JP5381042B2 - Loop gain estimation device - Google Patents

Description

  The present invention relates to a loop gain estimation apparatus that estimates a closed loop gain.

  Conventionally, various techniques for preventing howling have been proposed in loudspeaking systems for lectures, concerts, and the like. A general howling suppression method is to attenuate a frequency band causing howling with a filter when the occurrence of howling is detected.

Non-Patent Document 1 proposes a howling canceller using synchronous addition of M-sequence noise. The howling canceller of Non-Patent Document 1 emits high-level M-sequence noise in advance and trains the adaptive filter. During operation, it emits M-sequence noise at a weak level and applies the adaptive filter. It will continue to be updated. In addition, the howling canceller of Non-Patent Document 1 inhibits the update of the adaptive filter when the disturbance increases due to the speaker's continued speaking or the instrument continues to ring. Therefore, when the microphone input level is high, the disturbance The update of the adaptive filter is stopped.
Makoto Itami and Mitsutoshi Hatori, “Examination of howling removal in acoustic system”, IEICE Technical Report, 1989, EA89-4

  However, since the attenuation method using a general filter is a method of suppressing the frequency band in which howling has occurred, it cannot be suppressed until after howling has occurred once.

  In lectures and concerts, the position of the microphone often moves, and the closed-loop environment changes every moment as the position of the microphone changes. When the closed loop environment changes and the loop gain changes, the adaptive filter as described in Non-Patent Document 1 cannot keep up with the update of the filter coefficient and cannot suppress howling. Further, as in Non-Patent Document 1, if updating of the adaptive filter is stopped when the microphone input level is high, there is a possibility that howling cannot be suppressed.

  In addition, all of these are merely techniques for suppressing the frequency band that causes howling, and do not estimate the loop gain, and cannot prevent howling from occurring.

  Accordingly, an object of the present invention is to provide a loop gain estimation device capable of estimating a loop gain even when a closed loop environment changes in order to prevent howling from occurring.

  The howling prevention apparatus of the present invention includes an input unit for inputting a speech signal, a noise generation unit for generating pseudo noise, a superimposing unit for superimposing pseudo noise on the speech signal, a correlation calculation unit, and a loop gain estimation unit. The correlation calculation unit obtains a correlation between the input audio signal and the pseudo noise generated by the noise generation unit. The loop gain estimation unit estimates a closed-loop gain from the correlation peak calculated by the correlation calculation unit. For example, the loop gain estimation method is as follows.

  (1) The loop gain estimation unit, when the correlation generation unit first calculates a correlation value of a predetermined level or more after the noise generation unit outputs pseudo noise, is a correlation value in the time zone calculated first. A closed-loop gain is estimated based on the wave component.

  When the pseudo noise loops, the correlation calculation unit calculates a correlation value of a predetermined level or higher. The predetermined level is set according to the level of stationary noise. A correlation value equal to or higher than a predetermined level calculated first after the pseudo noise is output can be regarded as a direct sound (direct wave) in which the pseudo noise is looped. Therefore, when a correlation value of a predetermined level or higher is first calculated, the loop gain is estimated based on the calculated correlation value in the time zone (for example, from the sum of correlation values in this time zone). Alternatively, the loop gain may be estimated based only on the highest level correlation value in the time period. It is possible to estimate the loop gain because the direct wave component has a large influence on howling.

  (2) In addition to the direct wave component, the loop gain estimation unit calculates a correlation value of the direct wave component by the correlation calculation unit, and then converts the correlation value to an indirect wave component that is a correlation value of a predetermined level or more in a predetermined time length thereafter. Based on this, the closed-loop gain is estimated.

