Patent Translate Powered by EPO and Google Notice This translation is machine-generated. It cannot be guaranteed that it is intelligible, accurate, complete, reliable or fit for specific purposes. Critical decisions, such as commercially relevant or financial decisions, should not be based on machine-translation output. DESCRIPTION JP2014042103 Abstract: To provide a howling suppressor capable of adaptively suppressing howling even in an environment where a bandwidth has howling. SOLUTION: In an environment in which a bandwidth has a frequency howled, a bandwidth of one bandwidth is converged to a filter coefficient for suppressing a frequency by adaptively updating a filter coefficient related to a removal frequency and a removal bandwidth of an adaptive notch filter. Can suppress howling. Furthermore, filter coefficients for suppressing frequencies with a plurality of bandwidths by connecting such adaptive notch filters in multiple stages and using a multistage adaptive notch filter integrated into one and adaptively updating the filter coefficients Converge, and suppress howling. [Selected figure] Figure 3 Howling suppressor and program, adaptive notch filter and program [0001] The present invention relates to a howling suppression apparatus and program, and an adaptive notch filter and program, which can be applied to, for example, an audio receiving apparatus for a high-speed speech communication system such as a video conference or a teleconference. [0002] In a speech communication system such as a video conference or a telephone conference, sound echoed from a speaker (sound, voice, etc.) gets into a microphone and an acoustic echo is generated back to the transmitter side. 15-04-2019 1 Furthermore, when the gain of this feedback system becomes 1 or more, an oscillation state called so-called howling is achieved. Since howling is a significant hindrance to calls, much research and development have been conducted on how to suppress them. [0003] One method of suppressing howling is using an adaptive notch filter. The notch filter is a filter that removes (attenuates) a specific frequency, and the adaptive notch filter is a filter that adaptively updates filter coefficients to form a notch (removed frequency). By applying an adaptive notch filter to the howling suppressor and updating the filter coefficients of the adaptive notch filter, it is possible to adaptively suppress the howling frequency. Non-Patent Document 1 proposes an adaptive notch filter applicable to the howling suppressor. [0004] The adaptive notch filter described in Non-Patent Document 1 removes a specific frequency by obtaining an update amount of the filter coefficient that determines the removal frequency from the coefficient of the input signal of the adaptive filter and the denominator of the transfer function and updating it. . [0005] Masaki Kobayashi, Tomohiro Akagawa, Yoshio Itoh, "Algorithm and Convergence of Adaptive Notch Filter Using All-Pass Filter," Transactions of the Institute of Electronics, Information and Communication Engineers, Vol. J82-A, no. 3, pp. 325-332, March 1999 [0006] However, in the adaptive notch filter of Non-Patent Document 1, the applicable parameter is only the removal frequency, and the removal bandwidth needs to be set in advance. Because the howling characteristics differ depending on the room shape etc., if the removal bandwidth of the 15-04-2019 2 adaptive notch filter is the default setting, sufficient howling suppression performance may not be obtained in the howling suppression apparatus to which the adaptive notch filter is applied. . Adaptation of the removal bandwidth requires a separate updating equation, but the frequency and bandwidth are separately updated, and there is a problem that convergence of the adaptive notch filter is not guaranteed. [0007] The present invention has been made in consideration of the above points, and an object of the present invention is to provide a howling suppressor and program capable of suppressing howling whose suppression bandwidth is not a predetermined width. [0008] Furthermore, the present invention is intended to provide an adaptive notch filter and program that can cope with changes in one or more frequencies to be removed (the removal bandwidth is not limited to a predetermined width). [0009] A first aspect of the present invention is an adaptive notch filter unit having a transfer function Hf (z) represented by the following equation (A1), in an adaptive notch filter capable of adaptively changing the frequency to be removed, the frequency to be removed being different , And j (j is a natural number of one or more) connected in stages and characterized by having a transfer function equivalent to. [0010] In equation (A1), n is time, and a2 (n) and b1 (n) are filter coefficients, respectively, and are updated according to equations (A2) and (A3). The filter coefficient a2 (n) is a filter coefficient related to the removal bandwidth. In equations (A2) and (A3), μ is a step gain, and Δa2 and Δb1 are filter coefficient update amounts calculated according to the filter coefficient update algorithm. 15-04-2019 3 [0011] A second invention of the present invention is an adaptive notch filter program that causes a computer to function as an adaptive notch filter capable of adaptively changing the frequency to be suppressed, wherein the adaptive notch filter has different frequencies to be removed. The adaptive notch filter unit having a transfer function Hf (z) expressed by the equation (1) is characterized by having a transfer function equivalent to j (j is a natural number of 1 or more) stages connected. [0012] In equation (B1), n is time, and a2 (n) and b1 (n) are filter coefficients, respectively, and are updated according to equations (B2) and (B3). The filter coefficient a2 (n) is a filter coefficient related to the removal bandwidth. In equations (B2) and (B3), μ is a step gain, and Δa2 and Δb1 are filter coefficient update amounts calculated according to the filter coefficient update algorithm. [0013] According to a third aspect of the present invention, there is provided a howling suppression apparatus for suppressing a howling frequency using an adaptive notch filter section having an adaptive notch filter and a filter coefficient control section, the adaptive notch filter according to the first aspect of the present invention. A notch filter is applied. [0014] According to a fourth aspect of the present invention, a computer mounted on a howling suppressor for suppressing a howling frequency using an adaptive notch filter unit having an adaptive notch filter and a filter coefficient control unit functions as the above-mentioned adaptive notch filter unit. The adaptive notch filter program according to the second aspect of the present invention is applied as a howling suppression program having a program portion, which functions as an adaptive notch filter in the adaptive notch filter unit. [0015] According to the present invention, it is possible to provide an adaptive notch filter and an 15-04-2019 4 adaptive notch filter program capable of removing a frequency to be removed whose bandwidth is wide without presetting a removal bandwidth. [0016] Further, according to the present invention, it is possible to provide a howling suppression apparatus and a howling suppression program capable of suppressing the howling frequency having a wide bandwidth without setting the suppression bandwidth in advance. [0017] It is explanatory drawing which shows the zero and pole of the transfer function of the adaptive notch filter of 1st Embodiment, and the conventional