Signal recognition system and method

• US 5,210,820 A
• Filed: 05/02/1990
• Issued: 05/11/1993
• Est. Priority Date: 05/02/1990
• Status: Expired due to Term

• Current Assignee: Nielsen Company (US) LLC (RR Donnelley & Sons Company)
• Original Assignee: Broadcast Data Systems Limited Partnership
• Inventors: Kenyon, Stephen C.
• Primary Examiner(s): Knepper, David D.
• Application Number:
  US07/517,835
• Time in Patent Office: 1,105 Days
• Field of Search: 364/487, 364/604, 364/513.5, 358/84, 381/41-43, 395/2
• US Class Current: 704/200
• CPC Class Codes: G06F 2218/00 Aspects of pattern recognition specially adapted for signal processing
• CPC Class Codes: G10H 1/0033 Recording/reproducing or transmission of music for electrophonic musical instruments
• CPC Class Codes: G10H 2240/141 Library retrieval matching, i.e. any of the steps of matching an inputted segment or phrase with musical database contents, e.g. query by humming, singing or playing; the steps may include, e.g. musical analysis of the input, musical feature extraction, query formulation, or details of the retrieval process
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates

Abstract

Recognizing a signal begins by receiving the signal and sampling it at a plurality of sampling points to produce a plurality of signal values. A statistical moment of the signal is calculated using the signal values according to the formula: ##EQU1## where: N=the number of sampling points;

n=1<n<N

X=the signal value of the signal at the sampling point;

μ=a mean of the signal values;

σ=a standard deviation of the signal values; and

k=an integer greater than 1.

The calculated statistical moment is then compared with a plurality of stored signal identifications and the received signal is recognized as similar to one of the stored signal identifications. Preferably, the calculated statistical moment is used to create a feature vector which is quantized, and a weighted sum of the quantized feature vector is used to access a memory which stores the signal identifications.

Claims

1. A method of broadcast signal recognition, comprising the steps of:

receiving a broadcast signal and converting it into a plurality of electrical signals having different frequency bands;

sampling each of the plurality of electrical signals at a plurality of sampling points to produce a plurality of signal values for each frequency band;

calculating a plurality of statistical moments corresponding to the plurality of frequency bands, each statistical moment being calculated according to the formula;

##EQU9## where;

n=the number of sampling points;

n=1<

n<

N

X=the signal value of said electrical signal at a sampling point;

μ

=a mean of the signal values;

σ

=a standard deviation of the signal values; and

k=an integer greater than 1;

recognizing the received signal as similar to at least one of said stored signal identifications when said comparison step produces an indication that the calculated statistical moments are similar to the at least one stored signal identifications, and outputting a signal indicating that the received broadcast signal is similar to the at least one stored signal identification.

2. A method according to claim 1 wherein said calculating step includes the steps of calculating skew and kurtosis values, and wherein said comparing step includes the step of forming a feature vector from said skew and said kurtosis.

3. A method according to claim 1 wherein said receiving step includes the steps of:

bandpass filtering the received broadcast signal to provide the plurality of electrical signals;

rectifying said plurality of electrical signals;

low pass filtering the rectified signals; and

computing a plurality of linear combinations of the low pass filtered signals.

4. A method according to claim 3 wherein said sampling step includes the steps of:

sampling a first one of said linear combinations at the plurality of sampling points to produce a first plurality of signal values; and

sampling a second one of said linear combinations at the plurality of sampling points to produce a second plurality of signal values.

5. A method according to claim 4 wherein said calculating step includes the steps of:

calculating a first skew value and a first kurtosis value from said first plurality of signal values;

calculating a second skew value and a second kurtosis value from said second plurality of signal values; and

deriving a feature vector comprising said first and second skew values and said first and second kurtosis values.

6. A method according to claim 5 wherein said comparing step includes the steps of:

quantizing said feature vector in accordance with a plurality of predetermined quantization patterns to provide a quantized vector; and

accessing said library with said quantized vector to locate a signal identification corresponding to the quantized vector.

7. A method according to claim 6 wherein said accessing step includes the steps of:

forming a weighted sum of the values of said quantized vector using a non-decimal radix;

using the weighted sum as an address to access a pointer table to locate a pointer corresponding to said weighted sum; and

using said pointer to locate a signal identification list containing the signal identification corresponding to the quantized vector.

