Neural network method and apparatus for retrieving signals embedded in noise and analyzing the retrieved signals

• US 5,402,520 A
• Filed: 03/06/1992
• Issued: 03/28/1995
• Est. Priority Date: 03/06/1992
• Status: Expired due to Term

• Current Assignee: Noiseout, Inc.
• Original Assignee: Bonnie S. Schnitta
• Inventors: Schnitta, Bonnie S.
• Primary Examiner(s): Downs, Robert W.
• Assistant Examiner(s): Hafiz, Tariq R.
• Application Number:
  US07/847,776
• Time in Patent Office: 1,117 Days
• Field of Search: 395/22, 395/24, 395/2, 395/27, 395/2.35, 395/2.42, 395/21
• US Class Current: 706/16
• CPC Class Codes: G01S 7/292 Extracting wanted echo-signals
• CPC Class Codes: G01S 7/417 involving the use of neural networks
• CPC Class Codes: G01V 1/01 Measuring or predicting earthquakes
• CPC Class Codes: G01V 1/288 Event detection in seismic signals, e.g. microseismics G01V1/36 takes precedence
• CPC Class Codes: G06F 2218/04 Denoising
• CPC Class Codes: H03H 21/0012 Digital adaptive filters
• CPC Class Codes: H03H 2222/04 using neural networks
• CPC Class Codes: Y10S 706/925 Business
• CPC Class Codes: Y10S 706/929 Geological, e.g. seismology

Abstract

An apparatus for retrieving signals embedded in noise and analyzing the signals. The apparatus includes an input device for receiving input signals having noise. At least one noise filter retrieves data signals embedded in the input signals. At least one adaptive pattern recognition filter generates coefficients of a polynomial expansion representing the pattern of the filtered data signals. A storage device stores the coefficients generated. It is determined when an event has occurred, the event being located at any position within the data signals. An adaptive autoregressive moving average pattern recognition filter generates coefficients of a polynomial expansion representing an enhanced pattern of filtered data signals. At least one weighting filter compares the stored patterns with the enhanced pattern of data signals.

Claims

1. An apparatus for retrieving signals embedded in noise and analyzing the signals, comprising

an input device for receiving a plurality of input signals comprising noise;

noise filtering means, including at least one noise filter, for retrieving data signals embedded in the plurality of input signals received from said input device;

first pattern recognition means, including at least one adaptive pattern recognition filter, for generating first coefficients of a polynomial expansion representing the data signals received from said noise filtering means;

first storage means for storing the first coefficients generated by said first pattern recognition means;

event identifying means for calculating a predicted value of a next data signal based on the first coefficients stored in said first storage means, for comparing a next data signal to the predicted value to determine when an event has occurred, said event located at any position within the data signals, and for ascertaining an exact moment of the event;

second pattern recognition means, including an adaptive autoregressive moving average pattern recognition filter, for generating second coefficients of a polynomial expansion representing the data signals received from said noise filtering means;

second storage means for storing a plurality of predefined stored patterns each representing a known event type and having a known event location;

weighting means, including at least one weighting filter, for retrieving the plurality of predefined stored patterns from said second storage means, aligning the known event location of each of the predefined stored patterns with the exact moment of the even of the data signals and comparing the plurality of predefined stored patterns with the second coefficients generated by the second pattern recognition means, and for generating weights for each of the plurality of stored patterns based on said comparison; and

an output device for outputting output signals representing the weights generated by said weighting means.

2. The apparatus of claim 1 further comprising

ranking means for rearranging the output signals according to weight and for testing the output signals against existing rules;

and

means for outputting the ranked output signals to the output device.

3. The apparatus of claim 1 wherein a first one of said at least one noise filter comprises an adaptive noise filter.

4. The apparatus of claim 3 further including second noise filter means for retrieving data signals embedded in the input signals received from said input device, wherein said second noise filter means comprises a nonadaptive noise filter.

5. The apparatus of claim 4 further including third noise filter means for retrieving data signals embedded in the input signals received from said input device, wherein said third noise filter means includes a transparent filter.

