Digital tone decoder and method of decoding tones using linear prediction coding

• US 4,689,760 A
• Filed: 11/09/1984
• Issued: 08/25/1987
• Est. Priority Date: 11/09/1984
• Status: Expired due to Term

• Current Assignee: Digital Sound Corp.
• Original Assignee: Digital Sound Corp.
• Inventors: Lee, Chin-hui, Wong, David Y.
• Primary Examiner(s): Krass, Errol A.
• Assistant Examiner(s): Teska, Kevin J.
• Application Number:
  US06/670,585
• Time in Patent Office: 1,019 Days
• Field of Search: 364/576, 364/724, 364/726, 381/51, 381/34, 375/94, 375/96, 179/18 EB, 179/16 EC, 179/84 VF
• US Class Current: 370/526
• CPC Class Codes: H04Q 1/4575 which are transmitted in digital form

Abstract

An all digital feature based pattern recognition tone detection system distinguishes one or more tones from speech or other background noise and provides extremely reliable, low cost DTMF tone detection. The system includes a digital signal processing system which uses a modified PARCOR signal analysis technique to efficiently calculate inverse filter coefficients defining in functional form a smoothed LPC spectrum from a received frame of digital pulse code data. A Fourier transform provides high resolution evaluation of the LPC data at frequencies of interest to permit evaluation of the spectral response in accordance with known features of an expected tone signal.

Claims

1. The method of digitally detecting a tone at at least one predetermined tone frequency in a received signal represented by a frame of sequential digital data samples, the method comprising the steps of:

calculating an array of Z transform domain coefficients from the digial data samples approximately representing the frame of sequential digital data samples;

evaluating a frequency spectrum represented by the array of Z transform domain coefficients at at least one evaluation frequency in a vicinity of each at least one predetermined tone frequency to obtain a spectral value indication at each evaluation frequency;

testing each said spectral value indication for a presence or absence of a tone in the vicinity of each of the at least one predetermined tone frequency to determine for each predetermined tone frequency whether or not the received signal contains a tone at the predetermined tone frequency; and

providing an indication of whether any of the tones at at least one predetermined tone frequency is in the received signal represented by the frame, and if any of the tones are in the signal, the predetermined frequencies of the tones that are present in the received signal.

2. The method of detecting according to claim 1 above, wherein the at least one predetermined frequency includes at least 8 DTMF frequencies and the evaluating step includes evaluating the frequency spectrum of the frame at at least 6 evaluation frequencies in the vicinity of each DTMF frequency.

3. The method of detecting according to claim 2 above, wherein the frame represents a 22.5 millisecond time interval of the received signal and contains 180 digital data values.

4. The method of detecting according to claim 2 above, wherein the evaluating step includes calculating an array of autocorrelation coefficients, calculating an array of partial autocorrelation coefficients from the correlation coefficients using a recursive calculation technique and calculating the array of Z transform domain coefficient from the partial correlation coefficients using a recursive calculation technique.

5. The method of detecting according to claim 1 above, wherein the at least one predetermined frequency consists of a combination of exactly one frequency from a low group of frequencies and exactly one frequency from a high group of frequencies and wherein the providing step includes the step of providing an indication of whether a combination of exactly one tone at a low group frequency and exactly one tone at a high group frequency is present in the received signal.

6. The method of detecting according to claim 5 above, wherein the testing step includes the step of determining which spectral value indication at an evaluation frequency in the vicinity of the low group tone frequencies indicates a greatest low group spectral value, determining which spectral value indication at an evaluation frequency in the vicinity of the high group tone frequencies indicates a greatest high group spectral value, and determining that the received signal does not contain a valid combination of tones if the ratio of the indicated greatest low group spectral value and indicated greatest high group spectral value are not within a predetermined range.

7. The method of detecting according to claim 1 above, wherein the testing step includes the step of determining that the ratio is not within the predetermined range if the larger of the two indicated low group and high group spectral values divided by four is greater than the smaller.

