TONE DETECTOR WITH IMPROVED PERFORMANCE IN THE PRESENCE OF SPEECH
BACKGROUND OF THE INVENTION This invention is directed in general to tone detectors and more specifically to tone detectors that must function to detect tones in a communications channel where speech or other signals may be present. This invention specifically, although not exclusively, is suitable for digit detection with dual tone multifrequency (DTMF) such as in a telecommunications system. It is becoming increasingly important to be able to detect tones such as DTMF signals in a communications channel that can also carry other signals such as speech. An ideal detector detects a predetermined set of tones and does not confuse other signals such as speech as one of the predetermined tones. Services that are chosen by DTMF signaling in a telecommunications system, every time it becomes more popular. However, since other background noises or signals may be present in the same channel that carries the DTMF digits, the detector or tone receiver such as in a central office, must be able to reliably detect the DTMF digits, while minimizing simultaneously false detections that can be caused by other signals or voice in the channel.
DTMF digit detectors must operate under difficult conditions. First, the detector must be able to correctly detect a DTMF digit received over a range of characteristics defined by signaling standards that define a valid digit, including variations in tone frequency, pitch amplitude (both an absolute value, and relative to the amplitude of the dual tone pair) and subject to intermodulation distortions and other interference or noise with defined limits. Secondly, the DTMF detector should also minimize the number of disconnections of had or false detections due to the presence of speech. Many previous DTMF detectors have achieved reasonably good performance in one of these two requirements, but have typically not been fully successful in simultaneously achieving both of these requirements. In this way, there is a need for an improved tone detector that not only allows for required variations of the tones but also minimizes false detection such as due to speech disconnections. BRIEF DESCRIPTION OF THE INVENTION An object of the present invention is to provide a corresponding tone detector and method that provides improved performance while substantially satisfying both of the previously established requirements.
According to an illustrative embodiment of the present invention, a DTMF receiver segregates tone processing in a low frequency group and a high frequency group. A filter for band elimination with high tones, eliminates the high tones of the low frequency group, - and a filter for elimination of low tones band, eliminates the low tones of the high frequency signals. Separate bandpass filters for each low frequency tone operate on the filtered low frequency signal, - separate bandpass filters operate to filter each high frequency tone from the filtered high frequency signal. Filtered signals of low frequency and high frequency are used to generate low and high frequency variable threshold signals. The low frequency signals filtered in bandpass and the high frequency signals filtered in bandpass are compared to the variable threshold signals of low frequency and high frequency, respectively, to determine whether one or more low frequency or high frequency tones they are present. Processing with post-tone detection is also used before a final decision is made as to whether a valid DTMF tone has been received. Low frequency and high frequency tone decisions are validated to determine if a valid tone could have been received. The amplitudes of the respective tones of low frequency and high frequency, they are compared to ensure that they meet the amplitude requirements defined as a valid digit. Also, the duration of the low and high frequency tones is checked against synchronization requirements that also specify the requirements for a valid digit. A valid DTMF digit is reported only if appropriate low and high frequency tones have been determined and if the post-processing checks further validate the detection of a digit. BRIEF DESCRIPTION OF THE DRAWINGS Figures 1 and 2 illustrate a block diagram of a modality of a tone receiver, in accordance with the present invention. Figure 3 illustrates a telecommunications network that incorporates a DTMF digit receiver according to the present invention. Figure 4 illustrates an exemplary embodiment of a DTMF digit receiver as illustrated in Figure 3. Figures 5, 6, 7 and 8 are graphs illustrating the relationship of the amplitude of a detected signal to the threshold amplitude. variable for different signal environments. DETAILED DESCRIPTION In order to better appreciate the explanation of one embodiment of the present invention, it is important to appreciate the nature of the detected DTMF tones. Table 1 below illustrates the eight separate tones in four lower frequency tones and four higher frequency tones. Each valid DTMF digit consists of the