8. A method according to claim 7 wherein said recognizing step includes the steps of:

correlating a third one of said plurality of linear combinations with a plurality of stored signals which correspond to the signal identifications contained in said signal identification list; and

selecting one stored signal whose correlation with said third linear combination exceeds a predetermined threshold.

9. A method according to claim 6 wherein said quantizing step includes the steps of:

categorizing each skew value of said feature vector in accordance with a predetermined skew quantization pattern having categories non-uniformly distributed therein; and

categorizing each kurtosis value of said feature vector in accordance with a predetermined kurtosis quantization pattern having categories non-uniformly distributed therein.

10. A method of creating a signal library useful in broadcast signal recognition, comprising the steps of:

receiving a reference signal to be broadcast, and converting the reference signal into a plurality of electrical signals having different frequency bands;

sampling each of the plurality of electrical signals at a plurality of sampling points to produce a plurality of signal values for each frequency band;

calculating a plurality of statistical moments corresponding to the plurality of frequency bands using the plurality of signal values, each statistical moment being calculated according to the formula;

##EQU10## where;

N=the number of sampling points;

n=1<

n<

N

X=the sampled signal value at a sampling point;

μ

=a mean of the sampled signal values;

σ

=a standard deviation of the sampled signal values; and

k=an integer greater than 1;

deriving a feature vector from the calculated statistical moments; and

storing the feature vector or a representation thereof in a broadcast signal memory.

11. A method according to claim 10 wherein said calculating step includes the step of calculating two statistical moments of the received reference signal, and wherein said deriving step includes the step of deriving said feature vector from both of the calculated statistical moments.

12. A method according to claim 10 wherein said receiving step includes the steps of:

bandpass filtering the received signal to provide the plurality of electrical signals;

rectifying said plurality of signals;

low pass filtering the rectified signals; and

computing a plurality of linear combinations of the low pass filtered signals.

13. A method according to claim 12 wherein said sampling step includes the steps of:

sampling a first portion of one of said linear combinations at a plurality of points to produce a first plurality of signal values; and

sampling a second portion of said one linear combination at a plurality of points to produce a second plurality of signal values.

14. A method according to claim 13 wherein said calculating step includes the steps of:

calculating the skew and kurtosis of said first plurality of signal values to provide a first skew value and a first kurtosis value; and

calculating the skew and kurtosis of said second plurality of signal values to provide a second skew value and a second kurtosis value.

15. A method according to claim 14 wherein said deriving step includes the steps of:

forming a first feature vector including said first skew value and said first kurtosis value; and

forming a second feature vector including said second skew value and said second kurtosis value.

16. A method according to claim 15, wherein said deriving step further includes the step of quantizing both said feature vectors by replacing the skew and kurtosis values with integers in accordance with predetermined non-uniform quantization pattern to produce first and second quantized vectors.

17. A method according to claim 16 wherein said storing step includes the steps of:

computing a weighted sum of said first vector using a non-decimal radix;

computing a weighted sum of said second vector using said non-decimal radix;

using the weighted sums to address said memory; and

storing in said memory, at the addresses corresponding to the weighted sums, a signal identification code corresponding to said reference signal.

18. A method according to claim 15 wherein said deriving step further includes the steps of:

producing first and second overlap quantized vectors from said first feature vector in accordance with predetermined non-uniform overlap quantization patterns;

producing third and fourth overlap quantized vectors from said second feature vector in accordance with said predetermined non-uniform overlap quantization patterns;

forming a first plurality of permutations of said first and second overlap quantized vectors; and

forming a second plurality of permutations of said third and fourth overlap quantized vectors.

19. A method according to claim 18 wherein said storing step includes the steps of:

computing a weighted sum of the values of each one of said first and second pluralities of permutations;

accessing said memory at addresses corresponding to said weighted sums; and

storing in said memory, at areas corresponding to said addresses, a signal identification code identifying said reference signal.

20. A method of classifying a broadcast signal comprising the steps of:

receiving the broadcast signal and converting it into a plurality of electrical signals having different frequency bands;

sampling each of the plurality of electrical signals for each frequency band at a plurality of points to produce a plurality of signal values;

deriving a multi-value feature vector from said plurality of signal values derived from said plurality of electrical signals;

quantizing said feature vector by replacing each feature vector value with a category value determined from a predetermined quantization pattern which non-uniformly categorizes feature vector values; and

accessing a predetermined broadcast signal library with the quantized vector and outputting from said library a signal identification code corresponding to the received signal.

