GB2430832A

GB2430832A - Dual-tone multiple frequency (dtmf) symbol detection

Info

Publication number: GB2430832A
Application number: GB0519961A
Authority: GB
Inventors: Andrew Gough
Original assignee: Trinity Convergence Inc
Current assignee: Trinity Convergence Inc
Priority date: 2005-09-30
Filing date: 2005-09-30
Publication date: 2007-04-04
Anticipated expiration: 2025-09-30
Also published as: GB0519961D0; TW200729975A; GB2430832B

Abstract

A dual-tone detection system comprises a receiver and a digital-signal processor. The digital-signal processor comprises an arithmetic-logic unit as well as window, automatic-gain control, first filter, second filter and determination modules. The window module generates a representation of the input signal. The automatic-gain control module generates a scaled representation so the tones are at a desired power level. The first filter centered about a first frequency generates a first measurement. The second filter centered about a second frequency generates a second measurement. The arithmetic-logic unit uses an equation that models a power relationship between the first and second filtered tones to generate adjusted first and second measurements, respectively. The determination module compares each of the adjusted first and second measurements with a threshold value to identify when the input signal contains the low-frequency and high-frequency tones associated with a particular DTMF symbol.

Description

SYSTEMS AND METhODS FOR DUAL-TONE MULTIPLE FREQUENCY

SYMBOL DETECTION

Inventor Andrew Gough

BACKGROUND OF THE INVENTION

1. Field of the Invention

1] This invention relates generally to dual-tone multiple frequency (DTMF) symbol detection. More particularly, the invention relates to systems and methods that accurately identify a particular DTMF entry or symbol by detecting both the low- frequency tone and the high-frequency tone associated with the symbol.

2. Related Art [0002] DTMF' signaling is used in telephone dialing and other applications, such as electronic banking systems. A I)TMF signal corresponds to one of sixteen touchtone symbols (0-9, A-D, #, ) as shown in FIG. 1. Each symbol is represented by one of four frequencies in a low frequency band and one of four frequencies in a higher frequency band. In FIG. 1, the symbols are shown in a matrix format. Each symbol is represented by a frequency representing the column in which the symbol appears and by a frequency representing the row in which the symbol appears. The columns are represented by frequencies in a band between lkJIz (kilo-Hertz) and 2kHz, and the rows are represented by frequencies in a band between 500Hz and I kl-Jz. The first three columns of symbols form the telephone keypad layout familiar to consumers of voice telephone services. The last column includes symbols that are available for more particularized applications.

[00031 Whenever a key of a touch-tone keypad is depressed, the high frequency and the low frequency corresponding to the symbol assigned to that key are generated and transmitted to a receiving device. The device that receives this dual-tone signal must detect which one of the four low frequencies and which one of the four high frequencies have been received to determine which symbol has been transmitted by the communicating device.

4] The problem of DTMF signal detection is non-trivial for several reasons. T'he eight frequencies used to encode the symbols are within the spectrum of frequencies generated by voice-data. Therefore, when voice data is transmitted, symbol simulation, (also called digit simulation), may occur. A DTMF detector must be able to discriminate against these voice-simulated symbols. Also, the DTMF signal is attenuated by the transmission medium through which it is transmitted. Typical transmission media attenuate high frequencies more than low frequencies Thus, the higher frequency in the dual-tone pair may have significantly less power at the receiver than the low frequency in the pair. In addition, electronic devices do not typically generate all DTMF frequencies at the same power level. It is therefore possible for the lower frequency to be received at a lower power level than the high frequency. This disparity in power between the low and high frequency is called "twist." Further, both tones must be detected in the presence of noise power, which may be a significant fraction of the signal power of the received DTMF signal. A further problem is that not all devices will generate the exact dual-tone frequencies shown in FIG. I because of poor design or system degradation. Thus, the DT'MF receiver must be able to detect the DTMF signals at frequencies slightly offset from the nominal values while rejecting frequencies outside a given tolerance band. Because the nominal DTMF frequencies are closely spaced, the tolerance band must be very narrow. The problem of signal detection within narrow frequency bands is complicated by the fact that each signal is transmitted only for a short time of uncertain length with an nncertain delay time between transmissions of successive symbols.

[00051 To standardize the perfbrmancc of devices hr DTMF' signal generation and reception, the International Telecommunications Union (ITU) has developed a set of performance standards to which these devices should comply. These standards have achieved virtually worldwide acceptance, and refine the standards previously developed by Bell Communications Research, inc. ("Belicore"). The ITU standards are summarized in Table 1.

TABLE 1

Test Test Type Beilcore EIAiTIA-464A ITIJ-T Q.24, Results ID __________ TR-TSY-000 181 j______ _______ AT&T Values _____________ D- I Frequency 1 5% must accept, 1.5% must accept, 1.5% must PASS Deviation 3 5% must reject -13 5% must reject accept; All digits detected at 13.5% must reject 1.5% All Digits rejected at ______ 3.5% _______ D-2 Miiiiirnimn 40 ins must accept. 40 ins must accept 40 ins must accept PASS Tone 23 ms must relect 23 rns must reject All digits detected Duration with 40 ins pulse.

All digits rejected ______- with23mspulsc 1)-3 Minimum 40 ms 40 ms = 40 ins detect PASS 40 ms Interdigilal two digits Double digits Interval < 10 ins bridge detected for all digits gap (one digit) PASS 10 ms: Single digits detected for all --.- D-4 Minimuni 93 ms 93 ms 93 ms/digit PASS Cycle All digits detected.