simultaneous transmission of a tone for the lower frequency group and a tone for the higher frequency group. TABLE 1 Combination of Frequency DTMF Group of Ba to Frequency (Hz) Group of High Frequency (Hz) 1209 1336 1477 1633 697 1 2 3 A 770 4 5 6 B 852 7 8 9 C 941 * 0 # D It will be apparent that for a typical voice frequency channel such as 300-3000 Hz, the tones constitute band signaling. In this way, other signals, including speech, will also generate signals of the same frequencies used by the DTMF tones. Additional parameters specify requirements for valid digit detection. A frequency variation of each tone of ± 1.8% must be accepted. Maximum and minimum power levels of <; 0.8 dbm and > - 33.55 dbm, respectively are defined, the maximum difference (rotation) between amplitudes of the two tones should be between -8.75 and 7.25 dB. Also, a pitch duration of > 39.5 milliseconds Additional requirements are related to interference signals and noise requirements. For example, signal echoes delayed up to 20 milliseconds and 16 dB below the primary signal, should not interrupt the detection of a digit. Also, interference signals in the 300-480 Hz range with a total power level of -46 dbm should be tolerated as well as interference signals in the 480-3400 Hz range with a total power level of 20 dB below the low signaling frequency level or -46 dbm, whichever is greater. A receiver must operate at a signal-to-interference ratio of 16 dB with respect to the lowest amplitude. Although the signal distortion is not clearly defined in terms of acceptable parameters, certain amounts of signal distortion must be tolerated based on the allowable distortion, granted for corresponding DTMF transmitters. Thus, it will be apparent that the extent of conditions and variations to be tolerated impose substantial burdens on a DTMF receiver, especially considering the desirability of minimizing speech disconnection and other conditions of voice and signal interference. With reference to Figure 1, a communications channel 10 receives digital samples of data received in pulse code modulation (PCM) format such as at a rate of 8 kilohertz (kHz). This data may represent information received from a subscriber in a telecommunications system and shall include DTMF generated tones as well as information such as voice, interference or other signals that are carried on the channel. In the illustrated mode, the data is processed in digital format, to determine if a DTMF digit has been received. However, it will be apparent to those skilled in the art that analog components or a combination of analog and digital implementations can be employed to perform the steps and elements according to the illustrative embodiment of the present invention. A mechanism 12 for descending sampling converts 8 kilohertz PCM data into 4 kilohertz data that is transmitted to the high pass filter 14. The high pass filter 14 is used to attenuate 60 Hz power line frequency components and harmonics of these power line components. In applications where the dial tone is present, the high pass filter 14 can also be designed to attenuate it as well as the power line components. The high-pass filtered data is divided and provided as feeds to the high-tone band 16 filter and the low-tone band-removing filter 18. Both of these filters are similar in that their goal is to pass other frequencies while which eliminates a predetermined band of frequencies. In the case of filter 16, the frequencies of 1209 Hz-1633 Hz (high frequency tone group) are eliminated while passing higher and lower frequencies to this band. Similarly, the filter 18 rejects or attenuates signals at 697 Hz-941 Hz (low frequency tone group) while other frequencies pass. In this way, filters 16 and 18 serve to attenuate the other low frequency or high frequency tones of the tone pair that make up a single DTMF digit. The filtered output 20 of the filter 16 provides a power to the bandpass filters BPF1, BPF2, BPF3 and BPF4. Each of these bandpass filters is centered with respect to each of the four low frequency tones of the DTMF tone pair. The output 20 is also received by the energy detector 22 which determines the total energy present in the filtered signal on the line 20. The energy detector 22 may comprise a quadrature-type detector, which generates an average energy level and components of high frequency unwanted. The detected energy signal from the energy detector 22 is flattened by the low pass filter 24 and drifted to a lower frequency group variable threshold signal