notch filter. It is a block diagram which shows the detailed structure of the adaptive notch filter (multistage adaptive notch filter) of 2nd Embodiment. It is a block diagram which shows the structure of the howling suppression apparatus of 1st Embodiment. It is a block diagram which shows the detailed structure of the filter coefficient control part in the howling suppression apparatus of 1st Embodiment. It is a block diagram which shows the structure of the howling suppression apparatus of 2nd Embodiment. It is a block diagram which shows the detailed structure of the multistage filter coefficient control part in the howling suppression apparatus of 2nd Embodiment. [0018] Before describing the howling suppressor of the first and second embodiments, an embodiment of the adaptive notch filter according to the present invention, which is applied to the howling suppressor of these embodiments, will be described. First, the adaptive notch filter of the first embodiment will be described, and then the adaptive notch filter (multistage adaptive notch 15-04-2019 5 filter) of the second embodiment will be described. [0019] (A) Adaptive Notch Filter of First Embodiment The transfer function H Notch (z) of a conventional general notch filter is expressed by equation (1). In equation (1), f is a removal frequency, fs is a sampling frequency, and r is a parameter for determining a removal bandwidth. The removal frequency f corresponds to the howling frequency if the adaptive notch filter is applied to the howling suppressor. [0020] The transfer function of the adaptive notch filter according to the first and second embodiments is newly proposed by the inventor in consideration of the transfer function H (z) of the equation (1). [0021] The notch filter having the transfer function H Notch (z) shown in the equation (1) is understood from the configuration of the equation (1) that it is a filter of IIR (infinite impulse response) type. On the other hand, the transfer function HIIR (z) of a general IIR adaptive filter is expressed by equation (2). Reference 1 "http://www.lss.uni-stuttgart.de/matlab/notch/index.en.html" or Reference 2 "http://triplecorrelation.com/courses/fundsp/iirdesign.pdf" (2) Formula is disclosed. In the equation (2), n is the time, a1 (n), a2 (n), b0 (n), b1 (n), b2 (n) are filter coefficients of the IIR adaptive filter. [0022] Now, comparing the coefficients of the transfer function H Notch (z) shown in equation (1) with the coefficients of the transfer function HIIR (z) shown in equation (2), equations (3) to (7) are obtained, 3) Expression (8) is obtained for a1 (n) from Expressions (4) and (6). [0023] 15-04-2019 6 From these equations, the numerator constant term of the transfer function HIIR (z) of the IIR adaptive filter shown in the equation (2) and the coefficients of the z <−2> term (ie, b0 (n) and b2 (n)) 1 and adaptively update the coefficients related to the removal frequency and removal bandwidth (ie, b1 (n) and a2 (n)), and update a1 (n) with b1 (n) and a2 (n) (2), it can be found that the transfer function H.sub.llR (z) of the IIR adaptive filter shown in the equation (2) can be made the transfer function of the notch filter. The notch filter in this case is an adaptive notch filter, and the transfer function Hf (z) of this adaptive notch filter is expressed by equation (9). In equation (9), n is time, a 2 (n) and b 1 (n) are filter coefficients of the adaptive notch filter, and these filter coefficients are respectively expressed by equations (10) and (11). In equations (10) and (11), μ is a step gain, and Δa 2 and Δb 1 are filter coefficient update amounts. [0024] The filter coefficient update amounts Δa 2 and Δb 1 can be obtained using a filter coefficient update algorithm (for example, a minimum square error algorithm, a simple hyperstable adaptive recursive filter (SHARF) algorithm, etc.). Equations (12) and (13) show cases where the filter coefficient update amounts Δa 2 and Δb 1 are obtained using the SHARF algorithm. In equations (12) and (13), x (n) is an input signal to the adaptive notch filter at time n, and y (n) is an output signal from the adaptive notch filter at time n . [0025] The adaptive notch filter (see FIG. 3 described later) according to the first embodiment has coefficients a2 (n) and b1 (n (n) relating to the removal bandwidth and removal frequency adaptively updated according to equations (10) and (11). (9) to which Hf (z) is applied. [0026] The filtering process to which the transfer function Hf (z) shown in the equation (9) is applied is realized, for example, by the CPU executing a program for executing the calculation of the transfer function Hf (z) shown in the equation (9) You can also [0027] According to the adaptive notch filter of the first embodiment, it is possible to provide an 15-04-2019 7 adaptive notch filter that can remove (attenuate) the removal frequency even if the removal bandwidth is not a predetermined width. [0028] Further, according to the adaptive notch filter of the first embodiment, it is possible to provide an adaptive notch filter having a zero point and a pole in the same place as the conventional notch filter having the transfer function shown in equation (1). The filter can be constructed as a stable filter. The reason for achieving this effect will be described below. [0029] When the zero point of the transfer function Hf (z) shown in the equation (9) is confirmed, it is indicated that the numerator is 0 because the zero point of the transfer function Hf (z) is Hf (z) = 0 (14 ) Equation is obtained. By solving the equation shown in the equation (14), the equation (15) is obtained. Based on the equation (15), the norm | zzero | of zzero is obtained to obtain the equation (16). [0030] In the equation (16), | zzero | = 1 indicates that the zero point is on the unit circle of the z plane. In addition, since z conversion on the unit circle in the z plane represents frequency characteristics, by setting the constant term of the numerator and the coefficient of the z <-2> term to 1, zeros on the unit circumference of the z plane It can be seen that the frequency characteristic of the transfer function Hf (z) shown in equation (9) can be notched at a certain frequency. 15-04-2019 8 [0031] Furthermore, the pole of the transfer function Hf (z) indicates that the denominator is 0 from the point where the denominator of Hf (z) is 0, and equation (17) is obtained. By solving the equation shown in equation (17), equation (18) is obtained. [0032] Equation (18) indicates that the pole is located at √ {a2 (n)} times the zero point zzero. Here, FIG. 1 shows the zeros and poles of the equations (1) and (9) on the z plane, and the transfer functions shown in the equations (1) and (9) are positions of several times the zeros. It can be seen that a pole exists in That is, the adaptive notch filter according to the first embodiment is an adaptive notch filter having poles on the straight line of the origin and the zero point of the z plane as in the conventional notch filter having the transfer function shown in equation (1). [0033] Also, since the poles of the transfer function are stable if they exist in the unit circle, according to FIG. 1, by selecting a2 (n) to have a value within the range of 0 <a2 (n) <1, It can be seen that the adaptive notch filter of the first embodiment becomes a stable filter. [0034] (B) Adaptive Notch Filter of Second Embodiment Next, a second embodiment of the adaptive notch filter according to the present invention will be described. The adaptive notch filter of the second embodiment is a filter (multistage adaptive notch filter) in which the adaptive notch filters of the first embodiment described above are cascaded in multiple stages. [0035] FIG. 2 is a block diagram showing a detailed configuration of the adaptive notch filter (multistage adaptive notch filter) of the second embodiment. 