JPH01307321A - echo canceller - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a speaker that converts a received signal from a receiving line into voice and outputs the same, and a speaker that converts the voice to be transmitted into a transmitted signal and connects it to a transmitting line. A telephone that includes a microphone that outputs to
For example, in a hands-free telephone (such as a hands-free telephone), by creating a corresponding pseudo echo from the received signal and canceling out the echo that is generated when the sound output from the speaker is reflected directly or reflected on the walls of the room and wrapped around the microphone. This invention relates to an echo canceller that performs echo cancellation.

[Conventional technology]

2. Description of the Related Art Hands-free telephone devices, which allow telephone calls to be made using a microphone and speaker without using a conventional handset, are becoming popular for regular telephones in homes. This allows users to talk without having to hold the handset in their hands, so they don't have to worry about hand fatigue from long calls or ear pain from pressing the handset against their ears. Further, since the user's hands are freed, there are advantages such as being able to talk while doing something with the hands. Particularly when using a car phone while driving, this hands-free calling function has a great safety advantage.

By the way, telephone lines consist of 2-wire and 4-wire lines.
A hybrid circuit is used for the conversion. However, in this hybrid circuit, it is difficult to perfectly match impedances, so signal reflection occurs.

-a, when trying to talk using a microphone and speaker on a telephone, a signal loop is formed by the acoustic coupling between the microphone and speaker and the reflection of the signal from the 2-wire to 4-wire conversion circuit on the line side. The gain of this -circular loop is 1 (O
dB), a howling phenomenon occurs and the call is actually -F.
It becomes impossible. Therefore, as it is, it is impossible to raise the level of the speaker or the sensitivity of the microphone, making it impossible to use it for practical purposes. Therefore, it is necessary to prevent acoustic howling by using an echo suppressor, echo canceller, or the like. Further, since signal reflection in a 2-wire to 4-wire conversion circuit causes a large time delay in long-distance calls, even if it does not result in howling, it becomes a so-called echo, which interferes with conversation.

This is a method to prevent howling: the L-coupler suppressor uses a so-called voice switch method. this is,
The gain of the above-mentioned open loop is prevented from exceeding 1 by comparing the speech levels of the two parties, increasing the insertion loss amount of the smaller one, and effectively cutting off the signal path of the larger one.

However, with this method, there is a time lag when comparing call levels, resulting in a delay in changing the amount of loss, resulting in the beginning of a word being cut off, and in a noisy usage environment, noise may continue to enter. There are drawbacks such as a blocking phenomenon in which the insertion loss amount is not changed.

As an alternative method, an air cancellation method using digital signal processing has been attracting attention in recent years as the cost of semiconductors has declined and digital signal processing technology has progressed.

This echo cancellation method does not have the drawbacks of the echo suppressor described above, and is being put into practical use in teleconferencing systems and the like.

Next, the principle of this echo canceller will be explained using an explanatory diagram (FIG. 2) of a hands-free telephone using an echo cancellation method. In this method, out of the signal (s(t) +y(t)) input to the microphone 5, the speaker 2
The echo signal canceling circuit 3 cancels only the signal y(t) outputted from the room and reflected by the walls of the room. therefore,
The signal loop described above is not formed and howling is prevented.

In addition, with this system, there is no need to turn on or off the power switch, unlike a voice switch, so simultaneous calls can be made, and good call quality can be obtained without cutting off the beginning or end of a word.

First, consider a case where the input signal to the microphone 5 is only the echo sound from the speaker 2. In other words, the transmission signal 5 (
This is the case when t)=0.

Received signals x(t) from 10 telephone calls are inputted to an echo cancellation circuit (adaptive filter) 3 through an AD converter 4 and also inputted to a speaker 2. The sound emitted from the speaker 2 is reflected by the walls of the room and is input to the microphone 5 as a signal y(t).

The path along which the received signal x(t) becomes y(t) is called the anti-traditional path.

On the other hand, the received signal X(0) converted into a digital signal by the AD converter 4 is sequentially stored in the X register 8 composed of n register stages.

Each time the register 8 stores l samples of data, the data is sequentially moved to the next register stage, and the data in the last register stage is output and discarded. As a result, data of n samples of received signals (x(t) to x(-old 1)) are always stored in the X register 8, which is composed of n register stages.

The tap coefficient register 9 is composed of n register stages, the same number as the X register 8, and has a tap coefficient D (hO(t) to hn−+(
t), n; number of taps) are stored.

The convolution calculator 10 performs a convolution operation using each data from the tap coefficient register 9 and the X register 8 as input. That is, the pseudo-old signal y(t) is calculated by (1). subtractor l
In l, the input signal from the microphone 5 is sent to an AD converter 6.
The pseudo echo signal y(
t) as the result of the subtraction operation. In other words, it cancels the echo signal. This result is the estimation error of the echo path,
This is called the residual signal e(L), e (t) −y(t) −y (t)
...It can be expressed as (2). The adaptive filter (echo signal cancellation circuit) 3 modifies the previous tap coefficient using an algorithm such that this residual signal e(L) approaches O.

An example of this algorithm is the LMS method (Least Mean Square method).
There are well-known algorithms such as eMethod) or learning identification method. This involves modifying the tap coefficients one after another based on the residual signal e(t) and the received signal x(t),
This method provides this as a new tap coefficient. This can be expressed as a formula as follows.

Letting the correction amount of the tap coefficient be Δh, (t), h5(t+
1)=hi (t)+Δh, (t)...(3) Δh, (t)=G-xt(t)・e(t) −・”
(4) Here, G is a correction coefficient, which is a constant in the LMS method, and a value given by the following (5) in the learning identification method.

The correction amount calculator 12 successively calculates each Δh, (t), and the adder 13 adds this to the corresponding data (tap coefficient) read from the tap coefficient register 9.
It is stored again in the original location of the tap coefficient register 9.

Incidentally, this estimation of the tap coefficient was performed when the input to the microphone 5 was only the echo signal y(t). If there is a sound 5(t) (transmission signal) other than the echo signal y(t),
That is, if (y (t) + s (t) ), the tap coefficient estimation becomes inaccurate due to the transmit signal 5(t). Therefore, when there is a transmitting signal 5(t), it is necessary to prohibit updating of the tap coefficients. As a means for this, an input signal (s (t) + y (t
)) and the input signal to speaker 2, that is, the reception signal x
(t) and [s (t) + y
(t)] is larger than the power of y(t) by a certain amount, it is determined that the transmission signal 5(t) exists, and the output data of the correction amount calculator 12 is set to 0, that is, the tap coefficient update operation is performed. Temporarily cancel.

As a result, even when there is a sending signal S([), the echo canceling operation can be performed stably.

The following is the basic principle of an echo canceller.

One of the problems with this echo cancellation method is that when used for hands-free calls or conference calls, it depends on the impulse response of the room, which is determined by the size and shape of the room, the material of the walls, etc. In order to obtain this, it is necessary to cancel echoes that have a long delay time of several hundreds of meters. For this purpose, the telephone band (
The number of taps of an adaptive filter for signals in the range of 300 to 3400 Hz ranges from hundreds to thousands, and requires a huge amount of hardware. In addition, the hardware is DSP (Digital Signal
Even in the case of a configuration using arithmetic processing tC such as Processor), there is a limit to the delay time that can be canceled due to constraints from the calculation time.

As a countermeasure to this problem, there is a method described in Japanese Patent Publication No. 62-51528 and No. G in which a plurality of adaptive filters are connected in series to form an echo canceller.

[Problems to be Solved by the Invention] 1. However, in the above-mentioned prior art, each adaptive filter does not include a subtracter for subtracting a pseudo-echo signal from a microphone signal. Since the DSP itself has a subtraction function, when an adaptive filter is constructed using the DSP, there is no need to provide an external subtracter by using the subtraction function within the DSP.

An object of the present invention is to provide an echo canceller that performs echo compensation using optimal hardware for the purpose by connecting a plurality of adaptive filters having the same configuration.

[Means to solve the problem]

In order to achieve the above object, the present invention includes a speaker that converts a reception signal from a reception line into voice and outputs it, and a microphone that converts the voice to be transmitted into a transmission signal and outputs it to the transmission line. Hans 71J - The first echo canceller that performs echo cancellation by creating a corresponding pseudo echo from the received signal and canceling the echo caused by the sound output from the speaker going around to the microphone in a telephone, etc. a first input terminal (31) that inputs the signal; a first register (37) that sequentially stores the first signal input from the first input terminal (31) over a plurality of samples over time;
A first output terminal (34) that sequentially outputs each sample from the first register (37), a second register (38) that stores tap coefficients, and the first register (37).
sample data taken out from the second register (
a convolution operator (39) that performs a convolution operation using the tap coefficients taken out from the input terminal (38); a second input terminal (32) that inputs a second signal;
) from the second signal input from the convolution operator (
a subtracter (40) that subtracts the output of the subtractor (39);
A second output terminal (3) outputs the subtraction result in (40).
5) and a third input terminal (33) for inputting a third signal.
and a third register (41) that stores the third signal input from the third input terminal (33), and outputs the third signal input from the third input terminal (33). A third output terminal (36), a tap stored in the second register (38) from the output of the third register (41) and sample data taken out from the first register (37). Correction amount calculation unit (4) that calculates the correction amount of the coefficient
2) and a correction adder (43) that corrects the tap coefficient stored in the second register (38) according to the calculated correction amount. N pieces (N is any integer) were prepared.

[Effect]

In the first adaptive filter, the first input terminal (31
) and the output signal of the microphone from the second input terminal (32).
In the N-th adaptive filter, the output signal from the second output terminal (35) is input to the third input terminal (33) as the third input signal in the first adaptive filter. , for each of the second to N-1th adaptive filters, each of the first to third input terminals (3
1, 32, 33) as the first to third adaptive filters in the previous stage.
The first to third output terminals (34, 35, 36) are connected to the first to third input terminals (31, 36) of the subsequent adaptive filter.
2, 33), and extracts an output signal from the second output terminal (35) or the third output terminal (36) of the Nth adaptive filter and sends it to the transmission line.

〔Example〕

Embodiments of the present invention will be described in detail below with reference to the drawings.

A case will be described in which the echo canceller of the present invention is applied to prevent howling. FIG. 1 is a block diagram showing a first embodiment of the present invention.

In the figure, first, the configuration of the echo canceller of this embodiment will be explained.

The received signal x(t) from the other party is reproduced into an acoustic signal as an input signal to the speaker 2 and is input to the AD converter 4. Here the analog signal is converted into a digital signal. In this example, the telephone band (300 to 34
Since the target is 00t(z), the sampling frequency is 8.
K) Iz, the number of quantization bits is 16 bits. The output terminal of this AD converter 4 is the first stage adaptive filter 30A.
The first input of

It is connected to a terminal 31, and the received signal converted into a digital signal is input as first input data to the adaptive filter 30A.

On the other hand, as shown in FIG.
A to 3ON) have their corresponding input terminals and output terminals connected. Of these, the second output data of the final stage adaptive filter 3ON is input to the third input terminal 33 as the third input signal of the first stage adaptive filter 30A. Also,
The microphone input signal input from the microphone 5 is
After being converted into a digital signal by the converter 6 under the same conditions as the AD converter 4, the second signal of the first stage adaptive filter 30A is
The signal is input to the second input terminal 32 as an input signal.

Furthermore, the third output terminal 3 of the final stage adaptive filter 3ON
6 is connected to a DA converter 7, and the third output data is reproduced as an analog signal by the DA converter 7 and sent to the other party as a transmission signal.

Next, the operation of each adaptive filter will be explained using FIG. In FIG. 1, adaptive filters 30A to 3ON
Since they all have the same configuration, one of them is shown as 30 in FIG.

That is, 30 is an adaptive filter, and input terminals 31 to 3
3 and output terminals 34 to 36, respectively. The sample data input from the first input terminal 31 is transferred to the X register 37, which is composed of n 16-bit register stages.
are held sequentially.

Here, the data at time t (n pieces of data) held in the X register 37 is expressed as X 6 (t) ”-X n-t
(t) At time 0 (t+1), new data x (
t+1), the contents of the n register stages are moved to the next neighbor in turn. That is, X = - + (t + 1) = X t (t).

i=0.1. ......, n-2 ・・・・・・(6) becomes.

The last data x I'l-1(t) is passed as first output data from the first output terminal 34 to the first input terminal 31 of the next stage adaptive filter, where it is similarly The filters are processed and then passed to the next stage of adaptive filters connected in sequence. And the final stage adaptive filter 3 is turned on.
Therefore, the data output from the first output terminal 34 as the first output data is discarded as is.

Next, the tap coefficient register 38 consisting of n 16-pinto register stages similarly has values ha(t) to h, , -1 (0) in the n register stages at time t.