8. The method of detecting according to claim 5 above, wherein the testing step includes the step of determining that the received signal does not contain a valid tone combination if either the spectral value indication indicating a greatest spectral value at an evaluation frequency in the vicinity of the low group of tone frequencies indicates a spectral value less than a predetermined minimum threshold value or the spectral vaue indication indicating a greatest spectral value at an evaluation frequency in the vicinity of the high group of tone frequencies indicates a spectral value less than the predetermined threshold value.

9. The method of detecting according to claim 8 above, wherein the evaluating step includes evaluating the frequency spectrum at at least six frequencies in the vicinity of each predetermined tone frequency and wherein the testing step includes the step of increasing the threshold value to a second threshold value if the indicated strongest frequency in both the low group and high group of tone frequencies occurs at an evaluation frequency which is a lowest evaluation frequency in the vicinity of a predetermined tone frequency.

10. The method of detecting according to claim 5 above, wherein the testing step includes the steps of determining which spectral value indication at a frequency in the vicinity of a low group tone frequency indicates a highest spectral value, determining which spectral value indication at a frequency in the vicinity of a high group tone frequency indicates a highest spectral value, and determining that exactly one tone is present at a low group frequency and exactly one tone is present at a high group frequency when the determined low group highest spectral value indication and the determined high group highest spectral value indication both indicate a spectral value in excess of a selected threshold.

11. The method of detecting according to claim 5 above, wherein the evaluating step includes evaluating a frequency spectrum represented by the array of Z transform domain coefficients to obtain a spectral value indication at a frequency midway between the low group and high group tone frequencies and wherein the testing step includes determining that the received signal does not contain a valid tone combination if the midway frequency spectral value indication indicates a spectral value greater than either a greatest spectral value indicated by a spectral value indication at an evaluation frequency in the vicinity of a low group tone frequency or a greatest spectral value indicated by a spectral value indication at an evaluation frequency in the vicinity of a high group tone frequency.

12. The method of detecting according to claim 5 above, wherein the testing step includes the step of determining that the frame does not represent a valid tone combination if the spectral value indication for more than half of the evaluation frequencies in the vicinity of the low group tone frequencies indicate a spectral value less than a selected low group threshold value or if the spectral value indication for more than half of the evaluation frequencies in the vicinity of the high group tone frequencies indicate a spectral value less than a selected high group threshold value.

13. The method of detecting according to claim 5 above, wherein the testing step includes the step of determining separately for both the high group of tone frequencies and the low group of tone frequencies the spectral value indication which indicates the largest spectral value and determining that the received signal does not represent a valid tone combination if a spectral value indication for a second evaluation frequency in the vicinity of a tone frequency in the group indicates a spectral value that is a maximum and has an indicated spectral value within a threshold ratio of the indicated largest spectral value.

14. The method of detecting according to claim 5 above, wherein the evaluating step includes evaluating the frequency spectrum at at least six frequencies in the vicinity of each predetermined tone frequency and wherein the testing step includes the step of determinimg that the received signal does not represent a valid tone combination if the bandwidth of the indicated strongest tone frequency in either the low group or the high group has a bandwidth indicated by the spectral value indications greater than a predetermined maximum bandwidth.

15. The method of detecting according to claim 1 above, wherein the calculating step calculates inverse filter coefficients using a linear prediction coding technique.

16. The method of detecting according to claim 1 above, wherein the digital data samples are encoded with linear pulse code modulation (PCM) coding.

17. The method of detecting according to claim 1 above, wherein the array of Z transform domain coefficients is of order 6.

18. The method of detecting according to claim 1 above, wherein the evaluating step evaluates the frequency spectrum of the frame at at least three different evaluation frequencies in the vicinity of each predetermined frequency.

19. The method of detecting according to claim 1 above, wherein the at least one predetermined frequency includes a plurality of frequencies in each of a low frequency group and a high frequency group and the evaluating step includes evaluating the frequency spectrum of the frame at a frequency midway between the low and high frequency groups.