Time __________ D-5 Accept Levels 0 to -36 dBm must 0 to -25 dBni musl 0 to -25 clBm must PASS (Bellcore, accept; -55 dl3m accept accept EIAJTIA464A, and must reject -55 dBm reject ITU-TQ.24): Detect all digits in range of-lO to -48 dBmO (38 dB range) PASS (Bellcore and ITU-T Q.24): Reject all digits = -66 dBmO (-55 dBm).

D-6 Fwist (ratio of -8 to 4 dli -8 to +4 dli i4 to -8 dli must PASS high group accept Detects all tones br power to low) twist of +4 and -8 dli 1)-7 Bellcore lalk- Fewer than 670 - - PASS off total talkoffs; fewer (, total detections for Tape TR- than 330 talkoffs of Bellcore 3 hour test TSY-00763 digits 0-9; fewer than 170 talkoffs of ____ signals*and# -- -.. -.--- ______ I)-8 Mitel talk-off - - - PASS.

tape CM-7291 side B with I ________________ ___________ detection.

D-9 SNR 23 dli 15 dli - PASS.

All tones detected for and 23 dli SNR D-JO Impulse Noise Fewer than 14 Fewer than 10 - PASS missed or split errors in 10.000 No detections in digits in Belicore tones for EIA test Beilcore Impulse Impulse Noise Tape # 1, fewer than 500 Noise Tape No 201 errors in 10,000 tones for _______ ________ test #2 D-II Echo 16dB 10dB 10dB PASS Signal-to-Echo ratio Signal-to-Echo Signal-to-Echo Detected all digits ratio at 2Oms ratio up to 2Orns with 10 dB Signal to Echo ratio -.

6] Voice-simulated tones must be rejected as invalid tones. Signal frequencies that are within +1-i.5% of the nominal frequencies listed should be detected as valid DTMF tones. A signal frequency outside the band of +1-3.5% of a nominal frequency must be rejected as an invalid tone. Two twist parameters are also specified. The twist, which is the ratio of the low frequency power to the high frequency power in decibels (dB), is specified to be greater than -4dB and less than 8dB. A positive twist value is a forward twist condition, which is the case when the low frequency signal power exceeds the high frequency power. A negative twist value is a reverse twist condition which exists when the high frequency signal power exceeds the low frequency signal power. When the twist is within the range of -4dB to +8db, the signal must be accepted as valid. Also, according to Beilcore standards, a valid DTMF signal must be detected if the signal-to- noise ratio (SNR) is at least 15dB. In addition to frequency and power tolerances, temporal constraints are also imposed. A DTMF signal of duration at least 4Omsec (milliseconds) must be detected. A signal of duration 23msec or less must be rejected.

Also, if the time beiween the end of one [)TMF signal and the beginning of (he next Successive DTMF signal, the interdigit time, is at least than 40msec, the signals must he distinguished as two distinct symbols. Conversely, a signal interruption of I Omsec or less must not cause detection of two separate tones.

7] Within the telephone network, 1)TMF signals are typically transmitted digitally at a sampling rate of about 8khz (8000 samples per second), to give sample durations of approximately 0.Ol25msec. One way to detect the presence of a valid DTMF signal is by performing a digital-toanalog conversion fbllowed by applying bank of analog filters centered at the nominal DTMF frequencies. This method is not efficient because of the required conversion process and the size and complexity of analog filters. It is more desirable to achieve DTMF signal detection using digital methods which can be implemented by an integrated circuit digital signal processor.

[00081 The most common digital methods for DTMF detection involve repetitively or iteratively computing the frequency content of the received signal over a finite duration of time referred to as a frame. These digital DTMF detectors are based on the Goertzel filter as described below.

9] The Goertzel filter is a discrete Fourier transtbrm (DFT) which is centered on a particular target or center frequency. The output of the Goertzel filter is the magnitude of the energy, which a signal, applied to the filter, contains within a range of frequencies about the center frequency of the filter. The Goertzel filter is modified by coefficients and length. The coefficients determine which frequency the Goertzel filter is centered on, while the length determines the range of frequencies that the Goertzel filter encompasses. The longer the filter length, the greater the frequency resolution of the Goertzel filter.

0] Consider a signal of pure white noise. That is. a signal with every possible frequency, each signal component at exactly the same energy level. FIG. 2 illustrates the magnitude response plot 20 about a target or center frequency, labeled C1, for two different filter lengths. Frequency is plotted along the x-axis. Magnitude is plotted along the y- axis. The solid line represents the filter response when the filter length is samples. The dashed line represents the filter response when the filter length is 100 samples. It is clear from the magnitude response plot 20 that the frequency resolution of the magnitude response varies directly with the length of the filter. In FIG. 2, a Goertzel filter of length 200 has a magnitude response, which is 3dB below its peak at +1- 20Hz from its target or center frequency. A Goertzel filter with a length of 100 has a magnitude response which is 3dB below its peak at +1-40Hz from its target or center frequency. Note also that the area under the two curves in FIG. 2 is the same. The shorter the Goertzel filter length, the shorter in magnitude and wider in frequency the resulting response curve.

1] Conventional digital DTMF detectors calculate a magnitude using a Goertzel filter centered at each of the eight DTMF frequencies and at each of the eight 2' harmonics of the eight DTMF frequencies. Pure DTMF tones will produce a high magnitude at the fundamental frequencies and a low magnitude at the 2' harmonics.

Speech, which is similar to a DTMF tone is distinguishable because the filter at the 2nd harmonic contains a significant amount of energy. A state machine can then be used to determine whether a valid DTMF tone has been detected based on the sixteen Goertzel filter response magnitudes.