(LVT). Similarly, the output 26 of the filter 18 provides a power to the band pass filters BPF5, BPF6, BPF7 and BPF8 which are centered with respect to each tone in the high frequency group. The energy at the outlet 26 of the filter 18 is recorded by the energy detector 28. The detected energy signal is filtered with a low pass through the filter 30, to result in the variable threshold signal of the high frequency group (HVT). The filtered bandpass output signals BPF1-BPF8 as well as the LVT and HVT signals are generated at a rate of 4 kilohertz and then preferably are sampled in descending order at a rate of 1 kilohertz per the downstream sampling circuit 32. For example, downward sampling can be achieved by simply operating on every fourth signal sample. Likewise, the descending sampler 12 can operate in a similar manner by operating simply on any sampled sample that is provided on channel 10. The purpose of downward sampling is to minimize the structure, the number of comparison stages, and the computing power required. to process the signals, while still resulting in reliable DTMF digit detection decision. By employing descending sampling, in this way the number of instructions per second required to provide the indicated functionality is reduced, minimizing the number of calculations required as well as reducing the computational power required to implement the illustrative technique. The amplitude sensors 34, 36, 38 and 40 detect the amplitude of the output signals sampled in descending form of the bandpass filters BPF1, BPF2, BPF3 and BPF4, respectively. Similarly, the amplitude sensors 42, 44, 46 and 48 detect the amplitude of the output signals sampled in descending form of the bandpass filters BPF5, BPF6, BPF7 and BPF8, respectively. The outputs F1-F8 of the amplitude sensors 34-48 represent the energy associated with the signals, if any, that occur during the corresponding data sample for each of the eight discrete tones that can comprise a DTMF digit. The sampled LVT and HVT signals descending are provided as output signals LVT1 and HVT1. With reference to Figure 2, a maximum selector 54 receives feeds' of the signals LVT1 and LVT2 and chooses the signal with the greatest quantity that is sent out and receives as a power the maximum selector 56. The register 50 receives the output of the maximum selector 54 and store this value for an interaction (sample). This register value is a feed to the multiplier 51 that also receives a power from the decay factor register 55. The output of the multiplier 51 is LVT2. The decay factor stored in the register 55 is a constant value less than 1, which limits the rate of decay of the signal outside the maximum selector 54. The other supply of the selector 56 comprises a signal stored in the register 58 representing a threshold of minimum amplitude for low frequency group, to detect a valid low frequency tone. The selector 56 chooses the maximum value of its feeds to pass through its output 60. Similarly, a maximum selector 62 chooses the maximum amplitude signal between the signals HVT1 and HVT2, to provide an output to the maximum selector 64. The register 52, multiplier 53 and decay control register 57, operate in a manner similar to the previously described register 50, multiplier 51, and record decay factor 55. The selector 64 also receives as a power, a signal representative of the threshold of minimum allowed amplitude of a high-frequency tone group that is allowed to be detected as a valid one-digit DTMF tone. The maximum selector 64 chooses the largest of the two power signals and provides this selected signal as an output 68. The values stored in registers 58 and 66 adjust the minimum amplitude threshold below which the received tones will be rejected. Two sets of comparators use the reference signals 60 and 68 respectively, to determine individually if each of the possible signals F1-F8 is present. The comparators 70, 72, 74 and 76 use the dynamically variable threshold based on the LVT signal and compare the energy levels associated with the signal Fl, F2, F3 and F4 respectively. The output of these comparators is TRUE or a logic "1" if the associated frequency signal is larger than the threshold; the output is FALSE or a logic "0" if the respective signal is less than the threshold. Similarly, the comparators 78, 80, 82 and 84 use the dynamic threshold signal 68 based on the HVT signal, and make comparative determinations as to whether the tones associated with the signals F5, F6, F7 and F8, respectively, are present. Records 86 and 88 contain, for each sample, the respective output decisions for the set of low and high frequency comparators. In the illustrative mode, only the output of comparators 72 and 82 where TRUE (TRUÉ) represents an intermediate decision that the low frequency tone 697 Hz and the high frequency tone 1477 Hz were detected, ie the digit DTMF 3. A validity checker 90 determines whether the contents of registers 86 and 88 can contain a valid DTMF digit during each 1 kilohertz sample. The validation includes determining if one and only one tone is detected in each of the registers 86 and 88. To understand a valid DTMF digit, only a low frequency tone and a high frequency tone must be present during each sample. If the validity checker 90 determines that a valid DTMF tone may be present, a validation output is provided to an additional validity checker 92 that verifies the amplitudes of the tones with each other; that is, a tone must be within a predefined range in the amplitude of the other tone.
