JPS59214341A - Howling canceler - Google Patents

Abstract

PURPOSE:To cancel howling in any state by preparing the impulse response between a speaker and a microphone by a teaching type identifying method from a microphone input signal, and estimating and subtracting a speaker feedback signal therefrom. CONSTITUTION:A howling canceler cancels the generation of howling in the sound reinforcement device consisting of the microphone 8, an amplifier 10, the speaker 9, etc. The microphone input signal is inputted to an adding circuit 11, whose output signal is inputted to a delay circuit 13, the output of which is branched into two . The impulse response between the sepeaker and microphone is simulated by using the teaching type identifying method and the convolutional arithmetic between one branched signal and simulated impulse response is carried out to generate a signal to be fed back from the speaker 9 to the microphone 8 by an impulse response approximating circuit 12. The output signal of this approximating circuit 12 is inputted to the adding circuit 11 in an opposite phase and the other branch signal is outputted through the amplifier and speaker to cancel the howling.

Description

DETAILED DESCRIPTION OF THE INVENTION FIELD OF INDUSTRIAL APPLICATION The present invention relates to a noise canceler for canceling the occurrence of ringing in a public address system composed of a microphone, an amplifier, a speaker, etc.

1. Construction of a conventional example and its problems FIG. 1 shows a block diagram of the construction of a conventional canceller.

In Fig. 1, 1 is a sound source (such as a speaker), 2 is a microphone, 3 is a loudspeaker, 4 is an amplifier, 6 is an addition circuit, and the impulse response between the 6-digit speaker 3 and the micro 1172 is determined using a learning identification method. An impulse response is created by sequentially creating a simulation of the signal fed back from the speaker 3 to the microphone 2 by performing a convolution operation between the signal ej input to this circuit and the impulse response hj created here. This is an approximate circuit. Sj is the signal input to microphone 2 from sound source 1, and X is the signal input to microphone 2 from sound source 1.
yj is the signal output from the microphone 2△, yj is the output signal of the impulse response approximation circuit 6, ej is the difference signal between yj and yj, and hj is the signal output from the impulse response approximation circuit 6. created impulse response,
ham represents the impulse response of the amplifier 4, the impulse response of the hsp speaker 3, and the impulse response between the hsp speaker trichrophone, and the relationship between the respective signals is as follows. (However, these signals indicate temporally discrete signal sequences, and ko represents an arbitrary time point.) *, Yj = Sj + x *h
...(1,1)△ ej=y co -y co
・・Mark・(1,2)I]−(3j j※ ham
* hsp -... (1, 3) △ y ko e 1 *hj ・・
...(1, 4) From the above equation, the difference signal aj becomes as shown in the following equation.

ej-(sj+,z-ko*h”) (e-J
-1* hj)* ” fsj+ (ej-+ * hamp* hs
p) * h) - (e21 * hj ) * "" S] + ej + * (hamp * hsp * h-
hj)・・・・・・(1・5) * By sequentially correcting hj to approximate hsp*h, the signal 1dsj output from the speaker 3
(7) Mi,! : The signal returning from the speaker 3 to the microphone 2 can be removed, and the occurrence of howling can therefore be canceled.

Here, the method of approximating hj to barnp** hsp*h using the learning identification method is to use the correction amount Δhj of hj
As, △hj-tx-ej・sign (e 3-+
)/i,, ,)-(16) Here, 5j4n(
e2) is the length of the function pulse response hj representing the sign of 8j H) 2>α>0.

Find Δhj as in equation (1.6), hj++=i1j+Δhj
......(1, 7) and the method of sequential correction,
It can be approximated to hi f hamp** hsp*h.

However, although the learning identification method is used here to estimate the impulse response between the speaker 3 and the microphone 2, when the distance between the microphone and the speaker is small, sj and rj (corresponding to ej) are very Since there is a high correlation with
Identification is difficult, approximation is not performed well, and the speaker output may be zero.

OBJECTS OF THE INVENTION It is an object of the present invention to provide an adaptive howling canceller that can eliminate the drawbacks of conventional devices as described above and can cancel howling in any situation.