21. A method according to claim 20 wherein said predetermined quantization pattern is an overlap quantization pattern.

22. A method according to claim 20, wherein said quantizing step utilizes a plurality of different predetermined quantization patterns to quantize said feature vector.

23. A method according to claim 20 wherein said deriving step includes the steps of:

calculating two different statistical moments of said plurality of signal values; and

forming said feature vector using the two calculated statistical moments.

24. A method according to claim 23 wherein a predetermined quantization pattern is provided for each of the two statistical moments.

25. A method according to claim 23 wherein said two different statistical moments comprise skew and kurtosis.

26. A method according to claim 20 wherein said receiving step includes the step of envelope detecting said received signal.

27. A method according to claim 20 wherein said quantization step includes the steps of:

replacing said feature vector with a plurality of quantized vectors in accordance with at least one predetermined quantization pattern having overlapping categories; and

permuting said plurality of quantized vectors to produce a further plurality of permutated quantized vectors.

28. A method according to claim 27 wherein said accessing step comprises the steps of:

accessing said signal library with said plurality of permuted quantized vectors to reach a plurality of memory locations; and

writing in each said memory location a signal identification code corresponding to the received signal.

29. A method according to claim 20 wherein said receiving step includes the steps of:

spectrally analyzing said signal to provide the plurality of electrical signals having different frequency bands; and

forming linear combinations of said plurality of electrical signals;

and wherein said sampling step includes the step of sampling each linear combination to provide a plurality of sampling points for each linear combination;

and wherein said deriving step includes the steps of;

calculating skew and kurtosis values for each linear combination from the corresponding sampling points; and

forming said feature vector to include the skew and kurtosis values from all of said linear combinations.

30. A method of preparing a broadcast signal identification library useful in identification of broadcast signals, comprising the steps of:

receiving a signal to be broadcast and converting it into a plurality of electrical signals having different frequency bands;

sampling each of the plurality of electrical signals to provide a plurality of signal values for each frequency band;

deriving a plurality of multi-value feature vectors from said plurality of signal values;

quantizing each feature vector, including the substeps of;

establishing a plurality of quantization levels;

distributing the quantization levels non-uniformly over a predetermined statistical distribution of vector quantization values;

deriving a plurality of quantization thresholds in accordance with the distributed quantization levels; and

replacing each value of each multi-value feature vector with a corresponding quantization value determined by said quantization thresholds; and

storing, in a broadcast signal memory, a value representing each quantized vector as the signal identification for the corresponding received signal.

31. A method according to claim 30 wherein the step of distributing the quantization levels includes the step of distributing the quantization levels in an overlapped manner, and wherein the step of replacing each value of each feature vector includes the further steps of:

replacing each value of each feature vector with a plurality of quantization values determined by the overlapped quantization levels, each feature vector thus being replaced with a plurality of quantized vectors; and

permuting the plurality of quantized vectors of each feature vector to produce a plurality of permuted vectors;

and wherein said step of storing includes the step of storing, for each sampled signal, values corresponding to the plurality of quantized vectors.

32. A method according to claim 30 wherein said deriving step includes the step of calculating, for each of the plurality of electrical signals, two statistical moments, and forming each feature vector to include the two statistical moments.

33. A method of identifying broadcast information, comprising the steps of:

receiving broadcast information and converting it into an electrical signal;

spectrally analyzing the electrical signal to produce a plurality of analyzed waveforms having different frequency bands;

forming a plurality of linear combinations of said analyzed waveforms;

computing at least one statistical moment for each said linear combination;

deriving a feature vector from the computed statistical moments;

quantizing the feature vector;

using the quantized feature vector to assess a broadcast signal memory storing a plurality of identifications of broadcast information; and

identifying the received broadcast information based on the stored identifications, and outputting a signal corresponding to the received broadcast information.

34. A method according to claim 33 wherein said computing step includes the steps of:

sampling each analyzed waveform at a plurality of sampling points to provide a plurality of sampled values; and

calculating, for each analyzed waveform, two different statistical moments according to the formula;

##EQU11## where;

N=the number of sampling points;

n=1<

n<

N

X=the sampled value at a sampling point;

μ

=a mean of the sampled values;

σ

=a standard deviation of the sampled values; and

k=an integer greater than 1, a different value of k being provided for each of the two different statistical moments.