6. The apparatus of claim 1 wherein one of said at least one adaptive autoregressive pattern recognition filters comprises a hybrid image processing system (HIPS) filter.

7. The apparatus of claim 1 wherein one of said at least one adaptive autoregressive pattern recognition filters comprises a transient acoustic processing system (TRAPS) filter.

8. The apparatus of claim 1 wherein said at least one adaptive autoregressive pattern recognition filters includes a HIPS filter and a TRAPS filter.

9. The apparatus of claim 1 wherein said second pattern recognition means includes a self adaptive and variant error (SAVE) filter.

10. A method for retrieving signals embedded in noise and analyzing the signals comprising the steps of:

inputting a plurality of signals comprising noise;

filtering said plurality of input signals to retrieve a plurality of filtered data signals from the input signals;

passing said plurality of filtered data signals through at least one adaptive pattern recognition filter to generate a first set of coefficients of a polynomial expansion representing the plurality of filtered data signals;

storing the first set of coefficients;

calculating a predicted value of a next signal based on the first set of coefficients;

comparing a next data signal to the predicted value to determine when an event has occurred;

ascertaining an exact moment of the event for the plurality of filtered data signals;

passing said plurality of filtered data signals through at least one adaptive autoregressive moving average pattern recognition filter to generate a second set of coefficients of a polynomial expansion representing said plurality of filtered data signals;

retrieving a plurality of stored patterns from a storage device, each one of the plurality of stored patterns representing a known event type and having a known event location;

aligning the known event location of each of the plurality of stored patterns with the exact moment of the event of the plurality of filtered data signals;

comparing each one of the plurality of stored patterns with the second set of coefficients to generate weights for each of the plurality of stored patterns; and

outputting signals representing the generated weights.

11. The method of claim 10 further comprising the step of:

rearranging the output signals according to weight, testing the output signals against existing rules, and outputting the output signals in a ranked order.

12. The method of claim 10 wherein said step of passing said plurality of input signals through at least one noise filter includes passing said signals through an adaptive noise filter.

13. The method of claim 12 wherein said step of passing said plurality of input signals through at least one noise filter further includes passing said signals through a nonadaptive noise filter.

14. The method of claim 13 wherein said step of passing said plurality of input signals through at least one noise filter further includes passing said signals through a transparent noise filter.

15. The method of claim 10 wherein said step of passing said plurality of filtered data signals through at least one adaptive autoregressive pattern recognition filter includes passing said signals through a hybrid image processing system (HPS) filter.

16. The method of claim 10 wherein said step of passing said plurality of filtered data signals through at least one adaptive autoregressive pattern recognition filter includes passing said signals through a transient acoustic processing system (TRAPS) filter.

17. The method of claim 10 wherein said step of passing said plurality of filtered data signals through at least one adaptive autoregressive pattern recognition filter includes passing said signals through a HIPS filter and a TRAPS filter.

18. The method of any of claims 10, 15, 16, or 17 further comprising the step of:

outputting signals representing the generated coefficients of the polynomial expansion representing the pattern of data signals.

19. The method of claim 10 further comprising the step of:

outputting a signal flag when an event is determined to have occurred.

20. The method of claim 10 wherein said step of passing said plurality of filtered data signals through at least one adaptive autoregressive moving average pattern recognition filter further comprises the step of passing said signals through a self adaptive and variant error (SAVE) filter.

21. The method of claim 10 further comprising the steps of:

inputting to the input device a second plurality of signals containing information related to the first plurality of input signals; and

passing said second plurality of signals through at least one noise filter to retrieve discriminant signals embedded therein;

and wherein said step of ascertaining the exact moment of the event t_c further comprises the step of comparing each discriminant signal to the previous discriminant signals to determine the presence of a marked change in the value of said discriminant signals.

22. The method of any of claims 11, 16, 17, 17, 20, or 21 wherein the plurality of signals comprising noise represent prices of stock as a function of time.