2] To determine if a valid tone has been detected, the DTMF frequency in the high band at which the power is greatest is determined. Similarly, the DTMF frequency in the low band at which the power is greatest is also determined. Each of these signals must exceed a certain threshold power or a decision is made that no valid DTMF signal has been detected within the current frame. For static threshold testing, the threshold is a fixed, predetermined amount. For dynamic threshold testing, the threshold is the minimum amount by which the power in the strongest tone in the band must exceed the power of the signals at the other three DTMF frequencies in the band. Further, the power of the strongest tone in the high hand is compared to the power of the strongest tone in the low band to determine if the twist is within the range of -4dB to +8dB.

3] The processing power required to process each of the (Joertzel filters centered about the eight fundamental DTMF tones and the eight 2 harmonics of the eight DTMF tones is significant. Therefore, it would be desirable to provide a low cost, reliable and integrated solution that can he implemented using less processing power than conventional DTMF detectors.

SUMMARY

4] Embodiments of a dual-tone detection system comprise a receiver and digital signal processor. The receiver receives an input signal which may include a first tone at a first frequency and a second tone at a second frequency wherein the second frequency is different from the first frequency. The received input signal may also include speech and other audible infbrmation. The digital signal processor comprises an arithmetic logic unit as well as window, automatic gain control, first filter, second filter and determination modules. The window module generates a representation of the input signal. The automatic gain control module generates a scaled representation of the input signal such that the first and second tones are at a desired power level. The first filter module is centered about a first ideal tone, receives the scaled representation and generates a first measurement. The second filter module is centered about a second ideal tone, receives the scaled representation and generates a second measurement. The arithmetic logic unit uses an equation that models the power relationship between the first and second tones to generate adjusted first and second measurements, respectively.

The determination module compares each of the adjusted first and second measurements with a threshold value to identify when the input signal contains the low-frequency and high-frequency tones associated with a particular DTMF symbol. The dual-tone detection system effectively identifies the presence of a DTMF symbol when the corresponding tones are present in the input signal. I'hc dual-tone detection system also effectively discriminates between speech and other audible information components that overlap in frequency with the standard DFMF tones. That is, the dual-tone detection system does not identi' the presence of a I) TMF symbol when a D'I'MF symbol has not been entered.

100151 One embodiment of a method for detecting tones in a signal comprises receiving a signal that may include a first tone at a first frequency and a second tone at a second frequency wherein the second frequency is different from the first frequency, sampling the signal to generate a digital representation of the signal, applying an automatic gain control to the digital representation of the signal to generate a gain-controlled signal, applying the gain-controlled signal to a first filter and a second filter to generate a first filter response and a second filter response, respectively, the first filter having a first center frequency and the second filter having a second center frequency wherein the first center frequency is different from the second center frequency, determining an unmodified first filter response using the gaincontrolled signal, determining an unmodified second filter response using the gain-controlled signal, and using the unmodified first filter response and the unmodified second filter response to verif' that the first tone approximates the first center frequency and the second tone approximates the second center frequency.

6] An alternative embodiment of a method for detecting the presence of DTMF tones comprises receiving an input signal, reducing the impact of different signal power levels of first and second tones have on respective Goertzel filter response magnitudes when a first Goertzel filter is centered at the ideal low frequency and a second Goertzel filter is centered at the ideal high frequency of a particular DTMF tone pair and determining whether the input signal contains a particular DTMF tone pair based on the frequency of the first tone and the frequency of the second tone in relation to the ideal DTMF frequencies.

7] The figures and detailed description that follow are not exhaustive. The disclosed embodiments are illustrated and described to enable one of ordinary skill to make and use the systems and methods for accurate 1)TMF symbol detection. Other embodiments, features and advantages of the systems and methods will be or will become apparent to those skilled in the art upon examination of the following figures and detailed description. All such additional embodiments, features and advantages are within the scope of the systems and methods for accurate DTMF symbol detection as defined in the accompanying claims.

BRIEF DEscRIpTIoN OF THE FIGURES

8] The systems and methods for DTMF symbol detection can be better understood with reference to the following figures. The components within the figures are not necessarily to scale, emphasis instead being placed upon clearly illustrating the principles behind the systems and methods. Moreover, in the figures, like reference numerals designate corresponding parts throughout the different views.

9] FIG. 1 is a schematic diagram illustrating the association of the 16 DTMF symbols with four low frequencies and four high frequencies.

0] FIG. 2 is a plot illustrating Goertzel filter magnitude response for filters having two different filter lengths.

1] FIG. 3 is a plot illustrating Goertzel filter magnitude response of a low- frequency tone centered at 770 Hz when both the low-frequency and high- frequency tones are varied in frequency.

2] FIG. 4 is a plot illustrating Goertzel filter magnitude response of a high- frequency tone centered at 1209 I-Iz when both the low-frequency and high- frequency tones are varied in frequency.

[00231 FIG. 5 is a plot illustrating Goertzel filter magnitude response of the low- frequency lone centered at 770Hz when the signal power of the high- frequency tone centered at 1209Hz is varied when an automatic gain control (A(iC) is applied to the entire signal.

4] FIG. 6 is a plot illustrating Goertzel filter magnitude response of the high- frequency tone ccnteied at 1209Hz when the signal power of the high- frequency tone centered at 1209Hz is varied.

5] FIG. 7 is a plot illustrating the relationship between the Goertzel filter magnitude response of the low-frequency tone centered at 770Hz when the signal power of the high-frequency tone centered at 1209Hz is varied.

6] FIG. 8 is a plot illustrating the relationship between the Goertzel filter magnitude response of the high-frequency tone centered at 1209Hz when the signal power of the high-frequency tone centered at 1209Hz is varied.

[0027j FIG. 9 is a plot illustrating the relationship between the (Joertzel filter magnitude response of the low-frequency tone centered at 770Hz and the Goertzel filter magnitude of the high-frequency tone centered at 1209Hz.

[0028J FIG. 10 is a plot illustrating a normalized relationship between the Goertzel filter magnitude response of the low-frequency tone centered at 770Hz and the Goertzel filter magnitude of the high-frequency tone centered at 1209Hz.

9] FiG. 11 is a functional block diagram illustrating an embodiment of a DTMF symbol detection system.

0] FiG. 12 is a flow diagram illustrating an embodiment of a method for detecting a DTMF symbol.

1] FIG. 13 is a flow diagram illustrating an alternative embodiment of a method for detecting a DTMF symbol.

DETAILED DESCRIPTION

2] The described systems and methods produce a robust and reliable DTMF symbol detector using only eight (ioertzel filters centered on the fundamental DTMF frequencies. The disclosed approach has the advantage of reducing the processing power required to detect a D'FMF digit by 50%.

3] The problem with DTMF detection using only eight Goertzel filters is that it is impossible to know if a low (Joertzel filter response value is because the tone in question is low in power or is shifted in frequency away from the target frequency of the respective filter. For example, if there is a tone at half power at the DTMF fundamental frequency of 770Hz this could give the same result as a tone at full power at a frequency of 750Hz. By negating the impact that tones of differing signal power levels have on the Goertzel filter response magnitudes, a detection decision can be made based only on the frequency of the low frequency and high frequency tones in relation to the ideal DTMF frequencies.

Low-Frequency Detection [0034] Consider the case for 1)TMF symbol "4," where the low-frequency tone is 7701-lz and the high-frequency tone is 1209Hz. FIG. 3 is a plot 300 illustrating Goertzel filter magnitude of a low-frequency tone centered at 7701 Iz when both the low- frequency and high-frequency tones are varied in frequency. The plot 300 shows frequency of the low-frequency tone along the x-axis, frequency of the high-frequency tone along the y-axis and magnitude in the z-axis. Frequency of the high-frequency tone varies from about 1150Hz to 1265Hz. Frequency of the low-frequency tone varies from about 730Hz to about 810Hz. Each tone has the same power and an automatic gain control (AGC) was applied to bring the total energy of the signal to a known level, in this case 50% of maximum.

5] Plot 300 shows that as the low-frequency tone's frequency moves away from the 770Hz center or target frequency of the Goertzel filter, the (ioertzel filter response

II

decreases from a peak value of about 0.155 in the familiar normal distribution pattern.

Plot 300 also shows that as the high-frequency's tone varies in frequency the high- frequency tone has little effect on the magnitude of the 7701Iz Goertzel filter result.

high-Frequency Detection [0036] FIG. 4 is a plot 400 illustrating Goertzel filter magnitude of a high-frequency tone centered at 12091-Iz when both the low-frequency and high-frequency tones are varied in frequency. Plot 400 shows frequency of the low-frequency tone along the x- axis, frequency of the high-frequency tone along the y-axis and magnitude in the z-axis.

Frequency of the high-frequency tone varies from about 1150 lIz to about l2651-Iz.

Frequency of the low-frequency tone varies from about 730Hz to 810Hz. Note that the magnitude of the Goertzel filter response changes when the frequency of the high tone changes, there is very little effect on the magnitude as the low-frequency tone changes in frequency. For example, when the high-frequency tone is at exactly 1209Hz, the Goertzel filter response is approximately 0.155. However, as the frequency of the highfrequency tone moves away from 1209Hz towards 1250Hz or 1150Hz, the Goertzel filter response approaches zero.

7] The plots of FIG. 3 and FIG. 4 show that after application of an AGC, the Goertzel filter response remains relatively constant for both the low-frequency tone and the high-frequency tone no matter the frequency of the other tone.

Low-Frequency Detection with AGC and High-Frequency Power Variations [0038] FIG. 5 is a plot 500 showing the relationship between the signal power of the high-frequency tone and the Goertzel magnitude returned from the Goertzel filter centered about the low-frequency tone when both the high-frequency tone's frequency and the low-frequency tone's power are held constant. Power of the high-frequency tone, illustrated as a percentage, is shown along the x-axis. Frequency of the low- frequency tone is shown along the y-axis. Magnitude of the Goertzel filter centered about the low-frequency tone is shown in the z-axis. Frequency of the low-frequency tone varies from about 730Hz to about 810Hz. Power of the high-frequency tone is recorded from about 0% to about 200%. From the plots of FIGs. 3 and 4, it was shown that varying the frequency of the high-frequency tone has little effect on the low- frequency tone's Goertzel filter response.

9] Plot 500 shows that the familiar bell shape results as the frequency of the low- frequency tone is varied, but now there is a relationship between the high-frequency tones' power and the Goertzel filter response of the low-frequency tone. As the high- frequency lone increases in power, the Goertzel filter response of the low-frequency tone decreases. As the high-frequency tone decreases in power, the Goertzel filter response of the low-frequency tone increases.

0] This relationship exists because of the application of the AGC. Without the AGC, the Goertzel filter response of the filter centered at the low-frequency tone would not change as its power does not change. Without AGC, the plot illustrated in FIG. 5 would look much the same as the plot illustrated in FIG. 3 (if the low-frequency scale were illustrated on the same axis in the two plots). With an AGC, the power of the low- frequency tone changes in relation to the power of the high-frequency tone.

[00411 When the two tones have the same power, 100%, and the AGC is applied, the Goertzel values are approximately the same for both tones, at around 0.15, as shown in FIG. 5. If the high-frequency tone is doubled in power to 200% and the AGC is applied, then the low tone has a much lower power than if no AGC was applied, as the AGC brings the total power of both tones combined to the same level every time. The plot in FIG. 5 also shows that this relationship exists for each frequency of the lowfrequency tone.

High-Frequency Detection with AGC and High-Frequency Power Variations [0042] FIG. 6 is a plot illustrating Goertzel filter magnitude of the high-frequency tone centered at 1209Hz when the signal power of the highfrequency tone is varied and the frequency of the low-frequency tone centered at 770Hz is varied. Plot 600 shows power as a percentage of maximum signal power of the high-frequency tone along the x-axis, frequency of the low-frequency tone along the y-axis and magnitude in the z-axis.

Frequency of the low-frequency tone varies from about 730Hz to about 810Hz. As shown in FIG. 6, the high-frequency Goertzel filter response increases from 0 to approximately 0.225 as the high-frequency tone's power is varied between 0% and 200%. As further shown in FIG. 6, the frequency of the low tone does not affect the magnitude of the Goertzel filter centered at 1209Hz as the Goertzel filter response varies only minimally across the range of low frequencies.

Low-Frequency Magnitude Compared to High-Frequency Magnitude with AGC [0043] As the frequency does not affect the relationship between the lowfrequency tone's Goertzel filler response and the high-frequency tone's power, the Goertzel filter response can be adequately illustrated via a plot m two dimensions with the low- frequency tone centered at its center or target frequency of 770Hz as illustrated in FIG. 7 and the high-frequency tone centered at its center or target frequency of 1209Hz as illustrated in FIG. 8.

4] FIG. 7 is a plot 700 of the low-frequency Goertzel filter response. Signal power of the high-frequency tone is shown along the x-axis. Magnitude of the low-frequency Goertzcl filter is shown along the y-axis. As shown in FIG. 7, the low-frequency Goertzel filter response varies from about 0.25 when the high-frequency tone is at minimal signal power to just above 0 when the high-frequency tone signal power is amplified to 200%. Plot 700 shows how the low-frequency Goertzel filter response varies as a function of the high-frequency signal's power. Plot 700 is a slice of plot 500 (FIG. 5) through the x-axis at the center frequency of the low-frequency tone.

5] FIG. 8 is a plot 800 of the high-frequency Goertzel filter response. Signal power of the high-frequency tone is shown along the x-axis. Magnitude is shown along the y- axis. As shown in FIG. 8, the high-frequency Goertzel filter response varies from about 0 when the high-frequency tone is at minimal signal power to just about 0. 215 when the high-frequency tone signal power is amplified to 200%. Plot 800 shows how the high- frequency Goertzel filter response varies as a function of the high- frequency signal's power. Plot 800 is a slice of plot 600 (FIG. 6) through the x-axis at the center frequency of the low-frequency tone. [0046] FIG. 9 is a plot illustrating tl1e relationship between the

Goertzel filter magnitude of the low-frequency tone centered at 770Hz and the Goertzel filter magnitude of the high-frequency tone centered at 12091 Iz. Plot 900 illustrates high-frequency Goertzel filter magnitude along the x-axis and low-frequency Goertzel filter magnitude along the y- axis, As illustrated in FIG. 9, the low-frequency Goertzel magnitude is approximately 0.25 when the high-frequency Goertzel filter magnitude is 0. The comparison varies somewhat non-linearly over the range of filter magnitudes with the low-frequency Goertzel filter magnitude at a minimum when the high-frequency Goertzel filter magnitude is at approximately 0. 225.

7] It is now possible to compensate fhr the 12091-lz tone's variation in power, given the 770Hz Goertzel filter's magnitude. Because it is known that when both tones are of equal power they have a Goertzel value of 0.15, we can plot the relationship normalized to zero as illustrated in FIG. 10. FIG. 10 illustrates a normalized plot 1000 of the relationship between the low-frequency Goertzel filter response and the high-frequency Goertzel filter response. The x-axis depicts the high- frequency Goertzel filter response difference from an ideal tone. The x- axis varies from -0.06 to 0.04. The y-axis shows the low-frequency Goertzel filter response difference from an ideal tone. The y-axis varies from -0.05 to 0.05.

8] From FIG. 10 we can see that if the low frequencies tone's Goertzel filter magnitude is 0.04 above its ideal value then we can add about 0.048 to the high- frequency tone's Goertzel filter magnitude, in an ideal case this returns both values to 0.15, i.e., it has cancelled out the difference in powers between the two tones.

9] The relationship illustrated in the plot of FIG. 10 can be modeled using an equation. For simplicity; it is possible to model the relationship using a 2-order quadratic equation. If more accurate results are desired, the relationship can be modeled using a higher-order equation. For example, a cubic expression that models the above graph would have the form: y=-58x3 -8x2 -0.95x+0.0065 Eq. 1 [0050] Equation 1 is an implementation dependant approximation of normalized Goertzel filter magnitude relationships for the low-frequency and high-frequency tones associated with DTMF symbol "4." In the described example, the AGC was set to scale the signals to 50% of maximum and a cubic expression was used. It will be appreciated that many other variations of AGC signal scaling and multiple order equations could be used to model the relationship between the normalized Goertzel filter magnitudes of this and other low-frequency and high-frequency DTMF pairs.

[00511 Now that the energy difference has been compensated for, the only remaining variable to the Goertzel magnitudes is frequency deviation from the ideal frequency.

Using the graphs in FIGs. 2 and 3 a suitable level can be set for a pass threshold depending on the desired tolerance of the DTMF tone detector.

2] FIG. 11 is a block diagram that illustrates an embodiment of a DTMF symbol detection system 1100. As illustrated in the block diagram the DTMF symbol detection system comprises a receiver 1110 and a digital signal processor 1120. Receiver 1110 receives input signal 1103 and forwards detected signal 1105. Receiver 1110 could be configured to receive analog or digital input signals. When input signal 1103 is an analog signal, receiver 1100 may comprise an analog to digital converter. Whether input signal 1103 is analog or digital in nature, receiver 1100 may further include a low- pass filter to attenuate out-of-band signals and noise.

[00531 Digital signal processor 1120 receives detected signal 1105 and when both tones of a dual-tone pair are present, generates an encoded DTMF symbol at output bus 1135.

As illustrated in FIG. 11, digital signal processor 1120 compnses modules and units for processing detected signal 1105. More particularly, digital signal processor 1120 includes window module 1122, AGC 1124, a first or low-frequency Goertzel filter 1126, a second or high-frequency Goertzcl filter 1128, an arithmetic logic unit 1130, a determination module 1132, and an inputloutput module 1134. In addition, digital signal processor 1120 comprises memory 1140 and functional logic unit 1150.

4] Window module 1122 is configured to apply a Hamming window to the digital data stream received from receiver 1110. Sampled and window module processed data is fbi-warded to automatic gain control 1124, which in accordance with one or more parameters amplifies the processed data to account for signal attenuation due to signal transmission media. First Goertzel filter 1126 and second Goertzel filter 1128 are configured in accordance with filter coefficients and a desired filter length as stored in filter coefficient store 1144. In the example embodiment, first Goertzel filter 1126 is a discrete Fourier transform centered at 770Hz. Second Goertzel filter 1128 is a discrete Fourier transfbrm centered at 1209Hz. The windowed and gain-controlled data stream is applied to both the first Goertzel filter 1126 and second Goertzel filter 1128 to generate measures of the energy in the input signal within a close range of frequencies about each of 7701-Iz and 1209Hz.

5] In preferred embodiments, digital signal processor 1120 will include additional Goertzel filters centered at each of the remaining ideal DTMF frequencies as illustrated in FIG. 1. In alternative embodiments, filters centered at other frequencies may be further included in digital signal processor 1120 to detect the presence of these other signal tones in an input signal.

6] Arithmetic logic unit 1130 is configured to compare and normalize the magnitudes of the first and second Goertzel filters in accordance with an equation. In one embodiment, a cubic equation models the power relationship between the low- frequency and high-frequency filter responses when the input signal is gain-controlled to 50% of the maximum signal power. Equations with orders other than three may be used to model the power relationship between the tones in a D'FMF frequency pair. In the illustrated embodiment, the arithmetic logic unit 1130 receives coefficients from memory 1140. In other embodiments, the power relationship model may not he user configurable. In these other embodiments, the power relationship model store 1146 would not be required and the arithmetic logic unit, functional logic unit 1150 or some other portion of digital signal processor 1120 may he configured to implement the power relationship model, i.e., adjust the measured responses provided by first Goertzel filter 1126 and second (Joertzel filter 1128.

7] [)etermination module 1132 is configured to identify when the adjusted responses are at a level indicative of the presence of one or both DTMF frequencies in input signal 1103. Determination module 1132 compares the adjusted responses to a threshold value. Input / output module 1134 includes logic and buffers for converting the output from determination module 1132 to an encoded representation of an appropriate DTMF symbol when both the low-frequency and high-frequency DTMF tones are present in input signal 1103. Input / output module 1134 is coupled to and controls the condition of output bus 1135 when it is desired to communicate the presence of a particular DTMF tone outside of digital signal processor 1120.

8] In preferred embodiments, a Goertzel filter centered about each of the 8 DTMF standard frequencies will be run once. When the magnitude response of the respective Goertzel filter associated with one of the standard low frequencies and one of the high frequencies suggests the presence of a DTMF tone pair, the respective magnitude responses are forwarded to the arithmetic logic unit 1130 and determination module 1132 for confirmation of the DTMF tone pair. Otherwise, when the magnitude response of the Goertzel filters indicates that a DTMF tone pair is not present in the input signal, the arithmetic logic unit 1130 and determination module 1132 are bypassed until a later sample indicates that a D'FMF tone pair may be present in the input signal.

9] Memory 1140 includes system parameter store 1142, filter coefficient store 1144, power relationship model store 1146, and threshold store 1148. System parameter store 1142 includes one or more values used to initialize digital signal processor 1120.

These values may be used to manipulate window module 1122 and AUC 1124 among other digital signal processor modules and units. For example, system parameter store 1142 may include values that can be used to configure a Hamming or other type of window and direct the AGC to adjust the signal such that the first and second tones are adjusted to 50% of full power. Filter coefficient store 1144 includes the coefficients used to configure Goertzel filter 1126 and Goertzel filter 1128. Power relationship model store 1146 includes coefficients for a multiple order equation, such as -58, -8, - 0.95, and 0.0065, the coefficients in the cubic or 3d-order equation in Eq. 1 above.