  A correlation value equal to or higher than a predetermined level from when the first peak is calculated until a predetermined time length (time until the next direct wave) elapses can be regarded as a reflected wave (indirect wave) such as a wall. The predetermined time length is set according to the time from when the pseudo noise is output until the correlation value of the predetermined level or higher is calculated for the first time. The loop gain estimation unit estimates the closed-loop gain based on the components of the direct wave and the reflected wave (for example, from the sum of these correlation values). In this way, the loop gain estimation unit performs a process of estimating from the feedback component of the direct wave and the reflected component of the reflected wave that affects howling, so that the loop gain can be estimated with higher accuracy.

  (3) The loop gain estimation unit estimates a closed-loop gain from the sum of the peak value of the correlation value of the direct wave component and the peak value of the indirect wave component. In addition, the peak said here means the thing which shows the correlation value of the highest level in a certain time slot | zone among components more than a predetermined level. Since it is often the peak component that affects howling, loop gain estimation may be performed based on the sum of the correlation values of the peak components of the direct wave and the reflected wave.

  Various measures are taken to prevent howling from occurring based on the loop gain estimated as described above. For example, a threshold value is set to the estimated loop gain value, and when the loop gain approaches the threshold value, the gain of the audio signal is suppressed to prevent howling.

  Further, when the loop gain approaches the threshold value, a warning may be issued (LED is turned on, a warning is displayed on the display, etc.). In this case, further warning may be performed while suppressing the gain of the audio signal.

  According to the present invention, since the loop gain can be estimated even when the closed loop environment changes, various countermeasures can be taken before howling occurs.

  FIG. 1A is a block diagram showing the overall configuration of the loudspeaker of this embodiment, and FIG. 1B is a block diagram showing the configuration of a mixer incorporating the loop gain estimation device of the present invention. In the description of this embodiment, unless otherwise specified, all audio signals are digital signals, and the configurations of A / D conversion and D / A conversion are omitted.

  The mixer 1 includes a calculation unit 5 and a pseudo noise superimposing unit 7 for inputting a sound signal picked up by a microphone 11 (sound collecting unit). The pseudo noise superimposing unit 7 superimposes pseudo noise on the audio signal picked up by the microphone 11. The mixer 1 actually has a plurality of input channels and output channels, but in the present embodiment, only one channel system is shown for ease of explanation.

  The audio signal on which the pseudo noise is superimposed by the pseudo noise superimposing unit 7 is amplified by a subsequent amplification system (amplifier) and emitted from the speaker 3. The sound emitted from the speaker 3 returns to the microphone 11 to form a closed loop.

  The mixer 1 estimates the closed-loop gain in the arithmetic unit 5. The mixer 1 can prevent howling by suppressing the gain of the audio signal or giving a warning when the estimated loop gain approaches a predetermined threshold value.

  As shown in FIG. 1B, the mixer 1 includes an LPF 12, an audio signal volume 13, a superposition unit 14, an M-sequence generator 15, an N-times oversampling unit 16, an HPF 17, a pseudo noise volume 18, an HPF 19, and a correlation. A calculation unit 20, a timer 21, a loop gain estimation unit 22, and a gain control unit 23 are provided.

  The calculation unit 5 includes an M-sequence generator 15, an HPF 19, a correlation calculation unit 20, a timer 21, and a loop gain estimation unit 22. The pseudo noise superimposing unit 7 includes an audio signal volume 13, a superimposing unit 14, an M-sequence generator 15, an N-times oversampling unit 16, an HPF 17, a pseudo noise volume 18, and a gain control unit 23.

  The audio signal picked up by the microphone 11 is input to the LPF 12 of the pseudo noise superimposing unit 7 and the HPF 19 of the calculating unit 5. The configuration and function of the pseudo noise superimposing unit 7 will be described with reference to FIG. The lower column of each component shows the waveform of the signal output by each component.

  An audio signal picked up by the microphone 11 is input to the LPF 12 of the pseudo noise superimposing unit 7. Note that the waveform of the signal shown in the lower column of the microphone 11 is an example, and signals having various waveforms are actually input to the LPF 12.