23. An apparatus for retrieving signals embedded in noise and analyzing the signals, comprising

an input device for receiving a plurality of input signals comprising noise;

noise filtering means for retrieving a plurality of data signals embedded in the plurality of input signals received from said input device;

pattern recognition means for generating coefficients of a polynomial expansion representing the plurality of data signals received from said noise filtering means;

first storage means for storing the coefficients generated by the pattern recognition means;

event identifying means if or calculating a predicted value of a next data signal based on the coefficients stored in said first storage means, for comparing a next data signal to the predicted value to determine when an event has occurred, said event located at any position within the data signals, and for ascertaining an exact moment of the event;

second storage means for storing a plurality of predefined stored patterns each representing a known event type and having a known event location;

weighting means for retrieving the plurality of predefined stored patterns from said second storage means, aligning the known event location of each of the predefined stored patterns with the exact moment of the event of the plurality of data signals and comparing the plurality of predefined stored patterns with the coefficients generated by the pattern recognition means, and for generating weights for each of the plurality of stored patterns based on said comparison; and

an output device for outputting signals representing the weights generated by said weighting means.

24. The apparatus of claim 23 wherein the pattern recognition means includes at least one transparent filter.

25. The apparatus of claim 23 wherein the pattern recognition means includes at least one adaptive pattern recognition filter.

26. The apparatus of claim 23 wherein the pattern recognition means includes at least one adaptive autoregressive moving average pattern recognition filter.

27. An apparatus for retrieving signals embedded in noise and analyzing the signals, comprising

an input device for receiving a plurality of input signals comprising noise;

a noise filter for retrieving a plurality of data signals embedded in the plurality of input signals received from the input device;

at least one pattern recognition filter for generating coefficients of a polynomial expansion representing the plurality of data signals received from the noise filter;

a storage device for storing the coefficients generated by the pattern recognition filter and for storing a plurality of predefined stored patterns each representing a known event type and having a known event location;

means for calculating a predicted value of a data signal based on the coefficients;

means for comparing at last one of the plurality of data signals to the predicted value to determine when an event has occurred, said event located at any position within the data signals;

means for ascertaining an exact moment of the event;

means for retrieving the plurality of predefined stored patterns from the storage device;

means for aligning the known event location of each of the predefined stored patterns with the exact moment of the event of the plurality of data signals;

means for comparing the plurality of predefined stored patterns with the coefficients and generating weights for each of the plurality of stored patterns; and

an output device for outputting signals representing the weights generated.

28. The apparatus of claim 27 wherein the at least one pattern recognition filter comprises at least one transparent filter.

29. The apparatus of claim 27 wherein the at least one pattern recognition filter comprises at least one adaptive pattern recognition filter.

30. The apparatus of claim 27 wherein the at least one pattern recognition filter comprises at least one adaptive autoregressive moving average pattern recognition filter.

31. An method for retrieving signals embedded in noise and analyzing the signals, comprising the steps of:

receiving a plurality of input signals comprising noise;

filtering the plurality of input signals to obtain a plurality of data signals;

filtering the plurality of data signals to generate coefficients of a polynomial expansion representing the plurality of data signals;

calculating a predicted value of a data signal based on the coefficients;

comparing at least one of the plurality a data signals to the predicted value to determine when an event has occurred, said event located at any position within the data signals;

ascertaining an exact moment of the event;

providing a plurality of predefined stored patterns each representing a known event type and having a known event location;

aligning the known event location of each of the predefined stored patterns with the exact moment of the event of the plurality of data signals;

comparing the plurality of predefined stored patterns with the coefficients and generating weights for each of the plurality of stored patterns; and

outputting signals representing the weights generated.

32. The method of claim 31 wherein the step of filtering the plurality of data signals further comprises the step of filtering the plurality of data signals using at least one adaptive autoregressive moving average pattern recognition filter.

33. The method of claim 31 wherein the step of filtering the plurality of data signals further comprises the step of filtering the plurality of data signals using at least one adaptive pattern recognition filter.