15-04-2019 9 [0036] The adaptive notch filter (multi-stage adaptive notch filter) 10 according to the second embodiment has an adaptive notch filter having a transfer function shown in equation (9), as shown in FIG. , Adaptive notch filters 11-1 to 11-j (j is an integer of 2 or more) having different removal frequencies are cascaded in multiple stages to form one notch filter, and the filters of each adaptive notch filter 11-1 to 11-j By updating the coefficients, it is constructed as one adaptive notch filter. [0037] Each of the adaptive notch filters 11-1 to 11-j may be configured as hardware and connected in cascade. Further, multistage connection of each of the adaptive notch filters 11-1 to 11-j can also be performed by software. In this case, each of the adaptive notch filters 11-1 to 11-j may be described as different processing routines (programs), and the processing of each processing routine may be sequentially executed. The multistage connection of the adaptive notch filters 11-1 to 11-j may be realized by describing one processing routine in which the filters 11-1 to 11-j are merged. The latter case will be described below. [0038] The multistage connection of the adaptive notch filters 11-1 to 11-j is realized as a transfer characteristic in terms of integrated software by multiplying the transfer functions of the respective adaptive notch filters 11-1, ..., 11-j. can do. By multiplying the transfer function, the multistage adaptive notch filter 10 becomes a long notch filter, and when the number of adaptive notch filters is larger than the number of frequencies to be removed, the notch is a sharp notch filter. The coefficients converge. [0039] The multistage adaptive notch filter 10 according to the second embodiment may not necessarily 15-04-2019 10 be configured by connecting the adaptive notch filters 11-1 to 11-j in multiple stages. The point is that the adaptive notch filters 11-1 to 11- It is sufficient to have the same transfer function as in the case of connecting j in multiple stages. [0040] The relationship between the input and output X (z) and Y (z) of the multistage adaptive notch filter 10 shown in FIG. 2 can be expressed by using the transfer function of each of the adaptive notch filters 11-1,. As shown in the equation. In equation (19), H k (z) (k = 1, 2, ..., j) is the transfer function of each adaptive notch filter 11-1, ..., 11-j, and X (z) is a multistage adaptive notch filter The z conversion of the input signal of 10, Y (z) is the z conversion of the output signal of the multistage adaptive notch filter 10, and j is the number of adaptive notch filters constituting the multistage adaptive notch filter 10. For example, when j = 4, the multistage adaptive notch filter 10 is a filter that removes up to four different frequencies. [0041] The transfer function of the multistage adaptive notch filter 10 in which the adaptive notch filters 11-1, ..., 11-j are connected in multiple stages to form one adaptive notch filter is the transfer function of the equation (9) at the corresponding point of the equation (19). By substituting and organizing, it can be expressed by equation (20). A0 in equation (20) is expressed by equation (21). Formula (20) is calculated | required that the removal bandwidth of each adaptive notch filter 11 -1, ..., 11-j is the same. In the equations (20) and (21), n is the time, a2 (n), b1 (k, n) (k = 1, 2,..., J) is each adaptive notch filter 11-1,. Filter coefficients a (n), bi (n) (i = 1, 2,..., 2j-1) are filter coefficients of the multistage adaptive notch filter 10 for removing a plurality of frequencies, and updating of these filter coefficients is They are respectively executed according to the equations (22) and (23). In equations (22) and (23), μ is a step gain, and Δa, Δbi (i = 1, 2,..., 2j−1) are filter coefficient update amounts. [0042] The filter coefficient update amounts Δa, Δbi (i = 1, 2,..., 2j−1) can be obtained using a filter 15-04-2019 11 coefficient update algorithm. Equations (24) and (25) show cases where the filter coefficient update amounts Δa and Δbi are obtained according to the SHARF algorithm. In the equations (24) and (25), x (n) is an input signal to the multistage adaptive notch filter 10, and y (n) is an output signal from the multistage adaptive notch filter 10. [0043] From equation (20) described above, it can be seen that the constant term of the denominator, the constant term of the numerator, and the filter coefficient of the highest order are 1, respectively. From this, the multistage adaptive notch filter 10 for removing multiple frequencies sets the constant term of the denominator and the constant term of the numerator and the highest order filter coefficient of z to 1, and the highest order of the denominator and the constant term of the numerator and z It can be seen that an adaptive notch filter that removes multiple frequencies can be obtained by updating filter coefficients other than the highest order and obtaining constant terms of the denominator and filter coefficients other than the highest order from the updated filter coefficients of the numerator. [0044] After substituting the equations (20) and (21) into the equation (19), the equation is converted from the z plane to the time axis to obtain the equation (26). In equation (26), n is time, x (n) is an input signal to the multistage adaptive notch filter 10, and y (n) is an output signal from the multistage adaptive notch filter 10. [0045] When the multistage adaptive notch filter 10 has the transfer function shown in equation (20), it is possible to remove a plurality of frequencies having a bandwidth. When the multistage adaptive notch filter 10 is applied to the howling suppressor, the filter coefficient of the multistage adaptive notch filter 10 is updated so as to suppress the adaptive howling frequency in an environment in which a plurality of frequencies having bandwidth have howling. And howling can be properly suppressed. [0046] 15-04-2019 12 (C) Howling Suppression Device of First Embodiment Next, a first embodiment of the howling suppression device according to the present invention will be described with reference to the drawings. [0047] The howling suppression apparatus according to the first embodiment implements the adaptive notch filter according to the above-described first embodiment, performs adaptive notch filter processing on an input signal, and suppresses the frequency at which the howling is performed. It is a thing. [0048] (C-1) Configuration of Howling Suppression Device of First Embodiment FIG. 3 is a block diagram showing a configuration of the howling suppression device of the first embodiment. The units that process digital signals may be configured with a CPU and a program executed by the CPU, but even in this case, they can be functionally represented in FIG. [0049] In FIG. 3, the howling suppression apparatus 100 according to the first embodiment converts an analog sound signal input to the sound signal input terminal 101 to which an analog sound signal is input and the sound signal input terminal 101 into a digital sound signal. Converting digital sound signal output from the adaptive notch filter unit 103 into an analog sound signal, the converter 102, an adaptive notch filter unit 103 that performs processing of an adaptive notch filter with the