  The LPF 12 cuts the high range from the collected audio signal and outputs it to the audio signal volume 13 (refer to the lower waveform of the LPF 12 in the figure).

  The audio signal volume 13 outputs the signal collected by the microphone 11 to the superimposing unit 14 with the gain set by the gain control unit 23.

  The M-sequence generator 15 corresponds to the noise generation unit of the present invention, periodically generates a signal with high autocorrelation such as a PN code (M-sequence) as pseudo noise, and outputs the signal to the N-times oversampling unit 16 (Refer to the lower waveform of the M-sequence generator 15 in the figure, where the waveform in the lowermost column represents the time axis). In addition, you may use other random numbers, such as not only M series but Gold series.

  Note that the pseudo-noise output period is the time until the component of the reflected wave (indirect wave) decreases to a predetermined level or more (impulse in the acoustic transmission system) so that the loop gain estimation unit 22 described later can perform loop gain estimation processing. Longer than the response convergence time).

  The N-times oversampling unit 16 oversamples the pseudo noise. For example, 16 times oversampling is performed, the code period of each bit of the PN code is expanded, and the pseudo noise length is 16 times (refer to the lower waveform of the N-times oversampling unit 16 in FIG. Represents the time axis). The N-times oversampling unit 16 outputs the oversampled signal to the HPF 17.

  The HPF 17 cuts the low frequency range of the signal input from the N-times oversampling unit 16 (refer to the lower waveform of the HPF 17 in the figure, where the waveform in the lowermost column represents the time axis). The cutoff frequency is set to 10 kHz, for example.

  Note that LPF 12 and HPF 17 are not essential in the present invention. However, since the HPF 17 cuts the sound other than the high frequency range of the pseudo noise, even if the sound is emitted from the speaker 3, there is no sense of incongruity in hearing (noise is difficult to hear). Further, the LPF 12 prevents the high-frequency pseudo noise once input to the microphone from being output to the amplification system again, and the pseudo noise loop phenomenon can be suppressed. When the LPF 12 and the HPF 17 are omitted, the pseudo noise loop phenomenon may be suppressed by subtracting the pseudo noise component from the audio signal picked up by the microphone 11 and then outputting it to the amplification system.

  Note that oversampling by the N-times oversampling unit 16 is not essential in the present invention. However, by performing oversampling, the temporal redundancy of pseudo noise increases, and the accuracy of correlation calculation can be improved. Actually, the presence or absence of oversampling may be set according to the required accuracy and the code length of the pseudo noise.

  The signal output from the HPF 17 is input to the pseudo noise volume 18. The pseudo noise volume 18 outputs the output signal of the HPF 17 to the superposition unit 14 with the gain set by the gain control unit 23. The level of the pseudo noise may be a weak level that does not cause a sense of incongruity, but a level that can detect the peak value of the pseudo noise correlation is secured.

  The superimposing unit 14 superimposes the signal (pseudo noise) output from the HPF 17 on the audio signal output from the audio signal volume 13 and outputs the signal to the amplification system.

  Next, the configuration and function of the calculation unit 5 will be described with reference to FIG. The lower column of each component shows the waveform of the signal output by each component. The M-sequence generator 15 outputs the same pseudo noise as that output to the N-times oversampling unit 16 to the correlation calculation unit 20 (refer to the lower waveform of the M-sequence generator 15 in FIG. Represents the axis). Further, after outputting the pseudo noise, a signal indicating the output timing (timing signal) is transmitted to the timer 21. When receiving the timing signal, the timer 21 starts time counting and transmits a timer signal indicating the count time to the loop gain estimating unit 22. Note that the timer 21 is not essential in the present invention.

  The microphone 11 picks up sound including pseudo noise. An audio signal picked up by the microphone 11 is input to the HPF 19 of the calculation unit 5. The HPF 19 cuts the low frequency from the audio signal picked up by the microphone 11 and outputs it to the correlation calculation unit 20 (refer to the lower waveform of the HPF 19 in the figure, where the waveform represents the frequency axis). The cut-off frequency is determined corresponding to the HPF 17 (for example, 10 kHz).

