JP2585279B2 - Eco-Cancer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial application field) The present invention relates to an echo canceller used for, for example, a loudspeaker apparatus.

(Prior Art) In recent years, with the widespread use of audio conferences, it has been desired to provide a loudspeaker apparatus capable of ensuring simultaneous communication and having less echo (echo).

To meet such demands, there is a loudspeaker apparatus equipped with an echo canceller, which has, for example, the configuration shown in FIG.

The loudspeaker apparatus shown in FIG. 1 includes a reception signal input terminal 1 for receiving a reception signal, an amplifier 2 for amplifying a signal input to the reception signal input terminal 1, and a signal amplified by the amplifier 2 as an audible sound. A speaker 3 for outputting, a microphone 4 for converting audible sound into an electric signal, an amplifier 5 for amplifying a signal from the microphone 4, an echo for estimating a reverberation path characteristic between the speaker 3 and the microphone 4 and generating a pseudo echo signal. The main part is composed of a canceller 6, a subtracter 7 for subtracting the pseudo echo signal from the signal amplified by the amplifier 5, and a transmission signal output terminal 8 for outputting the signal from the subtracter 7 as a transmission signal.

Here, the pseudo echo signal generated by the echo canceller 6 is represented by S, the signal amplified by the amplifier 5, the signal of the transmitted voice portion among the signals included in the signal S, and the signal S included in the signal S. Let y be the signal of the echo portion of the resulting signal.

The signal S is S = y + n. Since the pseudo echo signal is subtracted by the subtracter 7 from the signal S, the signal e output from the subtractor 7 is obtained.
Is e = y− + n.

Then, when the estimation of the echo path characteristic between the speaker 3 and the microphone 4 is correct, y =, thereby e = n, and only the echo is completely removed.

In this apparatus, it is necessary to perform training in order to set the initial value of the echo canceller 6 at the initial stage. Reference numeral 9 is a training source for generating a training signal for this purpose, and a switch 10 and a control circuit 11 are means for setting the apparatus to a training state at the time of initial training. That is, at normal times, the control circuit 11
The switch 10 is dropped to the A side and the received signal from the received signal input terminal 1 is input to the amplifier 2. During the initial training, the switch 10 is dropped to the B side by the control of the control circuit 11 to receive the received signal from the training source 9. The signal is input to the amplifier 2.

Incidentally, examples of the learning algorithm for the initial training of the echo canceller 6 include a steepest descent method using white noise as a training signal, a learning identification method, and a chirp method using a chirp signal (frequency sweep signal) as a training signal.

Steepest descent method, learning identification method is sequential learning algorithm, as shown in FIG. 5, the training time T ₁
Has the advantage of freely setting can moreover be set short operation word length, it has the disadvantage that the time T ₂ is longer to converge to the contrary finally.

Meanwhile, since chirp method is an algorithm of collective learning, as shown in FIG. 6, it has the advantage of a short time T ₂ of the until finally converged, on the other hand training time T ₁ is not fixed Moreover, there is a disadvantage that the operation word length is long.

(INVENTION Problems to be Solved point) Thus, in the learning algorithm of the conventional initial training of the echo canceler in hands-free communication device, the final in the case where it can set short addition operation word length can freely set the training time T ₁ manner there is a problem that the time T ₂ of the until convergence is long, that the time T ₂ of the up eventually converge becomes long addition operation word length training time T ₁ is not fixed to the shorter There is a problem.

The present invention has been made in order to solve these problems, and it is an object of the present invention to provide an echo canceller in which the training time can be set freely, the operation word length can be set short, and the time until final convergence is short. It is an object.

[Structure of the Invention] (Means for Solving the Problems) In other words, the echo canceller of the present invention has an output means for outputting a reception signal to an echo path, and a transmission signal for outputting an audible sound as a transmission signal. Output means; pseudo echo signal generation means for estimating echo path characteristics between the output means and the transmission signal output means to generate a pseudo echo signal; signal subtraction for subtracting the pseudo echo signal from the transmission signal. Means, and a training signal source for generating a training signal, wherein the pseudo echo signal generation means estimates the echo path characteristics by using both a sequential tap coefficient calculation method and a collective tap coefficient calculation method during training. Features.

(Operation) In the present invention, the pseudo echo signal generation means estimates the echo path characteristics using both the sequential tap coefficient calculation method and the collective tap coefficient calculation method during training, so that the training time can be set freely and The operation word length can be set short, and the time until the convergence is finally shortened.

(Example) Hereinafter, details of an example of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing a configuration of a voice loudspeaker according to one embodiment of the present invention.

The voice loudspeaker shown in FIG. 2 has almost the same configuration as that of the conventional example shown in FIG.

That is, the apparatus comprises a receiving signal input terminal 1 for receiving a receiving signal, an amplifier 2 for amplifying a signal input to the receiving signal input terminal 1, and a speaker 3 for outputting the signal amplified by the amplifier 2 as audible sound. A microphone 4 for converting an audible sound into an electric signal; an amplifier 5 for amplifying a signal from the microphone 4; a reverberation path characteristic between the speaker 3 and the microphone 4 to generate a pseudo echo signal; An echo canceller 6 having both a chirp algorithm and a learning identification algorithm for estimating the echo path characteristic at the time, a subtracter 7 for subtracting a pseudo echo signal from a signal amplified by the amplifier 5, and a signal from the subtracter 7 A transmission signal output terminal 8 that outputs a transmission signal constitutes a main part.