Considering that the tones Fl and F7 are present, the validity checker 92 determines whether the amplitude of the tone Fl is within the range of amplitude required with respect to the amplitude of the tone F7. A validation output from the validity checker 92 is provided to a final validity checker 94 which verifies the duration of the tone against specified limits. That is, the number of samples during which registers 86 and 88 have remained constant are counted to establish a length of time of the received tone length that can represent a DTMF digit. Also, validity checker 92 must be satisfied during this account. In the illustrative embodiment, it is preferred that the tones are detected for a maximum length of time, consistent with the minimum length defined in order to effect a determination of a valid digit. This helps in distinguishing against false detections, by requiring a longer continuous duration in which the same tones must have been detected. According to an exemplary embodiment, approximately 25 milliseconds of a minimum duration of 40 milliseconds of a DTMF tone are required to be detected in order to constitute a valid decision by the validity checker 94 that a DTMF digit has been received. Further discussion concerning the synchronization criterion will be made with respect to Figures 5-8. The output channel 96 carries a digit output decision from the validity checker 94, which constitutes a final determination of the received digits. In accordance with the exemplary embodiment, output 96 will be indicative that the DTMF digit "3" was received. If the tones represented by the registers 86 and 88 do not remain in stable and constant state, or do not pass the validity checker 90 and 92, for the minimum time determined by the validity checker 94, then the validation process is readjusted. Although various implementations and elements may be employed to meet different design requirements, "the following specific elements are described for an illustrative embodiment: The high pass filter 14 which may comprise two filter sections (digitally implemented) having a cutoff frequency to minimize the power line and other very low frequency signals The filters for band 16 and 18 elimination, each can comprise three filter sections (digitally implemented) that have tuned drum frequencies to eliminate band tone sets respective high and low band bandpass filters BF1-BF8 can each be achieved with two-pole IIR bandpass filters.Although a two-pole filter does not provide extremely high selectivity, each filter achieves a response time relatively high (fast) This improves the ability to provide disconnect performance n appropriate speech. The attenuation provided by the filters at + 1.8% of the center frequency should be adjusted to provide equal amplitude with respect to the amplitudes of the variable threshold signal. The rejection or attenuation of the signal must be sufficient to satisfy a requirement to reject tones that are + 3% of the specified center frequency of each of eight tones. These filters provide a difference of 1.5 dB between frequencies to + 1.8% from the center frequency and frequencies to + 3.0% from the center frequency. In the illustrative embodiment, it is considered that the proper setting -from the bandpass filters with respect to the outputs and response characteristics of the low pass VTN filters 24 and 30, contribute to the improved performance of the digit detector. The energy detectors 22 and 28 may comprise conventional energy detection techniques such as employing a quadratic function in order to determine energy levels. Individual tone amplitude detectors 34-48 use a technique based on the following formula: Quadratic amplitude = xl * xl -2 * cos (wO * T) * xl * x2 + x2 * x2 where: wO = frequency of a sine wave T = sampling period xl = a sample value of the sine wave x2 = next sequential value of the sine wave This formula is used to determine the quadratic amplitude of the tones in each central frequency wO. The measured amplitude is substantially accurate within a constant multiplier and is independent of the selected sample values. If the frequency of a current signal is close to wO, then the amplitude of this sine wave is also measurable by this formula, but with an error that increases as the frequency of the current tone varies from the center frequency. Within the required bandwidth + 1.8%, the precision at which the amplitude can be measured is within a few tenths of a decibel (dB). In view of the 1 kilohertz down sampling that results in a sample every millisecond, this technique minimizes the number of calculations required to calculate the corresponding amplitudes for the outputs of the eight bandpass filters. Amplitude detectors 34-48 also provide a scaling function with respect to variable threshold levels. With reference to Figure 3, a central switch 100 supports a plurality of customer facility equipment 102 and 104 which may include conventional telephone equipment using DTMF signaling. The central switch 100 is connected by trunks 106 to a quota telecommunications network 108. The switch includes a DTMF digit receiver 110 that can be selectively coupled to receive digitized PCM data from a subscriber line, to determine if one or more DTMF digits they have been transmitted by a corresponding user. The output of the digit receiver 110 is supplied to the central switch which makes decisions concerning the addressing of a call or other services, based on the DTMF digits that have been received. Of course, services other than a central switch can also benefit from the present invention. The application of a DTMF receiver as described in Figures 1 and 2, can be employed as the DTMF digit receiver 110 as used in a telecommunications