20. The method of detecting according to claim 1 above, wherein the step of calculating an array of partial correlation coefficients uses the LeRoux and Gueguen recursion algorithm modified by terminating the recursion and setting all remaining partial correlation coefficients to zero when the value of the first term of the normalized residual energy array falls below a predetermined threshold.

21. The method of detecting according to claim 1 above, further comprising the step of checking for a full frame of similar information by counting the zero crossings in a first half frame and in a last half frame and eliminating the frame if the two counts differ by more than a threshold amount.

22. The method of detecting according to claim 1 above, wherein the testing step includes determining for each predetermined frequency whether the spectral value of each evaluation frequency proximate thereto satisfies selected criteria and determining that the received signal does not have a tone at the predetermined frequency if the criteria is not met, the criteria including the occurrence of a single maximum spectral value greater than a predetermined threshold value in the vicinity of each predetermined frequency.

23. The method of detecting according to claim 1 above wherein the testing step includes the step of determining that a tone is present at a predetermined tone frequency when a spectral value indication at an evaluation frequency in the vicinity of the predetermined tone frequency indicates a spectral value in excess of a selected threshold value.

24. The method of detecting according to claim 1 above, wherein the testing step includes the step of determining for each tone frequency that a tone is not present at the frequency if no spectral value indication for a frequency in the vicinity of the tone frequency indicates a spectral value in excess of a predetermined threshold value.

25. The method of detecting according to claim 1, above wherein the testing step includes the step of determining which evaluation frequency has a spectral value indication indicating the largest spectral value and determining that the received signal is not a tone signal if a spectral value indication at a second evaluation frequency indicates a spectral value that is a maximum and has a spectral value within a predetermined threshold range of the spectral value indicated by the largest spectral value indication.

26. The method of detecting according to claim 1 above, wherein the evaluating step includes evaluating the frequency spectrum at at least six evaluation frequencies in the vicinity of each predetermined tone frequency and wherein the testing step includes determining that the received signal does not represent a tone if the bandwidth of the spectral values indicated by the spectral value indications for evaluation frequencies surrounding an evaluation frequency having a spectral value indication indicating a largest spectral value is greater that a selected threshold.

27. The method of detecting according to claim 1 above, wherein the array of Z transform domain coefficients is an array of inverse filter coefficients and wherein the evaluating step includes the step of evaluating a frequency spectrum represented by the array of inverse filter coefficients at at least one evaluation frequency in a vicinity of each at least one predetermined tone frequency to obtain an inverse spectral value at each evaluation frequency.

28. The method of detecting a tone at each of a plurality of predetermined low group and high group frequencies in a received signal represented by a frame of sequential data samples, the method comprising the steps of:

calculating from the digital data samples an array of Z transform domain coefficients approximately representing the frame of sequential digital data samples;

evaluating a frequency spectrum represented by the array of Z transform domain coefficients at at least one evaluation frequency in a vicinity of each predetermined frequency to obtain a spectral value indication at each evaluation frequency;

testing each said spectral value indication in the vicinity of each predetermined frequency to determine whether or not the received signal contains exactly one tone at a predetermined low group frequency and exactly one tone at a predetermined high group frequency and

providing an indication of whether the received signal contains a combination of one tone at a low group frequency and one tone at a high group frequency and, if it does, which combination.

29. The method of detecting telephone dialing tones at predetermined DTMF frequencies in a received signal comprising the steps of:

reresenting a frame of the received signal with sampled digital data values;

calculating linear prediction coding coefficients defining an inverse filter from the digital data values;

evaluating the Fourier transform of the inverse filter to obtain spectral values at selected frequencies including at at least one frequency proximate each predetermined DTMF frequency;

processing the spectral values at the at least one frequency proximate each predetermined DTMF frequency to determine for each predetermined DTMF frequency whether or not a dialing tone exists; and

determining whether exactly one tone has been detected in each of two different frequency ranges and, if so, indicating a character corresponding to the detected tones.