  The correlation calculation unit 20 obtains the correlation between the pseudo noise input from the M-sequence generator 15 and the output signal of the HPF 19. Since the M-sequence code has a very high autocorrelation, if the output signal of the HPF 19 includes the same M-sequence pseudo noise, the level of the correlation value is as shown in the waveform of FIG. Get higher. The correlation calculation unit 20 outputs the timing (reception timing) at which the high level correlation value is calculated and the correlation value at that time to the loop gain estimation unit 22.

  When receiving the reception, the loop gain estimation unit 22 refers to the timer signal from the timer 21 and obtains a time difference from the timing at which the pseudo noise is output to the reception timing. This time difference corresponds to the delay time of the closed loop. Note that the output of the reception timing of the correlation calculation unit 20 is not essential unless the delay time of the closed loop is measured (when the timer 21 is not provided).

  The loop gain estimation unit 22 performs processing for estimating the loop gain. Although various modes can be considered as a method for estimating the loop gain, for example, it is performed in the following modes.

  First, the first estimation method will be described with reference to FIG. FIG. 4 is a diagram schematically showing the time axis characteristic of the correlation.

  When the loop gain estimation unit 22 first calculates a correlation value of a predetermined level or more from the timing when the pseudo noise is output, the loop gain estimation unit 22 regards the correlation value in the first calculated time zone as a direct wave and obtains a peak component of the direct wave. . That is, when the loop gain estimator 22 calculates a correlation value equal to or higher than a predetermined level, the loop gain estimation unit 22 then temporarily stores the correlation value in a predetermined time zone t1 in a memory (not shown), and the highest level in the predetermined time zone t1 A correlation value is extracted and set as a peak value a0. The predetermined level is set according to the level of stationary noise. The predetermined time period t1 for extracting the peak value is set according to the accuracy of correlation value calculation (code length of pseudo noise, etc.), the presence or absence of the HPF 19, and the cutoff frequency.

  When the loop gain estimation unit 22 first calculates a correlation value of a predetermined level or higher and then calculates the correlation value of the predetermined level or higher again after the predetermined time period t1 has elapsed, the loop gain estimation unit 22 converts the correlation value as a reflected wave. The peak component of the reflected wave is determined. Similarly to the above, when the correlation value of the predetermined level or more is calculated, the loop gain estimation unit 22 then temporarily stores the correlation value of the predetermined time zone t1 in the memory, extracts the correlation value of the highest level, and the peak value a1 And Similarly, the peak value (a1, a2,...) Of the reflected wave is extracted for a predetermined time length t2. The predetermined time length t2 referred to here corresponds to a pseudo-noise output period. When the reverberation time in the room is known to some extent, the time t2 may be set in advance or may be manually input by the user.

  Then, the loop gain estimation unit 22 obtains absolute values (| a1 |, | a2 |,...) Of the peak values of the extracted direct wave and reflected wave, and estimates the loop gain from the sum of the absolute values. Thus, the loop gain estimation unit 22 performs the process of estimating the loop gain from the feedback component of the direct wave and the reflected component of the reflected wave that affects howling, and thus can estimate the loop gain with high accuracy. The first estimation method assumes that it is often the peak component that affects howling, and performs loop gain estimation based on the sum of correlation values of the peak components of the direct wave and the reflected wave.

  Next, the second method will be described with reference to FIG. FIG. 5 is a diagram schematically showing the time axis characteristic of the correlation.

  The loop gain estimator 22 extracts all correlation values of a predetermined level or higher until a predetermined time length t2 elapses from the timing at which the pseudo noise is output, and calculates the sum of these absolute values (determines an integral value). The predetermined level in this case is also set according to the level of stationary noise. The predetermined time length t2 also corresponds to the pseudo-noise output period.