35. A method according to claim 34 wherein said deriving step includes the step of forming each feature vector to include values corresponding to each of the two calculated statistical moments.

36. A method according to claim 33 wherein said quantizing step includes the step of replacing each value of said feature vector with a quantization value derived from a predetermined non-uniform quantization pattern.

37. A method according to claim 36 wherein said using step includes the step of extracting from said memory a list of identifications of broadcast information, and wherein said identifying step includes the steps of:

correlating one of said analyzed waveforms with a plurality of information codes stored in a correlation library in said memory, the plurality of information codes corresponding to the list of identifications extracted in said using step; and

declaring the broadcast information to be similar to one of the plurality of information codes when the correlation therebetween exceeds a predetermined threshold.

38. A method of classifying broadcast information, comprising the steps of:

receiving broadcast information and converting it into a received electrical signal;

performing a spectral analysis of said received electrical signal and providing a plurality of electrical signals having different frequency bands;

computing a plurality of linear combinations of said plurality of electrical signals;

calculating a statistical moment for each said linear combination; and

classifying the received broadcast information based on the calculated statistical moments and a plurality of predetermined broadcast signal identifications stored in a broadcast signal memory.

39. A method according to claim 38 wherein said calculating step includes the steps of:

calculating two higher-order statistical moments of each said linear combination; and

forming a feature vector incorporating the calculated statistical moments for each linear combination;

and wherein said classifying step includes the step of classifying the received broadcast information based on the formed feature vectors.

40. A method of classifying a broadcast signal comprising the steps of:

receiving the broadcast signal and converting it into an electrical signal;

bandpass filtering the converted electrical signal to provide a plurality of filtered signals having different frequency passbands;

rectifying and low pass filtering the filtered signals to provide a plurality of low bandwidth signals;

computing a plurality of linear combinations of said low bandwidth signals;

sampling several of the linear combinations to provide several sets of sampled values;

calculating a skew value and a kurtosis value for each set of sampled values;

forming a feature vector form the skew and kurtosis values of the several sets of sampled values;

quantizing the feature vector by replacing each skew and kurtosis value with a category value derived from a plurality of stored predetermined non-uniform quantization patterns;

using the quantized vector to access a first broadcast signal memory data base storing a plurality of signal identification lists, each list capable of storing a plurality of signal identification codes;

extracting from the first data base the signal identification list corresponding to the quantized vector;

extracting from a second data base stored correlation signals corresponding to the signals in the signal identification list extracted from the first data base;

correlating the extracted correlation signals with one of the linear combinations computed in said computing step; and

classifying the broadcast information as similar to one of said correlation signals when its correlation with said one linear combination exceeds a predetermined threshold.

41. Broadcast signal recognition apparatus, comprising:

means for receiving a broadcast signal and converting it into a plurality of electrical signals having different frequency bands; and

processing means for (a) sampling each of the plurality of electrical signals at a plurality of sampling points to produce a plurality of signal values for each frequency band, (b) calculating a plurality of statistical moments corresponding to the plurality of frequency bands using the plurality of signal values, each statistical moment being calculated according to the formula;

##EQU12## where;

N=the number of sampling points;

n=1<

n<

N

X=the signal value of said electrical signal at a sampling point;

μ

=a mean of the signal values;

σ

=a standard deviation of the signal values; and

k=an integer greater than 1,

(c) storing, in a broadcast signal memory, a library containing a plurality of stored signal identifications, (d) comparing the calculated statistical moments with the plurality of stored signal identifications in said library, and (e) recognizing the received signal as similar to one of said stored signal identifications when said comparison step produces an indication that the calculated statistical moments are similar to the at least one stored signal identification, and outputting a signal indicating that the received broadcast signal is similar to the at least one stored signal identification.

42. Apparatus according to claim 41 wherein said processing means calculates skew and kurtosis of said values, and forms a feature vector from said skew and said kurtosis.

43. Apparatus according to claim 41 wherein said processing means includes:

means for bandpass filtering the received broadcast signal to provide the plurality of filtered signals;

means for rectifying said filtered signals;

means for low pass filtering the rectified signals; and

means for computing a plurality of linear combinations of the low pass filtered signals.

