlation coefficient matrix.
Other features and advantages of the present invention should be apparent from the following description of the preferred embodiments, taken in conjunction with the accompanying drawings, which illustrate, by way of example, the principles of the invention.
REFERENCES:
patent: 5057992 (1991-10-01), Traiger
patent: 5277053 (1994-01-01), McLane et al.
patent: 5652713 (1997-07-01), Chandler
patent: 5867807 (1999-02-01), Yamada et al.
patent: 5884245 (1999-03-01), Chandler
patent: 6278961 (2001-08-01), Kadtke et al.
patent: 2000059332 (2000-02-01), None
Broomhead and King, “Extracting Qualitative Dynamics From Experimental Data”Physica20D:217-236 (1986).
Crutchfield and McNamara, “Equations of Motion from a Data Series”Journal of Complex Systems 11-26 (1987).
Dzwinel, W., “How To Make Sammon's Mapping Useful For Multidimensional Data Structures Analysis”Pattern Recognition27 (7):949-959 (1994).
Gouesbet and Letellier, “Global vector-field reconstruction by using a multivariate polynominal L2approximation on nets”Physical Review E49 (6):4955-4972 (1944).
Kadtke and Kremliovsky, “Estimating statistics for detecting determinism using global dynamical models”Physics Letters A229:97-106 (1997).
Kadtke, J., “Classification of highly noisy signals using global dynamical models”Physics Letters A203:196-202 (1995).
Meraim, et al., “Blind System Identification”Proceedings of the IEEE85 (8):1310-1322 (1997).
Proakis et al.,Introduction to Digital Signal Proceedings, pp. 795-849, Macmillan (1988).
Ray and Turner, “Mahalanobis Distance-Based Two New Feature Evaluation Criteria”Information Sciences60:217-245 (1992).
Rössler, O.E., “An Equation For Continuous Chaos”Physics Letters57A (5):397-398 (1976).
Rotman, J., “Flipper's Secret”New Scientistfl-85:34-39 (1997).
Streit and Luginbuhl, “Maximum Likelihood Training of Probabilistic Neural Networks”IEEE Transactions on Neural Networks5 (5):764-783 (1994).
Kadtke James B.
Kremliovsky Michael N.
Fawcett Robroy R.
Hilten John S.
Nolan, Jr. Charles H.
Nonlinear Solutions, Inc.
LandOfFree
Detection and classification system for analyzing... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Detection and classification system for analyzing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Detection and classification system for analyzing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2956405