digital sound signal from the AD converter 102 as an input A DA converter 108 and a sound signal output terminal 109 for outputting an analog sound signal are provided. [0050] The adaptive notch filter unit 103 corresponds to the adaptive notch filter according to the first embodiment described above. The adaptive notch filter unit 103 can hold the digital sound signal output from the AD converter 15-04-2019 13 102 as the input signal, the filter processing unit 104 that performs processing of the adaptive notch filter, and the input signal to the filter processing unit 104 for the filter length. The filter coefficient is calculated from the input register unit 105, the output register unit 106 capable of holding the output signal from the filter processing unit 104 for the filter length, and the input / output signals held in the input register unit 105 and the output register unit 106. And a filter coefficient control unit 107. [0051] FIG. 4 is a block diagram showing the detailed configuration of the filter coefficient control unit 107. As shown in FIG. The filter coefficient control unit 107 updates the filter coefficient from the input / output signal held in the input register unit 105 and the output register unit 106 using the filter coefficient update algorithm such as the equation (12) or (13) described above. Using the filter coefficient update amount calculated by the filter coefficient update amount calculation unit 201 for calculating the amounts Δa 2 and Δb 1 and the filter coefficient update amount calculation unit 201, the filter coefficient a 2 (n) according to the equations (10) and (11) , B1 (n), and whether a2 (n + 1) of one of the updated filter coefficients is larger than 0 and smaller than 1 or not, and a value of 0 or less or In the case of a value of 1 or more, filter coefficients a2 (n + 1) and b1 (n + 1) are output while performing correction so as to be a value larger than 0 and smaller than 1. A filter coefficient correction unit 203, a filter coefficient register unit 204 that holds the filter coefficients a2 (n + 1) and b1 (n + 1) output from the filter coefficient correction unit 203 and supplies the filter coefficient update unit 202; The filter coefficient calculation unit 205 calculates b1 (n) ((a2 (n)) of the equation (9) from the coefficients and outputs all filter coefficients. [0052] (C-2) Operation of the howling suppression apparatus of the first embodiment Next, the operation of the howling suppression apparatus 100 of the first embodiment will be described. [0053] Analog sound such as a microphone input signal amplified by a microphone amplifier, a signal obtained by amplifying and adding a plurality of microphone input signals by an audio mixer etc., 15-04-2019 14 or a signal from which an echo is removed by an echo canceler, etc. A signal is input. The analog sound signal input to the sound signal input terminal 101 is folded in the AD converter 102 by passing it through a low pass filter (built in the AD converter 102) whose passband is set according to the sampling frequency. After the frequency of the high frequency component is cut so as not to occur, the analog sound signal is converted into a digital sound signal, and the digitized digital sound signal is output from the AD converter 102. The digital sound signal is input to the filter processing unit 104 and the input register unit 105 of the adaptive notch filter unit 103. [0054] The filter processing unit 104 receives the input signal of the past time held in the input register unit 105 (for example, refer to x (n−1) in the equation (13)) and the past held in the output register unit 106. An output signal at time (for example, y (n) of equation (13) or (12) or y (n-2) of equation (12)) and a filter coefficient output from the filter coefficient control unit 107 For example, the transfer function Hf (z) of the equation (9) is applied using a2 (n + 1) of the equation (10) and b1 (n + 1) of the equation (11), and a new function from the filter processing unit 104 is applied. The output signal (y (n + 1)) is obtained by applying the notation of equation (13). The determined new output signal is provided to the output register unit 106 and the DA converter 108. [0055] Here, when one sample of the new digital sound signal output from the AD conversion unit 102 is input, the input register unit 105 receives one sample of the digital sound signal stored most in the end of the input register unit 105. After deleting and shifting the value of the input register unit 105, a new one-sample digital sound signal is stored at the top of the input register unit 105. [0056] Further, when the output signal of the filter processing unit 104 is output, the output register unit 106 deletes the output signal of one sample stored at the end of the output register unit 106 and the value of the output register unit 106. , And store the output signal just obtained at the top of the output register unit 106. 15-04-2019 15 [0057] Next, the operation of the filter coefficient control unit 107 will be described. [0058] The filter coefficient update amount calculation unit 201 calculates update amounts Δa2 and Δb1 of the filter coefficient from the holding signals of the input register unit 105 and the output register unit 106 according to equations (12) and (13). [0059] The filter coefficient update unit 202 outputs the filter coefficient update amount calculation unit 201 with the constant term of the denominator of the transfer function, the constant term of the numerator, and the second-order coefficient of the numerator being 1 as shown in equation (9). Using the coefficient update amounts Δa 2 and Δb 1 and the filter coefficients a 2 (n) and b 1 (n) held in the filter coefficient register unit 204 and the fixed step gain μ according to the equations (10) and (11) The filter coefficients a 2 (n + 1) and b 1 (n + 1) are updated, and the updated filter coefficients are output to the filter coefficient correction unit 203. [0060] The filter coefficient correction unit 203 determines whether a2 (n + 1) out of the updated filter coefficients a2 (n + 1) and b1 (n + 1) is a value larger than 0 and smaller than 1. When the value is smaller, the filter coefficients a2 (n + 1) and b1 (n + 1) are output to the filter coefficient register unit 204 and the filter coefficient calculation unit 205 without correction. If the updated filter coefficient a2 (n + 1) is a value of 0 or less or 1 or more, it is corrected to be a value larger than 0 and smaller than 1 (for example, a value after correction is determined in advance), The filter coefficients a2 (n + 1) and b1 (n + 1) are output to the filter coefficient register unit 204 and the filter coefficient calculation unit 205. [0061] The filter coefficient register unit 204 holds the filter coefficients a2 (n + 1) and b1 (n + 1) output from the filter coefficient correction unit 203. [0062] 15-04-2019 16 The filter coefficient calculation unit 205 uses the corrected filter coefficients a2 (n + 1) and b1 (n + 1) to obtain b1 (n) √ (a2 (n)) in equation (9), and filters all filter coefficients. Output to unit 104. [0063] The DA converter 108 converts a digital sound signal, which is an output signal of the filter processing unit 104 of the adaptive notch filter unit 103, into an analog sound signal, and then incorporates a low pass filter (built in by the DA converter 108 according to the sampling frequency). To cut the frequency of the high-frequency component so that aliasing does not occur, give it to the sound signal output terminal 109 as an analog sound signal, and output the analog sound signal from the sound signal output terminal 109 to the speaker amplifier or communication path Be done. [0064] Next, an operation from when the operation of the howling suppression apparatus 100 according to the first embodiment starts to when the howling occurs will be described. [0065] When the howling suppression apparatus 100 starts operation, the analog sound signal passes through the sound signal input terminal 101, is converted into a digital sound signal by the AD converter 102, and is input to the adaptive notch filter unit 103. [0066] If the howling does not occur, the filter coefficients of the filter processing unit 104 do not converge, so that the digital sound signal is output to the DA converter 108 without the specific frequency being suppressed by the filter processing unit 104. In the DA converter 108, the digital sound signal output from the adaptive notch filter unit 103 is converted into an analog sound signal, and is output through the sound signal output terminal 109. [0067] 15-04-2019 17 Since the filter coefficient of the adaptive notch filter in the filter processing unit 104 has not yet converged to the filter coefficient that suppresses the generated frequency immediately after the single frequency starts to howl, the filter processing unit 104 performed the howl The digital sound signal is output to the DA converter 108 without suppressing the signal. Since the DA converter 108 converts the digital sound signal output from the adaptive notch filter unit 103 into an analog sound signal and outputs it via the sound signal output terminal 109, howling is suppressed when the howling starts to occur No analog sound signal is output. [0068] The filter coefficient control unit 107 calculates the update amount of the filter coefficient in the filter coefficient update amount calculation unit 201 based on the filter coefficient update algorithm using the holding signal of the input register unit 105 and the output register unit 106, and updates the filter coefficient. Output to the unit 202. The filter coefficient update unit 202 outputs the filter coefficient update amount calculation unit 201 while keeping the constant term of the denominator of the transfer function, the constant term of the numerator, and the filter coefficient of the highest order of the numerator as 1, shown in equation (9). The filter coefficient updated in the filter coefficient correction unit 203 is updated using the filter coefficient update amount, the filter coefficient of the filter coefficient register unit 204, and the fixed step gain to update the filter coefficients of the equations (10) and (11). Output The filter coefficient correction unit 203 determines whether one of the filter coefficients a2 (n + 1) is a value larger than 0 and smaller than 1. If the value is larger than 0 and smaller than 1, b1 (n + 1) is not corrected. Output to the filter coefficient register unit 204 and the filter coefficient calculation unit 205. If the updated filter coefficient a2 (n + 1) is a value of 0 or less or 1 or more, it is corrected to be a value larger than 0 and smaller than 1 and the filter coefficient register unit 204 and the filter coefficient together with b1 (n + 1) It is output to the calculation unit 205. 15-04-2019 18 The filter coefficient calculation unit 205 obtains b1 (n) √ (a2 (n)) of the equation (9) using the filter coefficient output from the filter coefficient correction unit, and outputs all the filter coefficients to the filter processing unit 104 Do. [0069] As the update proceeds, the filter coefficients of the adaptive notch filter in the filter processing unit 104 converge on the filter coefficients that suppress the howling frequency, and as a result, the filter processing unit 104 performs DA conversion on the digital sound signal whose howling is suppressed. Output to the signal generator 108. In the D / A converter 108, the digital sound signal whose howling is suppressed output from the adaptive notch filter unit 103 is converted into an analog sound signal, and is output through the sound signal output terminal 109. Since the output signal from the adaptive notch filter unit 103 becomes almost 0 (howling is suppressed to 0), the update amount calculated by the filter coefficient update amount calculation unit 201 of the filter coefficient control unit 107 becomes almost 0. The filter coefficient updating unit 202 converges without updating the filter coefficient. Thereby, the convergence state of the filter coefficient is continued, and howling continues to be suppressed. [0070] When different frequencies from the above state begin to howl, the adaptive notch filter in the filter processing unit 104 can suppress the howled signal because the filter coefficients are not converged so as to suppress the different frequencies that have started to howl. The digital sound signal is output to the D / A converter 108 instead. In the DA converter 108, since the digital sound signal output from the adaptive notch filter unit 103 is converted to an analog sound signal and output through the sound signal output terminal 109, when howling of different frequencies starts to occur, A howling analog sound signal is output. 15-04-2019 19 [0071] The filter coefficient control unit 107 calculates the update amount of the filter coefficient in the filter coefficient update amount calculation unit 201 based on the filter coefficient update algorithm using the holding signal of the input register unit 105 and the output register unit 106, and updates the filter coefficient. Output to the unit 202. The filter coefficient update unit 202 outputs the filter coefficient update amount calculation unit 201 with the constant term of the denominator of the transfer function, the constant term of the numerator, and the highest order filter coefficient of the numerator being 1 as shown in equation (9). The filter coefficients of the equations (10) and (11) are updated using the coefficient update amount, the filter coefficient of the filter coefficient register unit 204 and the fixed step gain, and the filter coefficient updated in the filter coefficient correction unit 203 Output The filter coefficient correction unit 203 determines whether one of the filter coefficients a2 (n + 1) is a value larger than 0 and smaller than 1. If the value is larger than 0 and smaller than 1, b1 (n + 1) is not corrected. Output to the filter coefficient register unit 204 and the filter processing unit 104. If the updated filter coefficient a2 (n + 1) is a value of 0 or less or 1 or more, it is corrected to be a value larger than 0 and smaller than 1 and the filter coefficient register unit 204 and the filter coefficient together with b1 (n + 1) It is output to the calculation unit 205. The filter coefficient calculation unit 205 obtains b1 (n) √ (a2 (n)) of the equation (9) using the filter coefficient output from the filter coefficient correction unit, and outputs all the filter coefficients to the filter processing unit 104 Do. [0072] As the update proceeds, the filter coefficients of the adaptive notch filter in the filter processing unit 104 converge on the filter coefficients that suppress different frequencies again, so the signal with the howling suppressed from the adaptive notch filter unit 103 is sent to the DA converter 108. It is output. In the DA converter 108, the digital sound signal whose howling is suppressed, which is an output signal from the adaptive notch filter unit 103, is converted into an analog sound signal, and is output through the sound signal output terminal 109. The filter coefficient control unit 107 calculates the filter coefficient update amount calculation unit 201 of the filter coefficient control unit 107 because the output signal from the adaptive notch filter unit 103 becomes almost 0 (howling of different frequencies is suppressed to 0). The updated amount becomes almost zero, the filter coefficient updating unit 202 does not update the filter coefficient, and the filter coefficient output from the filter coefficient control unit 107 converges. In this way, the filter coefficients converge and continue to suppress the frequency even for different frequencies. 15-04-2019 20 [0073] When a frequency having a bandwidth starts to feedback (not only a certain frequency but also frequency components before and after the frequency), the adaptive notch filter in the filter processing unit 104 converges on the filter coefficient that suppresses the bandwidth. Since it is not possible to suppress the howled signal, the digital sound signal howled to the DA converter 108 is output. In the DA converter 108, since the digital sound signal output from the adaptive notch filter unit 103 is converted to an analog sound signal and is output through the sound signal output terminal 109, when the frequency with a bandwidth starts to feedback Is output as an howling analog sound signal. [0074] The filter coefficient control unit 107 calculates the update amount of the filter coefficient in the filter coefficient update amount calculation unit 201 based on the filter coefficient update algorithm using the holding signal of the input register unit 105 and the output register unit 106, and updates the filter coefficient. Output to the unit 202. The filter coefficient update unit 202 outputs the filter coefficient update amount calculation unit 201 while keeping the constant term of the denominator of the transfer function, the constant term of the numerator, and the filter coefficient of the highest order of the numerator as 1, shown in equation (9). The filter coefficient of the equations (10) and (11) is updated using the filter coefficient update amount, the filter coefficient of the filter coefficient register unit 204, and the fixed step gain, and the filter updated in the filter coefficient correction unit 203 Output coefficients. The filter coefficient correction unit 203 determines whether one of the filter coefficients a2 (n + 1) is a value larger than 0 and smaller than 1. If the value is larger than 0 and smaller than 1, b1 (n + 1) is not corrected. Output to the filter coefficient register unit 204 and the filter coefficient calculation unit 205. If the updated filter coefficient a2 (n + 1) is a value of 0 or less or 1 or more, it is corrected to be a value larger than 0 and smaller than 1 and the filter coefficient register unit 204 and the filter coefficient together with b1 (n + 1) It is output to the calculation unit 205. When a frequency having a bandwidth is howling, even in such an updating process, the output signal is still large and the amount of updating is large, the convergence of the filter coefficient is delayed, and the update of the filter coefficient is continued for a considerable time. The filter coefficient calculation unit 205 obtains b1 (n) フ ィ ル タ (a2 (n)) of the equation (9) using the filter coefficient output from the filter coefficient correction unit 204, and transmits all filter coefficients to the filter processing unit 104. Output. 15-04-2019 21 [0075] Nevertheless, when the update proceeds, the filter coefficients of the adaptive notch filter in the filter processing unit 104 converge so as to suppress a certain frequency with the bandwidth, and the signal whose howling is suppressed from the adaptive notch filter unit 103 is the DA converter 108 Output to In the DA converter 108, the digital sound signal whose howling is suppressed, which is an output signal from the adaptive notch filter unit 103, is converted into an analog sound signal, and is output through the sound signal output terminal 109. Since the filter coefficient control unit 107 has almost zero output signal from the adaptive notch filter unit 103 (howling of frequencies with a bandwidth is suppressed to be 0), the filter coefficient update amount calculation unit of the filter coefficient control unit 107 The update amount calculated in 201 becomes almost 0, the filter coefficient updating unit 202 does not update the filter coefficient, and the filter coefficient output from the filter coefficient control unit 107 converges. In this manner, the filter coefficient converges and continues to suppress the frequency even for the frequency with the bandwidth. [0076] (C-3) Effects of the howling suppressor of the first embodiment According to the howling suppressor of the first embodiment, an environment in which a single frequency is howling, an environment in which the howling frequency changes, and a bandwidth in the howling frequency Howling can be suppressed even in an environment having [0077] (D) Howling Suppression Device of Second Embodiment Next, a second embodiment of the howling suppression device according to the present invention will be described with reference to the drawings. [0078] The howling suppression apparatus according to the second embodiment is an implementation of the adaptive notch filter (multi-stage adaptive notch filter) according to the second embodiment described above, which performs adaptive notch filter processing on an input signal, and Even in an environment where frequency is howling, it suppresses multiple frequencies that are howling. [0079] 15-04-2019 22 (D-1) Configuration of the howling suppressor of the second embodiment FIG. 5 is a block diagram showing the configuration of the howling suppressor of the second embodiment, and is a diagram according to the howling suppressor of the first embodiment. The same parts as 3 and the corresponding parts are shown with the same reference numerals. The units that process digital signals may be configured with a CPU and a program executed by the CPU, but even in this case, they can be functionally represented in FIG. [0080] In FIG. 5, the howling suppression apparatus 300 of the second embodiment is an AD that converts an analog sound signal input to the sound signal input terminal 101 to which an analog sound signal is input and the sound signal input terminal 101 into a digital sound signal. From the multistage adaptive notch filter unit 301 which performs processing of an adaptive notch filter (the adaptive notch filter of the second embodiment) with the converter 102, a digital sound signal from the AD converter 102, and from the multistage adaptive notch filter unit 301 A digital-to-analog converter 108 converts the output digital sound signal into an analog sound signal, and a sound signal output terminal 109 to which the analog sound signal is output. [0081] The multistage adaptive notch filter unit 301 takes the digital sound signal output from the AD converter 102 as an input signal and performs processing of the adaptive notch filter shown in equation (26), and an input of the multistage filter processing unit 302 Filter coefficients from the input register unit 105 that can hold signals by the filter length, the output register unit 106 that can hold the output signals of the multistage filter processing unit 302 by the filter length, and the holding signals of the input register unit 105 and the output register unit 106 A multistage filter coefficient control unit 303 that calculates [0082] FIG. 6 is a block diagram showing a detailed configuration of the multistage filter coefficient control unit 303. As shown in FIG. The multistage