Composition of the Invention The present invention is a public address system consisting of a microphone, an amplifier, and a speaker, in which an impulse response between the speaker and the microphone is approximated by a learning identification method, and a signal fed back from the speaker to the microphone is calculated by a convolution operation. It has an impulse response approximation circuit that simulates the
The simulated speaker feedback signal output from this circuit is subtracted from the microphone input signal, and the difference signal is delayed and output from the speaker to cancel the occurrence of howling.

DESCRIPTION OF EMBODIMENTS An embodiment of the present invention will be described below with reference to FIG. In Fig. 2, 7 is a sound source, 8 is a microphone, 9 is a speaker, 10 is an amplifier, 11 is an addition circuit, and 12 is a model of the impulse response between the speaker 9 and the microphone 8 using a learning identification method. The signal 19j7 input to the circuit and the impulse response hj created here
An impulse response approximation circuit generates a signal to be fed back from the speaker to the microphone by performing a convolution operation with . 13 is a delay circuit. In addition, sj is a signal input from the sound source 7 to the microphone 8, and I is a signal input to the speaker 9.
Signal output from 1. The output signal from the Yjid microphone △8, yj is the output signal from the impulse response approximation circuit △12, ej is the difference signal between yj and yjO, and e-co.

is a signal obtained by delaying ej output i'' from the delay circuit 13 by a time τ, h3 is an innopulse response that approximates the impulse response between the speaker 9 and the micropon 8 created in the impulse response approximation circuit 12, and ham is The impulse response of the amplifier 10, hsp is the impulse response of the speaker 9, and h* is the impulse response between the speaker 9 and the microphone 8.

The relationship between the respective signals is as follows, similar to the formula shown in the conventional example.

yco S fu+eco*h*
・・・・・・ (21)△ e co y co y J
(2, 2) From these equations, the difference signal ej can be expressed as follows.

△ ej=yj−yj= (sj+,,rj *h
) (e; −r*hj)=(sj+(ej
, * hamp * hsp) * h) - (eko7 * h
j ) 18] +fllj , -* (hamp* hs
p※h−hj)・・・・・・(2,5) Here, as in the conventional example, using the learning identification method, hj is
By approximating amp*hsp*h, the signal output from the speaker is only sj, so the signal fed back from the speaker to the microphone can be removed, and howling can therefore be canceled. When hj is approximated by hamp※※ hsp※h K using the learning identification method, 8 j -7 delays the difference signal ej by τ time, so the correlation with sj is small, so identification is easy. can be done.

However, by setting the delay time of the difference signal aj to about 20 ms, the influence of the time delay on the speaker can be eliminated.

However, the method of approximating the impulse response using the learning identification method is performed as shown in the following equation, similar to the case of ρ shown in the conventional example.

Assuming that the amount of correction of hj is Δhj, Δhj−α, ef, sign(6corr)/1e;
, 1---G2. e5:) Here, sign (e
J-7) is a function representing the sign of 6j 7
, Also, l5j−, rl=, Σ” j −r−i l
(N is -1 length of hj) 2〉α〉0, hco+, = hco+Δhco ・
By making corrections as (2, 7), hj can be approximated to hamp*hsp*h*.

Effects of the Invention As described in detail in the embodiments, the present invention uses a microphone input signal learning identification method to create an impulse response between a speaker and a microphone, and then estimates and subtracts a speaker feedback signal. This provides an adaptive howling canceller that can eliminate feedback of signals from the speaker to the microbon.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a conventional howling canceller, and FIG. 2 is a block diagram of an embodiment of the present invention. 7...Sound source, 8...Microphone, 9
...Speaker, amplifier for 10..., 11...
group/addition circuit, 12...impulse response approximation circuit, 13...delay circuit.

Claims (1)

[Claims] Howling canceller cancels howling by removing the signal fed back from the speaker output to the microphone from the microphone input signal of a public address system consisting of a microphone, lamp, speaker, etc.The microphone input signal is input to an adder circuit, and this The output signal of the adder circuit is input to a delay circuit, this output is branched into two, the other signal is used to simulate the impulse response between the speaker and the microphone using a learning identification method, and the output is divided into two. an impulse response approximation circuit that sequentially creates a signal that is fed back from the speaker to the microphone by performing a convolution operation between the branch signal and the simulated impulse response;
The output signal of this impulse response approximation circuit is input to the adder circuit in reverse phase, and the other branch signal which is branched into two is amplified through an amplifier and a speaker, thereby canceling howling. No Uring Canceller.