application. However, it will be apparent to those skilled in the art that these applications of this digit receiver can be used in a variety of applications where tone or digit detection is required, especially in an environment where other signals are present and that must be distinguished from the detection of these tones. Figure 4 illustrates an architectural implementation of the exemplary embodiment of the detector illustrated in Figures 1 and 2. A digital signal processor such as a DSP32C available from AT & amp;; T, is employed to implement the functionality illustrated in Figures 1 and 2. A read-only memory (ROM) 114 provides control instructions for DSP 112. An associated random access memory (RAM) 116, is coupled to the DSP 112 and provides a location where the received data and other DSP 118 parameters can be stored. The digital data channel 118 is coupled to DSP 112 and the input / output interface 120 which provides an interface that receives the digital input data on channels 122 and 124, such as in PCM format and provides a mechanism for sending out digital data on line 126, representing DTMF decisions that can be driven by the auxiliary equipment (not shown). Preferably, DSP 112 is capable of simultaneously processing a plurality of power PCM data samples, ie more than a single channel, and can make tone decisions concerning more than one channel at a time. The output channel 126 (which corresponds to the output decision channel 96 in Figure 2) likewise provides decision information concerning a concurrent plurality of power data processed concurrently by the digital signal processor 112. The number of channels that can be to be served concurrently by DSP, will depend on the particular design implementation, the rate of downward sampling used, and the particular implementation of the filters and energy calculation required according to the illustrative modality. Figures 5-8 are graphs showing the relative magnitude of a variable threshold signal with respect to the amplitude of corresponding tone outputs. The horizontal axis represents the time and in the exemplary mode it represents a total time of approximately 50 milliseconds with a pitch duration of 40 milliseconds. In these graphs it is considered that the signal value corresponds to the quadratic amplitude of tone Fl and that the variable threshold signal corresponds to the output 60 of the maximum selector 56. Figure 5 illustrates the variable threshold signal 130 (quadratic amplitude) with respect to a quadratic energy level detected Fl, represented by graph 132. It is observed that the time of ascent to curve 130 representing the variable threshold is faster than that of signal 132, - this is due to the slightly wider bandwidth of the low pass filter VTN, compared to the bandpass filter BPFl. The curve 132 initially crosses and becomes larger than the curve 130 at the point 134. The curve 132 decreases and becomes less than the curve 130 at the point 136. The time T5 between points 134 and 136 defines the time in the that the data samples are determined to meet the appropriate tone criteria, that is, the curve 132 remains continuously on the variable threshold curve 130, during this time interval T5. This reflects a continuous period of detection of the corresponding tone generated by the curve 132. In the exemplary curve as illustrated in Figure 5, the tone has an exact center frequency and is transmitted with a higher frequency tone paired properly, in an environment relatively free of frequencies. The vertical distance between the curves 132 and 130 at the point of maximum separation represents a small relative amount such as 1.7 dB. According to a preferred embodiment of the present invention, the proximity of these two curves during the time in which the curve 132 exceeds the curve 130, aids in the performance of the illustrative mode to reject speech and false disconnection signals. Figure 6 is similar to Figure 5, where the variable threshold curve 140 is illustrated with respect to a curve 142 representing the signal detection frequency Fl. It will be noted that the curve 142 is relatively heterogeneous and has what may be termed fluctuation or associated noise. This is due to the fact that Figure 6 illustrates that the tone associated with curve 142 was 1.8% above the appropriate center frequency. As you will see, this results in some choppy or uncertainty in relation to curve 142 with respect to curve 132, especially in the area where curve 142 exceeds curve 140, which is determined between points 144 and 146 that define the time interval T6. This 1.8% variation represents the maximum frequency deviation that a tone can deviate from and constitute a valid digit. The complementary high-frequency tone that would have been generated concurrently is not illustrated. The time interval T6 is substantially equivalent to T5 and will result in meeting the criteria required for a valid DTMF digit detection. It will also be noted that the magnitude of the curve 142 with respect to the curve 140 is substantially closer during the time T6 than the distance between the curve 132 and the curve 130 in the time interval T5. This is primarily caused by the fact that the off-center frequency is further attenuated by its corresponding bandpass filter since it is not a center frequency. The vertical distance between curves 142 and 140 at the point of maximum separation is 1 dB. Figure 7 illustrates a plot where both variable threshold signals are 150 and the detected energy signal 152 have substantial oscillations even during the interval between crossing points 154 and 156 defining the time interval T7. This graph illustrates a received tone that is + 1.8% high from its center frequency, a higher