In this apparatus, it is necessary to perform training in order to set the initial value of the echo canceller 6 at the initial stage. Reference numeral 9 denotes a training source for generating a chirp signal (frequency sweep signal), which is a training signal for this purpose, and a switch 10 and a control circuit 11 are means for setting the apparatus to a training state at the time of initial training. That is, in normal operation, the switch 10 is dropped to the A side under the control of the control circuit 11, and the received signal from the received signal input terminal 1 is input to the amplifier 2. A reception signal from the training source 9 is input to the amplifier 2. Reference numeral 12 denotes a minus input terminal of the subtractor 7 which is connected to the echo canceller 6 side (C side) or the ground level (D
Side), which is switched by the control of the control circuit 11.

The initial training of the echo canceller 6 is performed as shown in FIG.

First, when this device is set to the ON state, the generation of a chirp signal, which is a training signal, is started from the training source 9 (section (d) in FIG. 2), and the switch 10 is controlled to the B side by the control of the control circuit 11. , Switch 12 is C
(C section in FIG. 2). At this time, learning is not yet started in the echo canceller 6, and dummy training by the chirp method is performed for N samples to load the chirp sequence in the echo canceller 6 (section (a) in FIG. 2). The value of N is a shift register (not shown) provided in the echo canceller 6.
Is set in accordance with the tap length of.

When such an operation is completed, the switch 12 is dropped to the D side under the control of the control circuit 11 (section (c) in FIG. 2), and the minus input terminal of the subtractor 7 is set to the ground level. Then, learning of the echo canceller 6 is started (section (b) in FIG. 2), and training by the chirp method is performed for N samples (section (a) in FIG. 2).

That is, the tap coefficient sequence of the echo canceller 6 is The echo is y (k), and the input signal sequence of the echo canceller 6 is When the step gain is α (0 <α <2, usually α = 1), Is performed N times and the average value is obtained.

When such an operation is completed, the switch 12 is again dropped to the C side under the control of the control circuit 11 (section (c) in FIG. 2), and the minus input terminal of the subtractor 7 is connected to the echo canceller 6.
Connected to the side. Then, training by the learning identification method is performed for mN samples (m ≧ 1, FIG. 2 (a)).
section).

That is, the tap coefficient sequence of the echo canceller 6 is The residual signal is e (k), and the input signal sequence of the echo canceller 6 is When the step gain is α (0 <α <2, usually α = 1), Is performed, and the tap coefficient sequence is set each time the calculation result is calculated. These operations are performed mN times.

In this manner, apparatus of this embodiment, when performing initial training of the echo canceller 6, as shown in FIG. 3, the section performs training by the time to converge T ₂ is short chirp method, interval training time T ₂
Since training is performed by a learning identification method that can set the calculation word length freely and the operation word length can be set shorter, the time until the convergence is finally shorter than that of the conventional learning identification method.
Also, after the section, the training time can be set freely and the operation word length can be set short.

In the above-described embodiment, the training is performed on the section by the learning identification method, but this may be the steepest descent method. The arithmetic expression in that case is the tap coefficient sequence of the echo canceller 6. The residual signal is e (k), and the input signal sequence of the echo canceller 6 is When the step gain is α (0 <α <2, usually α = 1), Becomes

[Effects of the Invention] As described above, according to the echo canceller of the present invention, the training time can be set freely, the operation word length can be set short, and the time until the convergence is finally shortened.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a loudspeaker apparatus according to an embodiment of the present invention. FIG. 2 is a diagram for explaining an initial training operation of an echo canceller in the loudspeaker apparatus shown in FIG. FIG. 3 is a diagram for explaining an effect produced by an initial training operation of the echo canceller in the loudspeaker apparatus shown in FIG. 1, FIG. 4 is a block diagram showing a configuration of a conventional loudspeaker apparatus, and FIG. Is a diagram for explaining the steepest descent method and the learning identification method, and FIG. 6 is a diagram for explaining the chirp method. 1 reception signal input terminal 2 amplifier 3 speaker 3 microphone 5 amplifier 5 6 echo canceller 7 subtracter 8 transmission signal output terminal 8 9 training source 10 …… Switch 11 …… Control circuit 12 …… Switch

 ──────────────────────────────────────────────────続 き Continuing from the front page (72) Inventor Shigenobu Minami 70 Yanagimachi, Saiwai-ku, Kawasaki City, Kanagawa Prefecture Inside the Toshiba Yanagimachi Plant (72) Inventor Takashi Saeki 3-3-1 Asahigaoka, Hino-shi, Tokyo Stock Company Toshiba Hino Plant (56) References JP-A-61-71727 (JP, A) JP-A-58-204633 (JP, A)

Claims (4)

(57) [Claims] 1. An output means for outputting a reception signal to an echo path,
Transmission signal output means for outputting an audible sound as a transmission signal,
A pseudo echo signal generating means for estimating an echo path characteristic between the output means and the transmission signal output means and generating a pseudo echo signal; a signal subtracting means for subtracting the pseudo echo signal from the transmission signal; A training signal source for generating a signal, wherein the pseudo echo signal generation means estimates the echo path characteristic by using both a successive tap coefficient calculation method and a collective tap coefficient calculation method during training. Canceller. 2. The method according to claim 1, wherein the pseudo echo signal generation means estimates the echo path characteristics by a collective tap coefficient calculation method during training.
2. The echo canceller according to claim 1, wherein the echo path characteristic is estimated by a successive tap coefficient calculation method. 3. The method according to claim 1, wherein the successive tap coefficient calculation method is a learning identification method or a steepest descent method.
The echo canceller described in the item. 4. The echo canceller according to claim 1, wherein the collective tap coefficient calculation method is a chirp method.