44. Apparatus according to claim 43 wherein said processing means (a1) samples a first one of said linear combinations at the plurality of sampling points to produce a first plurality of signal values, and (a2) samples a second one of said linear combinations at the plurality of sampling points to produce a second plurality of signal values.

45. Apparatus according to claim 44 wherein said processing means (b1) calculates a first skew value and a first kurtosis value from said first plurality of signal values, calculates a second skew value and a second kurtosis value from said second plurality of signal values, and (b2) derives a feature vector comprising said first and second skew values and said first and second kurtosis values.

46. Apparatus according to claim 45 wherein said processing means (d1) quantizes said feature vector in accordance with a plurality of stored predetermined quantization patterns to provide a quantized vector, and(d2) accesses said library with said quantized vector to locate a signal identification corresponding t the quantized vector.

47. Apparatus according to claim 46 wherein said processing means (d2a) forms a weighted sum of the values of said quantized vector using a non-decimal radix, (d2b) uses the weighted sum as an address to access a pointer table stored in said library to locate a pointer corresponding to said weighted sum and uses said pointer to access said library to locate a signal identification list containing the signal identification corresponding to the quantized vector.

48. Apparatus according to claim 47 wherein said processing means (e1) correlates a third one of said plurality of linear combinations with a plurality of stored signals which correspond to the signal identifications contained in said signal identification list and (e2) selects one stored signal whose correlation with said third linear combination exceeds a predetermined threshold.

49. Apparatus according to claim 46 wherein said processing means (d1a) categorizes each skew value of said feature vector in accordance with a stored predetermined skew quantization pattern having categories non-uniformly distributed therein, and (d1b) categorizes each kurtosis value of said feature vector in accordance with a stored predetermined kurtosis quantization pattern having categories non-uniformly distributed therein.

50. Apparatus for creating a signal library useful in signal recognition comprising:

means for receiving a reference signal to be broadcast and converting it into a plurality of electrical signals having different frequency bands;

a broadcast signal memory;

processing means for (a) sampling each of said plurality of electrical signals at a plurality of sampling points to produce a plurality of signal values for each frequency band, (b) calculating a plurality of statistical moments corresponding to the plurality of frequency bands using the plurality of signal values, each statistical moment being calculated according to the formula;

##EQU13## where;

N=the number of sampling points;

n=1<

n<

N

X=the sampled signal value at a sampling point;

μ

=a mean of the sampled signal values;

σ

=a standard deviation of the sampled signal values; and

k=an integer greater than 1,

(c) deriving a feature vector from the calculated statistical moments, and (d) storing the feature vector or a representation thereof in said memory.

51. Apparatus according to claim 50 wherein said processing means calculates two statistical moments for each of the plurality of electrical signals, and derives each said feature vector from both of the calculated statistical moments.

52. Apparatus according to claim 50 wherein said processing means bandpass filters the received signal to provide the plurality of electrical signals, rectifies said filtered signals, low pass filters the rectified signals, and computes a plurality of linear combinations of the low pass filtered signals.

53. Apparatus according to claim 52 wherein said processing means (al) samples a first portion of one of said linear combinations at a plurality of points to produce a first plurality of signal values, and (a2) samples a second portion of said one linear combination at a plurality of points to produce a second plurality of signal values.

54. Apparatus according to claim 53 wherein said processing means (b1) calculates the skew and kurtosis of said first plurality of signal values to provide a first skew value and a first kurtosis value, and (b2) calculates the skew and kurtosis of said second plurality of signal values to provide a second skew value and a second kurtosis value.

55. Apparatus according to claim 54 wherein said processing means (c1) forms a first feature vector including said first skew value and said first kurtosis value, and (c2) forms a second feature vector including said second skew value and said second kurtosis value.

56. Apparatus according to claim 55, wherein said processing means (c3) quantizes both said feature vectors by replacing the skew and kurtosis values with integers in accordance with predetermined non-uniform quantization patterns stored in said memory to produce first and second quantized vectors.

57. Apparatus according to claim 56 wherein said processing means (d1) computes a weighted sum of said first vector using a non-decimal radix, (d2) computes a weighted sum of said second vector using said non-decimal radix, (d3) uses the weighted sums to address said memory, and (d4) stores in said memory, at addresses corresponding to the weighted sums, a signal identification code corresponding to said reference signal.