  As described above, the second method performs the loop gain estimation with high accuracy by summing all the components of the direct wave and the indirect wave.

  Next, the third method will be described with reference to FIG. FIG. 6A is a diagram showing the time-axis characteristic (absolute value) of the correlation, and FIG. 6B schematically shows the time-axis characteristic.

  The loop gain estimation unit 22 first extracts the peak value of the direct wave and obtains the absolute value | a0 | as shown in the first method. Then, the loop gain estimation unit 22 acquires the absolute value | b0 | of the correlation when the time t3 further elapses from the peak. The time t3 is obtained as the time (closed loop delay time) from the timing at which the pseudo noise is output to the timing at which the correlation peak is first calculated. (In this method, the timer 21 is indispensable.) The absolute value | b0 | is not limited to the value of the timing at which the time t3 has elapsed from the first peak, but after the time t3 has elapsed and in the vicinity thereof (for example, It may be a value when the absolute value of the correlation is the largest at around several tens of μsec. Then, the loop gain estimation unit 22 estimates a ratio (| b0 | / | a0 |) between the absolute value | a0 | and the absolute value | b0 | as a loop gain.

  In the third method, after first extracting the peak component of the direct wave, the waveform around the timing when the time t3 has passed is determined as a direct wave in which the pseudo noise output from the speaker 3 is looped again, and the loop The gain is estimated.

  As a modification of the third method, the peak component of the direct wave extracted first may be estimated as the loop gain. Since the direct wave component has a large influence on the occurrence of howling, the loop gain can be easily estimated.

  In addition, since the output period of the pseudo noise is set to be longer than the convergence time of the impulse response in the acoustic transmission system in each of the first method, the second method, and the third method, after outputting the pseudo noise, Next, until the pseudo noise is output, dummy noise may be output to eliminate the silent section. By always outputting a noise sound, the pseudo noise becomes inconspicuous and there is no sense of incongruity in hearing.

  The loop gain estimated by the loop gain estimation unit 22 as described above is output to the gain control unit 23. When the estimated loop gain approaches the predetermined threshold value th, the gain control unit 23 instructs to suppress the gain of the audio signal volume 13 because the possibility of howling is high. Further, the gain control unit 23 may perform a warning (lights the LED, displays a warning on the display, etc.) when the loop gain approaches the threshold value. Note that only one of the gain suppression processing and the warning processing may be performed, and further warning may be performed while suppressing the gain of the audio signal. Alternatively, a warning may be given first, and then gain suppression processing may be performed.

  Here, the predetermined threshold th varies depending on the loop gain estimation technique. The predetermined threshold th may be any value, but a certain margin is set in advance. For example, before the actual use, the user performs an operation of increasing and decreasing the gain, and when howling occurs, an operation of inputting howling is performed at the operation unit (not shown) of the mixer 1. Alternatively, in any processing unit of the mixer 1, the frequency characteristics of the audio signal are analyzed, and when a single frequency component becomes a high level for a predetermined time or more, it is detected as howling. The gain control unit 23 sets the estimated value of the loop gain input at this time as the maximum value thmax of the threshold, and sets th = α × thmax using a certain coefficient α (0 <α ≦ 1).

  As described above, the calculation unit 5 estimates the closed-loop gain, and when the estimated loop gain approaches a predetermined threshold, performs processing for suppressing the gain of the audio signal and processing for performing a warning, Howling can be prevented in advance.

  Since the mixer 1 of the present embodiment can predict howling based on the estimated loop gain, even if the position of the microphone often moves such as in a presentation or live performance, the howling is preferably performed. Can be prevented.

  The gain control unit 23 instructs to suppress the gain of the audio signal volume 13 and also suppresses the gain of the pseudo noise volume 18. However, a gain equal to or higher than a predetermined value is held so that the first peak of the pseudo noise correlation can always be detected. For this predetermined value, a value measured in advance in a laboratory or the like may be used, or a test is performed before actual use in the installation environment to obtain a limit gain that can calculate a correlation peak, and a certain margin is obtained. You may set the value which saw.