filter coefficient control unit 303 calculates multistage filter coefficient update amounts from the holding signals of the input register unit 105 and the output register 106 using a multistage filter coefficient update algorithm such as the equations (24) and (25). Multi- 15-04-2019 23 stage filter coefficient update unit that updates filter coefficients according to equations (22) and (23) using the filter coefficient update amount calculated by the filter coefficient update amount calculation unit 401 and the multi-stage filter coefficient update amount calculation unit 401 Check if 402 and one of the updated filter coefficients a (n + 1) is a value larger than 0 and smaller than 1. If the value is 0 or less or 1 or more, 1 or more and 0 or more A multistage filter coefficient correction unit 403 that corrects to a smaller value and then outputs the updated filter coefficient, and a multistage filter The multistage filter coefficient register unit 404 for holding the filter coefficient output from the number correction unit 403 and the updated filter coefficients a0b1 (n), a0 <2> b2 (n), a0 <2j− in equation (20) The multistage filter coefficient calculation unit 405 is configured to obtain 1> b2j-1 (n) and a0 of the equation (21), and output all the filter coefficients. [0083] (D-2) Operation of the Howling Suppression Device of the Second Embodiment Next, the operation of the howling suppression device 300 of the second embodiment will be described. In addition, each part of the code | symbol same as 1st Embodiment performs the same operation | movement in 2nd Embodiment. [0084] The multistage filter processing unit 302 holds the holding signals of the input register unit 105 (for example, x (n), x (n−i), x (n−2j), etc. of equation (26)) and the holding of the output register unit 106. A signal (for example, y (n−i), y (n−2j) in equation (26), etc.) and a filter coefficient output from multistage filter coefficient control section 303 (for example, bi (n) in equation (26) The output signal (y (n) of the equation (26)) from the multistage filter processing unit 302 is determined using a (n), a0 (n), and the like. Also, the output signal y (n) thus obtained is given to the output register unit 106 and the DA converter 108. [0085] Next, each part of the multistage filter coefficient control unit 303 will be described. 15-04-2019 24 [0086] The multistage filter coefficient update amount calculation unit 401 calculates update amounts Δa and Δbi of multistage filter coefficients from the holding signals of the input register unit 105 and the output register unit 106. [0087] The multistage filter coefficient update unit 402 outputs the multistage filter coefficient update amount calculation unit 401 while keeping the constant term of the denominator of the transfer function, the constant term of the numerator, and the highest order coefficient of the numerator as 1 shown in equation (20). Using the updated filter coefficient update amount, the filter coefficient of the multistage filter coefficient register unit 404, and the predetermined step gain, according to the equations (22) and (23), the other filter coefficients a (n + 1), bi (N + 1) is updated, and the updated filter coefficient is output to the multistage filter coefficient correction unit 403. [0088] The multistage filter coefficient correction unit 403 determines whether the updated filter coefficient a (n + 1) is a value greater than 0 and less than 1. If the value is greater than 0 and less than 1, the correction is not performed. The multistage filter coefficient register unit 404 and the multistage filter coefficient calculation unit 405 are output together with the other filter coefficients. When the updated filter coefficient a (n + 1) is a value of 0 or less or 1 or more, the multistage filter coefficient correction unit 403 corrects the value to a value larger than 0 and smaller than 1 (for example, after correction in advance) And the updated other filter coefficients are output to the multistage filter coefficient register unit 404 and the multistage filter coefficient calculation unit 405. [0089] The multistage filter coefficient register unit 404 holds the filter coefficient output from the multistage filter coefficient correction unit 403. 15-04-2019 25 [0090] The multistage filter coefficient calculation unit 405 uses the updated filter coefficients output from the multistage filter coefficient correction unit 403 and uses a0b1 (n), a0 <2> b2 (n), a0 <2j− in equation (20). 1> b 2 j −1 (n) and a 0 in equation (21) are calculated, and are output to the multistage filter processing unit 302. [0091] In the howling suppression apparatus 300 of the second embodiment, an operation from the start of operation to the time when howling occurs, an operation when a signal including a single frequency howling is input, an operation when the howling frequency changes, The operation when a frequency having a bandwidth is input is similar to the operation of the howling suppression apparatus 100 according to the first embodiment. [0092] When a plurality of frequencies begin to feedback, the multistage adaptive notch filter in the multistage filter processing unit 302 does not converge to the filter coefficients that suppress the plurality of frequencies, so the feedback signal can not be suppressed. A howling digital sound signal is output. In the DA converter 108, the digital sound signal output from the multistage adaptive notch filter unit 301 is converted to an analog sound signal and is output via the sound signal output terminal 109, so howling of a plurality of frequencies starts to occur. At the same time, the howling analog sound signal is output. [0093] The multistage filter coefficient control unit 303 calculates the update amount of the filter coefficient in the multistage filter coefficient update amount calculation unit 401 based on the filter coefficient update algorithm using the holding signals of the input register unit 105 and the output register unit 106. It is output to the filter coefficient update unit 402. The multistage filter coefficient update unit 402 outputs the multistage filter coefficient update amount calculation unit 401 while keeping the constant term of the denominator of the transfer 15-04-2019 26 function, the constant term of the numerator, and the coefficient of the highest order of the numerator as 1 shown in equation (20). The filter coefficients of the equations (22) and (23) are updated using the filter coefficient update amount, the filter coefficient of the multistage filter coefficient register unit 404, and the fixed step gain, and the multistage filter coefficient correction unit 403 is updated. Output the filtered filter coefficients. The multistage filter coefficient correction unit 403 determines whether one of the updated filter coefficients, a (n + 1), is a value greater than 0 and less than 1. If the value is greater than 0 and less than 1 Is output to the multistage filter coefficient register unit 404 and the multistage filter coefficient calculation unit 405 together with the updated other filter coefficients. The multistage filter coefficient correction unit 403 corrects the filter coefficient a (n + 1) to be a value larger than 0 and smaller than 1 when the filter coefficient a (n + 1) is a value of 0 or less, or 1 or more, along with other updated filter coefficients. The multistage filter coefficient register unit 404 and the multistage filter coefficient calculation unit 405 are output. When the time has not passed so much from the start of the update, the output signal from the multistage adaptive