Threshold store 1148 includes a value that corresponds to the magnitude of the Goertzel response at the lower and upper frequency limit. For example, a threshold of 0.13 corresponds to a frequency that is within 1. 5% of the ideal frequency. Accordingly, the corresponding DTMF tone is present in the received signal when the automatic gain- controlled and modified result is greater than 0.13.

0] It should be understood that one or more of the above mentioned stores and the values contained therein may be found within the contents of a memory device communicatively coupled to digital signal processor 1120. Stated another way, memory 1140 may be internal or external to digital signal processor 1140.

1] Functional logic unit 1150 may include one or more applicationspecific solutions to handle any of a number of desired signal processing tasks.

2] FIG. 12 illustrates an embodiment of a method 1200 for detecting tones in a signal. Method 1200 begins with block 1202 where a signal having a first frequency and a second tone at a second frequency different from the first frequency is received. In block 1204, the received signal is sampled to generate a digital representation.

Thereafter, as illustrated in block 1206, an AGC is applied to the digital representation to generate a signal of known power. In block 1208, the gain-controlled or amplified signal is applied to Goertzel filters configured with center or target frequencies that match the ideal low frequency and the ideal high frequency of a DTMF frequency pair.

Next, as indicated in block 1210, an unmodified first or low-frequency Goertzel filter response is determined from the applied gain-controlled signal. Substantially simultaneously, an unmodified second or highfrequency Goertzel filler response is determined from the applied gaincontrolled signal. As shown in block 1214, the unmodified first response and the unmodified second response are used to verify that the first and second tones each approximate the respective center frequencies of ideal DTMF tones. As is further illustrated in the flow diagram of FIG. 12, the functions within blocks 1202 through 1214 may be repeated as may be desired to determine when a particular DTMF tone pair associated with a DTMF symbol is present in the input signal.

[00631 FiG. 13 illustrates an alternative embodiment of a method 1300 for detecting tones in a signal. Method 1300 begins with block 1302 where a signal is received. In block 1304, the impact of different signal power levels of first and second tones in the received signal on respective Goertzel filter response magnitudes when a first Goertzel filter is centered at the ideal low-frequency and a second Goertzel filter is centered at the ideal high-frequency of a particular DTMF tone pair is reduced. This can be accomplished as described above by comparing and normalizing the responses of the respective Goertzel filters. Thereafter, as indicated in 1306, a determination is made whether the input signal contains a particular DTMF tone pair based on the frequency of the first tone and the frequency of the second tone in relation to the unmodified power levels that are expected from the respective Goertzel filters when ideal DTMF frequencies are present in the input signal.

Symbol "4" Detection [0064] As in the above discussion, we continue to use DTMF symbol "4" to illustrate operation of the systems and methods for DTMF symbol detection. In accordance with FIG. 1, DTMF symbol "4" is represented by a low- frequency tone centered at 770Hz and a high-frequency tone centered at 12091 Iz. The DTMF detector receives a signal comprising a first tone centered at a first frequency of 770Hz and a second tone centered at a second frequency of 1209Hz. The DTMF detector samples the signal and applies a hamming window to the sampled signal. l'hereafter, the sampled signal is scaled by an AGC to adjust for "twist" due to amplitude variation over frequency. th the example described above, the AGC scales the input signal to bring the power level to 50% of a maximum power. After application of the AGC, the amplified signal is applied to a first Goertzel filter having a target frequency of 770Hz and a second Goertzel filter having a target frequency of 12091 lz. The first and second Goertzel filters generate a first filter response and a second filter response. In an example, the low-frequency Goertzel filter returned a magnitude of 0.1 and the high-frequency Goertzel filter returned a magnitude of 0.2.

5] At this stage it appears as if the response from the low-frequency Goertzel filter would be too low to pass a detection decision. However, because an AGC has been applied, the response from the high-frequency Goertzel filter is higher than it should be if the energy was spread equally between the tones.

6] The DTMF symbol detector detennines an unmodified first filter response and an unmodified second filter response using the amplified signal. This determination can be accomplished by applying both the first and second filter responses to equation 1.

This results in a normalized high-frequency Goertzel filter magnitude of 0.045, which, when applied to equation 1, generates an adjusted low magnitude of 0.1577 and an adjusted high magnitude of 0.1550. The DTMF detector then uses the adjusted first filter response and the adjusted second filter response to verify that the first tone approximates the first center frequency and the second tone approximates the second center frequency. When the frequency confirmation range is set to 1.5%, adjusted (Ioertzel filter response magnitudes of greater than about 0.13 are indicative of tones in the input signal within 1.5% of the ideal center frequencies of 770Hz and 1209Hz.

Accordingly, a valid DT'MF "4" symbol is present in the input signal.

7] In summary, two (ioertzel filters centered at the ideal lowfrequency and the ideal high-frequency tones (see FIG. 1) are used to reliably detect the presence of the low-frequency and the high-frequency tones associated with a DTMF symbol "4," in an input signal, whereas conventional detection methods would use an additional two Goertzel filters centered on the 2'' harmonic frequencies of the low-frequency and the high-frequency tones to confirm presence of the tones.

8] The above method produces a robust detector with virtually no false detects during speech when limits are placed on how much compensation is provided for signal energy and frequency drift. For example, if the low-frequency tone signal power is between -4dB and +8dB different from the high-frequency tone signal power and the threshold for frequency deviation is set to 1.5% of the ideal DTMF frequency, there are no false detects.

9] Any process descriptions or blocks in the flow diagram presented in FIGs. 12 and 13 should be understood to represent modules, segments, or portions of code or logic, which include one or more executable instructions for implementing specific functions or steps in the associated process. Alternate implementations are included within the scope of the present systems and methods in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art after having become familiar with the teachings of the present systems and methods for DTMF symbol detection.