accompanying frequency tone (not shown) of 6 dB over the lower frequency tone, and intermodulation products that are 16 dB below the lower frequency tone magnitude (intermodulation products) occurring at 1813 hertz and 767 hertz. The impact of the intermodulation products, especially the frequency product of 767 hertz which is relatively close to the lower frequency tone Fl, results in a sinusoidal variation during the time period T7. However, it will be noted that at no time during period T7, curve 152 intercepts or goes below curve 150. In this way, even in the presence of these interference signals and on the higher frequency tone pair of amplitude (not shown), the signal represented by 152 does not intercept or cross the corresponding variable threshold curve 150 during the time interval T7. That is, T7 represents a continuous interval in which the tone 152 will be detected in accordance with the present invention. Figure 8 illustrates a graph in which both the variable threshold signal 160 and the detected energy signal 162 have substantial interference during the interval between the crossing points 164 and 166, defining the time interval T8. This graph illustrates a signal-to-interference ratio of 16 dB, using white Gaussian noise. It will be noted that even during the presence of this noise, curve 162 does not intersect or intersect curve 160 during period T8. In this way, even in the presence of this white noise, the signal represented by the curve 162 does not intersect or cross the corresponding variable threshold curve 160 during the time interval T8. That is, T8 represents a continuous interval where tone 162 will be detected according to the present invention. Figures 7 and 8 show that with the addition of signals outside the bandwidth of the desired tone, the margin or difference between the signal curve and the variable threshold curve tends to narrow. The wider margins illustrated in Figures 5 and 6, between the received signal and the variable threshold signal, are provided in such a way that even in the presence of strange signals and interference conditions, as illustrated in Figures 7 and 8 , there will be enough margin to maintain consistent tone detection. With reference to Figure 5, it will be noted that the rear edges of the curves 130 and 132 are arranged such that the curve 132 has a faster fall rate than the curve 130. This is generally convenient in order to provide measurements repetitive which would be more difficult to achieve if both curves had approximately the same slope at the point of intersection at point 136. The collapse of the slope of curve 130 is controlled by the feedback loop of the elements 50, 51 and 55 illustrated in Figure 2 and by the selection of the maximum selector 54. This provides a more gradual slope for the curve 130 than would have occurred had only a current sample been used. The presence of substantial voice energy, not attenuated by band rejection filters 16 and 18, with respect to the magnitude of a tone that may occur simultaneously, will nominally result in the variable threshold curve being greater than the pitch energy curve. Narrow band, due to the additional energy components that are provided by the in-band voice characteristics. This will of course prevent a detection of a valid DTMF tone if voice signals of a significant magnitude are present. This represents a compromise between the desired ability to detect a tone during other strange signal conditions and the ability to not falsely detect a tone when other significant signals are present. Several factors are considered that positively impact the good performance of the DTMF receiver according to the modality of the present invention in both detection tones, within the various parameters defined for acceptable tones while simultaneously minimizing foreign false detections and speech disconnection. . The separation of tones into groups of high and low filters with the corresponding opposite group removed by a bandpass filter is considered to be beneficial. By using these high and low filtered bands to generate a dynamically variable threshold against which the narrowest bandwidths are compared, they contribute to the results achieved. The ability to sample in a downward manner minimizes the number of instructions and computing power required to implement the functions according to the present invention. The validation stages, so-called post-processing 90, 92 and 94, provide additional criteria ensuring that a DTMF digit is recognized that has been received. A corresponding exemplary method also in accordance with the present invention has been explained with respect to the illustrative embodiment. This invention also contemplates additional variations and modifications. Although descending sampling at 4 kHz and then 1 kHz is described, other downward sampling rates may be used depending on performance and computational compensations. Bandpass filters with more poles can be used to crush the band frequency response feature to provide a more consistent vertical distance between the tone response curve and the VTN curve: However, the time response of The bandpass filters with respect to the time response of the low pass filters associated with the variable thresholds will have to be maintained. Although one embodiment of the present invention has been described above and illustrated with the accompanying drawings, the scope of the invention is defined by the claims that follow. It is noted that in relation to this date, the best method known to the applicant to carry out the aforementioned invention, is that which is clear from the present description of the invention. Having described the invention as above, property is claimed as contained in the following: