System for detection and estimation of periodic patterns in a noisy signal

• US 7,610,324 B2
• Filed: 07/08/2004
• Issued: 10/27/2009
• Est. Priority Date: 07/10/2003
• Status: Expired due to Fees

• Current Assignee: Arnon Breuer, Lidror Troyansky
• Original Assignee: Arnon Breuer, Lidror Troyansky
• Inventors: Breuer, Arnon, Troyansky, Lidror
• Primary Examiner(s): Mai; Tan V
• Application Number:
  US10/885,605
• Publication Number:
  US 20050033146A1
• Time in Patent Office: 1,937 Days
• Field of Search: 708/422, 708/426
• US Class Current: 708/422
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle
• CPC Class Codes: A61B 5/726 using Wavelet transforms
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 2218/08 Feature extraction

Abstract

A method for the detection and estimation of periodicities in noisy signals. First the signal is sampled and preprocessed in a discrete succession of equal periods. Then, short term characteristics vectors are generated for each period. The short term characteristics vectors are then grouped into sets, followed by the obtainment of at least one representative vector for each set. Periodic structures are estimated using the representative vectors. In a preferred embodiment, the short term characteristics are the the short term spectra of the signal. Practically, an autocorrelation matrix is constructed to assess autocorrelation along a diagonal.

Claims

1. A method for discerning periodicities in signals, comprising the succession of steps of:

sampling and pre-processing of a raw-signal in discrete succession of equally spaced slices of time, the sampling and pre-processing being respectively executed by a sampling device and an analog pre-processor;

generating short-term characteristics vectors for such slice of time;

grouping said-short term characteristics vectors into sets;

obtaining at least one representative vector for each of said sets of short-term characteristics vectors; and

estimating periodic structures using said representative vectors,

wherein said generating, grouping, obtaining, and estimating are executed by computerized processors.

2. A method for discerning periodicities in signals as in claim 1, wherein said short-term characteristics vectors are grouped into sets each having an initial time and a period, and further:

reducing each of said at least one of said representative vectors into one number respectively;

constructing a vector of the plurality of said numbers; and

estimating periodicities in said signal,

wherein said reducing, constructing, and estimating periodicities are executed by at least one computerized processor.

3. A method for discerning periodicities in signals as in claim 2, wherein said reducing is achieved by estimating at least one statistic of said vector with respect to its elements.

4. A method for discerning periodicities in signals as in claim 3, wherein said at least one statistic is selected from a group containing mean, variance, skew, kurtosis, and entropy.

5. A method for discerning periodicities in signals as in claim 1, wherein said short-term characteristics are the short-term spectra of the signal.

6. A method for discerning periodicities in signals as in claim 1, wherein said short-term characteristics are the short-term continuous-wavelets-transform (CWT) of the signal.

7. A method for discerning periodicities in signals as in claim 1, wherein each statistical characteristic of a corresponding set of vectors is averaged with respect to the initial time segment of the set of characteristic.

8. A method for discerning periodicities in signals as in claim 1 that further comprises the steps of:

dividing the set of vectors into at least two clusters;

obtaining a representative vector for each of the clusters;

reducing each representative vector to one number; and

taking the largest number as the output number,

wherein said dividing, obtaining a representative vector, and taking are executed by at least one computerized processor.

9. A method for discerning periodicities in signals as in claim 1, wherein said short-term characteristics vectors are grouped into sets each having an initial time and a period, and further:

reducing each said representative vector of at least one set of said sets of short-terms vectors into a number;

constructing at least two vectors of a plurality of said numbers; and

estimating periodicities in each of the said vectors,

wherein said reducing, constructing, and estimating periodicities are executed by at least one computerized processor.

10. A method for discerning periodicities in signals as in claim 9, wherein reducing said representative vectors of said at least one set of vectors into at least two numbers is achieved using a method comprising the succession of steps of:

dividing the set of vectors into at least two clusters; and

obtaining a representative vector with respect to each of said clusters.

11. A method for discerning periodicities in signals, comprising the steps of:

sampling a signal of interest using a sampling device;

pre-processing said sampled signal using an analog pre-processor;

obtaining short-term spectra of said signal at equally spaced in slices of time;

constructing an autocorrelation matrix between said short term spectra; and

associating each secondary diagonal of said matrix with a time lag,

wherein said obtaining, constructing, and associating are executed by computerized processors.

12. A method for discerning periodicities in signals as in claim 11, wherein at least one statistic of the elements of said diagonals is used to assess the auto-correlation along a diagonal.

13. A method for discerning periodicities in signals as in claim 12, wherein said at least one statistic is selected from a group containing mean, variance, skew, kurtosis, and entropy.

14. An apparatus for estimating periodicities in a signal, comprising:

an analog preprocessor that processes a sampled signal in discrete succession of equally spaced slices of time, using a band-pass filter on a wide-band signal; and

one or more processors that

generate short-term characteristics vectors for such slice of time,

group the short term characteristics vectors into set;

obtain at least one representative vector for each of said sets of short-term characteristics vectors, and

estimating of periodic structures using said representative vectors.