notch filter unit 301 is still large, and as a result, the update amount also becomes large, and the filter coefficients continue to be updated without convergence. The multistage filter coefficient calculation unit 405 obtains other filter coefficients from the updated filter coefficients, and outputs all the filter coefficients to the multistage filter processing unit 302. [0094] When the update progresses, the filter coefficients of the adaptive notch filter in the multistage filter processing unit 302 converge so as to suppress a plurality of frequencies, and the multistage adaptive notch filter unit 301 outputs a signal whose howling is suppressed to the DA converter 108. Be done. In the DA converter 108, the digital sound signal output from the multistage adaptive notch filter unit 301 is converted into an analog sound signal, and an analog sound signal whose howling is suppressed is output via the sound signal output terminal 109. In the multistage filter coefficient control unit 303, the output signal from the multistage adaptive notch filter unit 301 becomes almost 0, so the update amount calculated by the multistage filter coefficient update amount calculation unit 401 becomes almost 0, and the multistage filter 15-04-2019 27 coefficient updating unit At 402, the filter coefficients converge without being updated. Thereby, the convergence state of the filter coefficient is continued, and howling continues to be suppressed. [0095] The multistage adaptive notch filter in the multistage filter processing unit 302 suppresses a plurality of frequencies when a frequency having a bandwidth at a plurality of frequencies starts to feedback (also feedback not only a certain frequency but also frequency components before and after the frequency) Since the feedback signal is not converged, the digital sound signal that has been howled to the D / A converter 108 can not be suppressed. In the DA converter 108, the digital sound signal output from the multistage adaptive notch filter unit 301 is converted to an analog sound signal and output through the sound signal output terminal 109, so that the frequency having a bandwidth in frequency is howl When started, the howling analog sound signal is output. [0096] The multistage filter coefficient control unit 303 calculates the update amount of the filter coefficient in the multistage filter coefficient update amount calculation unit 401 based on the filter coefficient update algorithm using the holding signals of the input register unit 105 and the output register unit 106. It is output to the filter coefficient update unit 402. The multistage filter coefficient update unit 402 outputs the multistage filter coefficient update amount calculation unit 401 while keeping the constant term of the denominator of the transfer function, the constant term of the numerator, and the coefficient of the highest order of the numerator as 1 shown in equation (20). The filter coefficients of the equations (22) and (23) are updated using the filter coefficient update amount, the filter coefficient of the multistage filter coefficient register unit 404, and the fixed step gain, and the multistage filter coefficient correction unit 403 is updated. Output the filtered filter coefficients. The multistage filter coefficient correction unit 403 determines whether one of the updated filter coefficients, a (n + 1), is a value greater than 0 and less than 1. If the value is greater than 0 and less than 1 Is output to the multistage filter coefficient register unit 404 and the multistage filter coefficient calculation unit 405 together with the updated other filter coefficients. The multistage filter coefficient correction unit 403 corrects the filter coefficient a (n + 1) to be a value larger than 0 and smaller than 1 when the filter coefficient a (n + 1) is a value of 0 or less, or 1 or more, along with other updated filter coefficients. The multistage filter coefficient register unit 404 and the multistage filter coefficient calculation unit 405 are output. When the time has not passed so much from the start of the 15-04-2019 28 update, the output signal from the multistage adaptive notch filter unit 301 is still large, and as a result, the update amount also becomes large, and the filter coefficients continue to be updated without convergence. The multistage filter coefficient calculation unit 405 obtains other filter coefficients from the updated filter coefficients, and outputs all the filter coefficients to the multistage filter processing unit 302. When a frequency having a bandwidth at multiple frequencies is feedback, even with such an updating process, the output signal is still large and the amount of updating is large, the convergence of the filter coefficient is delayed, and the update of the filter coefficient is considerable The time will be continued. [0097] Nevertheless, when the update proceeds, the filter coefficients of the adaptive notch filter in the multistage filter processing unit 302 converge so as to suppress a plurality of frequencies, and the signal whose howling is suppressed from the multistage adaptive notch filter unit 301 is the DA converter 108 Output to In the DA converter 108, the digital sound signal output from the multistage adaptive notch filter unit 301 is converted into an analog sound signal, and an analog sound signal whose howling is suppressed is output via the sound signal output terminal 109. In the multistage filter coefficient control unit 303, the output signal from the multistage adaptive notch filter unit 301 becomes almost 0, so the update amount calculated by the multistage filter coefficient update amount calculation unit 401 becomes almost 0, and the multistage filter coefficient updating unit At 402, the filter coefficients converge without being updated. Thereby, the convergence state of the filter coefficient is continued, and howling continues to be suppressed. [0098] (D-3) Effects of the howling suppression apparatus of the second embodiment According to the howling suppression apparatus of the second embodiment, an environment where a plurality of frequencies are howling, an environment where one or more of a plurality of howling frequencies change, Howling can be suppressed even in an environment where one or more of the plurality of howling frequencies have a bandwidth. [0099] (E) Other Embodiments In the howling suppressor of the first and second embodiments, the adaptive notch filter of the first or second embodiment is applied to the howling suppressor. Or the application apparatus of the adaptive notch filter of 2nd Embodiment is not limited to the howling suppression apparatus. 15-04-2019 29 [0100] DESCRIPTION OF SYMBOLS 10 ... Multistage adaptive notch filter, 11-1-11-j ... Adaptive notch filter, 100, 300 ... Howling suppression apparatus, 103 ... Adaptive notch filter part, 104 ... Filter processing part, 105 ... Input register part, 106 ... Output register Unit 107: filter coefficient control unit 201: filter coefficient update amount calculation unit 202: filter coefficient update unit 203: filter coefficient correction unit 204: filter coefficient register unit 205: filter coefficient calculation unit 301: multistage adaptive Notch filter unit 302 multi-stage filter processing unit 303 multi-stage filter coefficient control unit 401 multi-stage filter coefficient update amount calculation unit 402 multi-stage filter coefficient update unit 403 multi-stage filter coefficient correction unit 404 multi-stage filter coefficient Register unit, 405: Multi-stage filter coefficient calculation unit. 15-04-2019 30

1/--страниц