  Note that a plurality of pseudo noise patterns generated by the M-sequence generator 15 may be prepared, and these patterns may be switched. For example, by switching the pseudo noise pattern for each microphone (for each input channel), even when multiple microphones are used at the same time, the correlation can be calculated with high accuracy without interfering with each other's pseudo noise. Can do. Since the loop gain of the closed loop can be estimated individually for each microphone, howling can be suitably prevented even when a plurality of microphones are used simultaneously.

  In particular, when a Gold sequence is used as the pseudo-noise, it is possible to generate many types of code sequences by switching the tap position of the code generation circuit (shift register), so that it is compatible with a large-scale PA system. be able to.

  Note that the arithmetic unit 5 and the pseudo noise superimposing unit 7 may be built in the microphone instead of the mixer 1. Further, an adapter including the arithmetic unit 5 and the pseudo noise superimposing unit 7 is configured, a microphone is connected to the input unit (input interface) of the adapter, and the signal input from the input unit is input to the arithmetic unit 5 and the pseudo noise superimposing unit. It is good also as a structure supplied to 7. In any case, it suffices if it is provided before the amplification system such as an amplifier device.

It is the block diagram which showed the structure of the howling prevention apparatus. It is the block diagram which showed the structure and process content of the pseudo noise superimposition part. It is the block diagram which showed the structure and processing content of the calculating part. It is the figure which showed the time-axis characteristic of correlation. It is the figure which showed the time-axis characteristic of correlation. It is the figure which showed the time-axis characteristic of correlation.

Explanation of symbols

1-mixer 3-speaker 5-calculation unit 7-pseudo-noise superimposing unit 11-microphone

Claims (6)

An input unit for inputting an audio signal;
A noise generator that generates pseudo noise;
A superimposing unit that superimposes the pseudo noise on the audio signal input by the input unit and outputs it to an amplification system;
A correlation calculation unit for obtaining a correlation between the audio signal input by the input unit and the pseudo noise generated by the noise generation unit;
From the correlation value calculated by the correlation calculator, a loop gain estimator for estimating a closed loop gain;
With
When the correlation calculation unit first calculates a correlation value of a predetermined level or more after the noise generation unit outputs pseudo noise, the loop gain estimation unit directly calculates the correlation value in the first calculated time zone. A loop gain estimation device that estimates a closed loop gain based on a wave component . In addition to the direct wave component, the loop gain estimating unit calculates a correlation value of the direct wave component by the correlation calculating unit, and then an indirect wave component that is a correlation value equal to or higher than a predetermined level for a predetermined time length thereafter. The loop gain estimation apparatus according to claim 1 , wherein a closed-loop gain is estimated based on. The loop gain estimation device according to claim 2 , wherein the loop gain estimation unit estimates a closed loop gain from a sum of a peak value of the correlation value of the direct wave component and a peak value of the indirect wave component. If the gain of the closed loop approaches a predetermined threshold value, the loop gain estimating apparatus according to any one of claims 1 3 including a gain to suppress the gain control unit of the audio signal. An input unit for inputting an audio signal;
A noise generator that generates pseudo noise;
A superimposing unit that superimposes the pseudo noise on the audio signal input by the input unit and outputs it to an amplification system;
A correlation calculation unit for obtaining a correlation between the audio signal input by the input unit and the pseudo noise generated by the noise generation unit;
From the correlation value calculated by the correlation calculator, a loop gain estimator for estimating a closed loop gain;
And a gain control unit configured to suppress a gain of the audio signal when the closed-loop gain approaches a predetermined threshold value .   The loop gain estimation apparatus according to claim 1, further comprising a warning unit that issues a warning when the gain of the closed loop approaches a predetermined threshold value.