30. The method of detecting a sinusoidal tone signal of a given tone frequency in a received data signal comprising the steps of:

digitally sampling the data signal at a predetermined sample rate to produce a sequence of sampled data points;

grouping the sequence of sampled data points into a sequence of data frames, each representing a predetermined number of sampled data points; and

for each frame in succession evaluating the frame to determine whether or not the sinusoidal tone is present, said evaluating including the steps of;

performing time domain analysis of the data frame including the generation of an inverse filter transfer function of the form

space="preserve" listing-type="equation">A(z)=1+SUM {a.sub.i z.sup.-i };

(i=1,2,,P)

where z is the Z transform operator;

performing frequency domain analysis including a Fourier transform of the inverse filter transfer function to obtain spectral data at at least one frequency in the vicinity of the tone frequency; and

evaluating the spectral data to determine the presence or absence of the sinusoidal tone.

31. The method of detecting a sinusoidal tone signal according to claim 30 above, wherein the inverse filter transfer function is generated using linear prediction coding.

32. The method of detecting a sinusoidal tone signal according to claim 30 above, wherein the inverse filter transfer function is generated using a linear prediction coding autocorrelation technique.

33. The method of detecting a sinusoidal tone signal according to claim 30 above, wherein the inverse filter transfer function is generated using a linear prediction coding partial correlation technique.

34. The method of detecting a sinusoidal tone signal according to claim 30 above, wherein the inverse filter transfer function is generated using a linear prediction coding partial correlation technique in which reflection coefficients and normalized residual energy terms are generated recursively, the recursion being terminated when the value of a generated residual energy term falls below a selected threshold which is greater than zero.

35. The method of detecting at least one sinusoidal tone signal, each of a given frequency, in a frame or sampled data representing a data signal comprising the steps of:

generating in response to the frame of sampled data a time domain finite polynomial sequence containing information that is indicative of a frequency spectrum of the frame of sampled date;

performing a Fourier transform of the finite polynomial sequence to obtain spectral frequency data at at least one frequency in the vicinity of the given frequency of each sinusoidal tone signal;

evaluating the spectral frequency data to determine the presence or absence of each of the sinusoidal tone signals.

36. A digital computer program for evaluating a frame of sampled data representing a received signal to determine the presence or absence of a tone signal in the received signal occurring at a frequency which is one of at least one predetermined tone frequency, the program comprising the steps of:

calculating from the sampled data an array of Z transform domain coefficients approximately representing the frame of sampled data;

evaluating a frequency spectrum represented by the array of Z transform domain coefficients at at least one evaluation frequency in a vicinity of each at least one predetermined tone frequency to obtain a spectral value indication at each evaluation frequency;

testing each said spectral value indication in the vicinity of each said at least one predetermined tone frequency to determine for each predetermined tone frequency whether or not the received signal contains a tone at the predetermined tone frequency; and

providing an indication of whether the received signal carries a tone at one of the at least one predetermined frequency, and if the received signal does carry at least one tone, the predetermined frequency associated with each tone.

37. A decoder system comprising:

a data bus coupled to communicate digital data including sampled data representing a frame of a received signal; and

a digital signal processing system including a digital data processor, the digital signal processing system including

means for calculating from the sampled data an array of Z transform domain coefficients approximately representing the frame of sampled data;

means for evaluating a frequency spectrum represented by the array of Z transform domain coefficients at at least one evaluation frequency in a vicinity of each of at least one predetermined tone frequency to obtain a spectral value indication at each evaluation frequency;

means for testing each said spectral value indication in the vicinity of each predetermined tone frequency to determine whether or not the received signal carries a tone at the predetermined tone frequency; and

means for communicating over the data bus an indication of whether or not the received signal has a tone at a predetermined tone frequency and if the received signal has a tone at a predetermined frequency, the frequency of each tone.

38. The decoder system according to claim 37 above, wherein the means for calculating includes means for using linear prediction coding to calculate the Z transform domain coefficients.

39. The decoder system according to claim 38 above, wherein the means for using includes means using partial correlation to recursively calculate the Z transform domain coefficients.

40. The decoder system according to claim 37 above, wherein the means for calculating is a 16 bit fixed point digital data processor.