58. Apparatus according to claim 55 wherein said processing means (c3) produces first and second overlap quantized vectors from said first feature vector in accordance with predetermined non-uniform overlap quantization patterns stored in said memory, (c4) produces third and fourth overlap quantized vectors from said second feature vector in accordance with said predetermined non-uniform overlap quantization patterns stored in said memory, (c5) forms a first plurality of permutations of said first and second overlap quantized vectors, and (c6) forms a second plurality of permutations of said third and fourth overlap quantized vectors.

59. Apparatus according to claim 58 wherein said processing means (d1) computes a weighted sum of the values of each one of said first and second pluralities of permutations, (d2) accesses said memory at addresses corresponding to said weighted sums, and (d3) stores in said memory, at areas corresponding to said addresses, a signal identification code identifying said reference signal.

60. Apparatus for classifying a broadcast signal comprising:

a receiver for receiving the broadcast signal and converting it into a plurality of electrical signals having different frequency bands;

a broadcast signal memory;

a processor for (a) sampling each of the plurality of electrical signals at a plurality of points to produce a plurality of signal values for each frequency band, (b) deriving a multi-value feature vector from said plurality of signal values derived from said plurality of electrical signals, (c) quantizing said feature vector by replacing each feature vector value with a category value determined from a predetermined quantization pattern which non-uniformly categorizes feature vector values, said pattern being stored in said memory, and (d) accessing a signal library with the quantized vector to output a signal identification code corresponding to the received signal, said library being stored in said memory.

61. Apparatus according to claim 60 wherein said predetermined quantization pattern is an overlap quantization pattern.

62. Apparatus according to claim 60, wherein said processor utilizes a plurality of different predetermined quantization patterns to quantize said feature vector.

63. Apparatus according to claim 60 wherein said processor (b1) calculates two different statistical moments of said plurality of signal values, and (b2) forms said feature vector using the two calculated statistical moments.

64. Apparatus according to claim 63 wherein a predetermined quantization pattern is provided for each of the two statistical moments.

65. Apparatus according to claim 63 wherein said two different statistical moments comprise skew and kurtosis.

66. Apparatus according to claim 60 wherein said processor envelope detects said received signal.

67. Apparatus according to claim 60 wherein said processor (c1) replaces said feature vector with a plurality of quantized vectors in accordance with at least one predetermined quantization pattern having overlapping categories, and (c2) permutes said plurality of quantized vectors to produce a further plurality of permutated quantized vectors.

68. Apparatus according to claim 67 wherein said processor (d1) accesses said signal library with said plurality of permutated quantized vectors to reach a plurality of memory locations, and (d2) writes in each said memory location a signal identification code corresponding to the received signal.

69. Apparatus according to claim 60 wherein said receiver (a) spectrally analyzes said received broadcast signal to provide a plurality of electrical signals having different frequency bands, and (b) forms linear combinations of said plurality of electrical signals;

and wherein said processor samples each linear combination to provide a plurality of sampling points for each linear combination, calculates skew and kurtosis values for each linear combination from the corresponding sampling points, and forms said feature vector to include the skew and kurtosis values from all of said linear combinations.

70. Apparatus for preparing a broadcast signal identification library useful in identification of broadcast signals, comprising:

receiver means for receiving a signal to be broadcast and converting it into a plurality of electrical signals having different frequency bands;

a broadcast signal memory;

a processor for (a) sampling each of the plurality of electrical signals to provide a plurality of signal values for each frequency band;

(b) deriving a plurality of multi-value feature vectors from said plurality of signal values, and (c) quantizing each feature vector, including the substeps of;

(c1) establishing a plurality of quantization levels;

(c2) distributing the quantization levels non-uniformly over a predetermined statistical distribution of vector quantization values;

(c3) deriving a plurality of quantization thresholds in accordance with the distributed quantization levels; and

(c4) replacing each value of each multi-value feature vector with a corresponding quantization value determined by said quantization thresholds; and

said processor (d) storing a value representing each quantized vector in said memory as the signal identification for the corresponding received signal.

71. Apparatus according to claim 70 wherein the quantization levels are distributed in an overlapped manner, and wherein said processor (c4a) replaces each value of each feature vector with a plurality of quantization values determined by the overlapped quantization levels, each feature vector thus being replaced with a plurality of quantized vectors, and (c4b) permutes the plurality of quantized vectors of each feature vector to produce a plurality of permuted vectors;

and wherein said processor (d1) stores in said memory, for each sampled signal, values corresponding to the plurality of quantized vectors.