0] The foregoing description has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the scope of the claims to the precise forms disclosed. Modifications or variations are possible in light of the above teachings. The embodiments discussed, however, were chosen and described to enable one of ordinary skill to utilize various embodiments of the systems and methods for DTMF detection. All such modifications and variations are within the scope of the appended claims when interpreted in accordance with the breadth to which they are fairly and legally entitled.

Claims

What is claimed is: I I. A niethod for detecting tones in a signal, comprising: 2 receiving a signal comprising a first tone at a first frequency and a second tone at 3 a second frequency wherein the second frequency is different from the first frequency; 4 sampling the signal to generate a digital representation of the signal; applying an automatic gain control to the digital representation of the signal to 6 generate a gain-controlled signal at a desired level; 7 applying the gain-controlled signal to a first filter and a second filter to generate 8 a first filter response and a second filter response, respectively, the first filter having a 9 first center frequency and the second filter having a second center frequency wherein the first center frequency is different from the second center frequency;

11 determining an unmodified first filter response using the gain- controlled signal; 12 determining an unmodified second filter response using the gain- controlled 13 signal; and 14 using the unmodified first filter response and the unmodified second filter response to verify that the first tone approximates the first center frequency and the 16 second tone approximates the second center frequency.

1 2. The method of claim 1, wherein determining an unmodified first filter 2 response comprises identifying a first difference between the gaincontrolled first filter 3 response and a nominal signal energy value.

1 3. The method of claim 2, wherein the nominal signal energy value 2 comprises the energy in a first filter response when the first frequency substantially 3 matches the first center frequency and the signal strength is at a desired level.

1 4. The method of claim 2, wherein the first difference is applied in an 2 equation representing the relationship between the unmodified first filter response and 3 the unmodified second filter response.

1 5. The method of claim 4, wherein the equation is derived from 2 experimental data.

1 6. The method of claim 1, wherein determining an unmodified second 2 response comprises identifying a second difference between the gaincontrolled second 3 filter response and a nominal signal energy value.

1 7. The method of claim 6, wherein the nominal signal energy value 2 comprises the energy in the second filter response when the second frequency 3 substantially matches the second center frequency and the signal strength is at a desired 4 level.

8. The method of claim 6, wherein the second difference is applied to an 2 equation representing the relationship between the unmodified first filter response and 3 the unmodified second filter response.

1 9. The method of claim 8, wherein the equation is derived from 2 experimental data.

1 ft A dual-tone detection system, comprising: 2 a receiver configured to receive an input signal having a first tone at a first 3 frequency and a second tone at a second frequency wherein the second frequency is 4 different from the first frequency; and a digital signal processor coupled to the receiver, the digital signal processor 6 comprising: 7 a window module configured to generate a representation of the input 8 signal; 9 an automatic gain control module configured to generate a scaled representation such that the tones are at respective desired amplitudes; 11 a first filter centered about a first frequency and configured to receive the 12 scaled representation and generate a first measurement; 13 a second filter, centered about a second frequency different from the first 14 frequency, configured to receive the scaled representation and generate a second measurement;

16 an arithmetic logic unit configured to use an equation that models the 17 power relationship between the first and second tones to generate adjusted first 18 and second measurements, respectively; and 19 a determination module responsive to a comparison of each of the adjusted first and second measurements with a threshold value to identify when 21 the input signal contains the low-frequency and high-frequency tones associated 22 with a particular DTMF symbol.

1 11. The system of claim 10, wherein the equation represents the relationship 2 between the signal energy that would have been identified by the first filter had the 3 representation of the signal been applied to the first filter and the signal energy that 4 would have been identified by the second filter had the representation of the signal been applied to the second filter.

1 12. The system of claim 10, wherein the nominal value comprises the signal 2 energy in one of the first filter response and the second filter response when a tone 3 present in the signal has a frequency that substantially matches a center frequency of one 4 of the first filter and the second filter and the signal strength is at the desired level.

I 13. The system of claim 10, wherein the equation is derived from 2 experimental data.

1 14. The system of claim 10, wherein the equation is a quadratic equation.

I 15. The system of claim 10, wherein the equation has an order greater than a 2 quadratic equation.

1 16. A dual-tone detection system, comprising: 2 means for receiving an input signal; 3 means fir reducing the impact of different signal power levels of first and 4 second tones have on respective Goertzel filter response magnitudes when a first Goertzel filter is centered at the ideal low-frequency and a second Goertzel filter is 6 centered at the ideal high-frequency of a particular DTMF tone pair; and 7 means for determining whether the input signal contains a particular DTMF tone 8 pair based on the frequency of the first tone and the frequency of the second tone in 9 relation to the ideal DTMF frequencies.

17. The system of claim 16, wherein the means for reducing the impact of 2 different signal power levels of the first and second tones on respective Goertzel filter 3 response magnitudes comprises comparing and normalizing the Goertzel filter response 4 magnitudes.

18. The system of claim 17, wherein comparing and normalizing the 2 Gocrtzel filter response magnitudes comprises modeling the relationship between a low- 3 frequency Goertzel filler response difference from the magnitude of the response of the 4 low-frequency Goertzel filter when a first ideal tone is present and a high-frequency Goertzel filter response difference from the magnitude of the response of the high- 6 frequency Goertzel filter when a second ideal tone is present.

I 19. The system of claim 18, wherein modeling comprises generating an 2 equation.

1 20. The system of claim 19, wherein the equation has at least a 2 order.