72. Apparatus according to claim 70 wherein said processor (b1) calculates, for each of the plurality of electrical signals, two statistical moments, and (b2) forms each feature vector to include the two calculated statistical moments.

73. Apparatus for identifying broadcast information, comprising:

means for receiving broadcast information and converting it into an electrical signal;

a broadcast signal memory storing a plurality of identifications of broadcast information;

means for spectrally analyzing the electrical signal to produce a plurality of analyzed waveforms having different frequency bands; and

processing means for (a) forming a plurality of linear combinations of said analyzed waveforms, (b) computing at least one statistical moment for each said linear combination, (c) deriving a feature vector from the computed statistical moments, (d) quantizing the feature vector, (e) using the quantized feature vector to access said plurality of identifications of broadcast information stored in said memory, and (f) identifying the received broadcast information based on the stored identifications in said memory, and outputting an identification signal corresponding to the identified received broadcast information.

74. Apparatus according to claim 73 wherein said processor (b1) samples each analyzed waveform at a plurality of sampling points to provide a plurality of sampled values, and (b2) calculates, for each analyzed waveform, two different statistical moments according to the formula:

##EQU14## where;

N=the number of sampling points;

n=1<

n<

N

X=the sampled value at the sampling point;

μ

=a mean of the sampled values;

σ

=a standard deviation of the sampled values; and

k=an integer greater than 1, a different value of k being provided for each of the two different statistical moments.

75. Apparatus according to claim 74 wherein said processor (c1) forms each feature vector to include values corresponding to each of the two calculated statistical moments.

76. Apparatus according to claim 73 wherein said processor (d1) replaces each value of said feature vector with a quantization value derived from a predetermined non-uniform quantization pattern stored in said memory.

77. Apparatus according to claim 76 wherein said processor (e1) extracts from said memory a list of identifications of broadcast information, and wherein said processor (f1) correlates one of said analyzed waveforms with a plurality of information codes stored in a correlation library in said memory, the plurality of information codes corresponding to the extracted list of identifications, and (f2) declares the broadcast information to be similar to one of the plurality of information codes when the correlation therebetween exceeds a predetermined threshold.

78. Apparatus for classifying broadcast information, comprising:

means for receiving broadcast information and converting it into an electrical signal; and

processing means for (a) performing a spectral analysis of said electrical signal and providing a plurality of electrical signals having different frequency bands, (b) computing linear combinations of said plurality of electrical signals analyzed waveforms, (c) calculating a statistical moment for each said linear combination, and (d) classifying the received broadcast information based on the calculated statistical moments and a plurality of predetermined broadcast signal identifications stored in a broadcast signal memory, and outputting an identification signal corresponding to the classified received broadcast information.

79. Apparatus according to claim 78 wherein said processor (c1) calculates two higher-order statistical moments of each said linear combination, (c2) forms a feature vector incorporating the calculated statistical moments, and (d1) classifies the received broadcast information base don the formed feature vector.

80. Apparatus for classifying a broadcast signal comprising:

a receiver for receiving the broadcast signal and converting it to an electrical signal;

a broadcast signal memory;

a computer for (a) bandpass filtering the converted electrical signal to provide a plurality of filtered signals having different frequency passbands, (b) envelope detecting and low pass filtering the filtered signals to provide a plurality of low bandwidth signals, (c) computing a plurality of linear combinations of said low bandwidth signals, (d) sampling several of the linear combinations to provide several sets of sampled values, (e) calculating a skew value and a kurtosis value for each set of sampled values, (f) forming a feature vector from the skew and kurtosis values of the several sets of sampled values, (g) quantizing the feature vector by replacing each skew and kurtosis value with a category value derived from a plurality of stored predetermined non-uniform quantization patterns stored in said memory, (h) using the quantized vector to access in said memory a first data base storing a plurality of signal identification lists, each list capable of storing a plurality of signal identification codes, (i) extracting from the first data base the signal identification list corresponding to the quantized vector, (j) extracting from a second data base stored in said memory correlation signals corresponding to the extracted signal identification list, (k) correlating the extracted correlation signals with one of the computed linear combinations, and (l) classifying the broadcast information as similar to one of said correlation signals when its correlation with said one linear combination exceeds a predetermined threshold, and outputting a signal corresponding to the classification.