This is the tap coefficient mentioned in the known example and corresponds to the impulse response of the echo path.

In the convolution calculator 39, the tap coefficient register 38 and
Using the data in register 3, the calculation shown in equation (1) is performed to calculate the output y(t). From the second input data input from the second input terminal 32, this y([
) is subtracted by the subtracter 40, and the value is used as second output data from the second output terminal 35 to the next stage adaptive filter 30B.
hand over to.

The adaptive filter 30B at the next stage performs the same processing using this as second input data.

In the final stage adaptive filter 3ON, the second output data is returned to the third input terminal 33 as the third input data of the first stage adaptive filter 30A, and this value is transferred to the E register 41.
After storing it in the third output terminal 3 as the third output data,
6 and then handed over to the next stage adaptive filter 30B. In the final stage adaptive filter 3ON, this third output data is D
The signal is input to the A converter 7 and becomes a transmission signal.

Next, in the LMS method, the third input data (i.e. e(t)) stored in the E register 41 and X+(t)
Based on this, the correction amount calculator 42 calculates the correction amount of the tap coefficient shown in equation (41) above. Thereafter, the values of the tap coefficient register 38 are read out one after another, and after the correction amount is added in the 'ζ adder 43, the values are stored in the same position of the tap coefficient register 38 again.

Each adaptive filter performs the above two series of operations in parallel within one sampling time (125
(μs).

As a result, even if a single adaptive filter can only have 128 taps, by cascading multiple adaptive filters, an adaptive filter with as many as 11,000 taps can be constructed, eliminating long echo signals. It is possible to configure an echo canceller. Further, there is an effect that the number of adaptive filters can be set depending on the time of the echo signal to be canceled.

Next, a second embodiment is shown in FIG. The difference from the first embodiment is that the input data to the DA converter 7 is taken from the second output terminal 35 of the final stage adaptive filter 3ON. The third output data of the final stage adaptive filter 3ON is then discarded as is. Although the data sent to the DA converter 7 is the same as in the first embodiment, it has the advantage of being able to receive data at a faster time.

Further, a third embodiment is shown in FIG. FIG. 5 is a functional Bronc diagram when the learning identification method is used to update the tap coefficients.

In comparison with FIG. 3, a square sum calculating circuit 44 is provided to calculate the square sum of the data stored in the X register 37, and the result is input to the correction amount calculator 42.

The correction amount calculation circuit 42 uses the calculation formula of the learning identification method (the above formula (
4) and (5)), the correction amount is calculated and used for updating the stamp coefficient. This method has the advantage of shortening the convergence time compared to the LMS method of the first embodiment.

A fourth embodiment is shown in FIG. This is a method in which the second output of the last-stage adaptive filter 3ON is temporarily stored in the E-buffer 18 in FIG. 1, and then delivered to the first-stage adaptive filter 30A. This allows the first-stage adaptive filter 30A to start the next operation without waiting for the second output from the final-stage adaptive filter 3ON.

However, in that case, a signal from one sample earlier will be used. However, this does not significantly affect the operation of the echo canceller itself (each adaptive filter can operate, and the number of adaptive filters that can be connected in series can be increased).

Further, in the fifth embodiment shown in FIG. 7, compared to the embodiment shown in FIG. 3, the value of the E register 41 is used as the third output. This E register 41 is the third
is updated each time an input signal is received. This allows the tap coefficient register 38 to be updated without waiting for the third input signal. However, in this case, as in the fourth embodiment, values before l samples are used.

In the above embodiments, each of the input and output terminals of the adaptive filter has been explained as a single cotton terminal.

However, since the data handled is 16-pin digital data, the transfer method may be parallel or serial. It is also possible to input and output three types of human output data in a time-sharing manner. Therefore, the apparent number of input/output terminals is not limited to six in total as shown in FIG.

Furthermore, in all of the embodiments, each block of the adaptive filter has been described as being constituted by hardware. However, with recent advances in semiconductor technology, DSP (Digital Signal Processor)
LS capable of high-speed calculation processing called cesSor)
I appeared. Using this, the adaptive filter of this embodiment can be constructed using a DSPI chip.

In this case, the signal processing block shown in FIG. 3 will be performed by software. Even in this case, the adaptive filters used in this embodiment can be operated using exactly the same software. Therefore, DSP
This program has an economical effect as you only have to pay for the mask once. Furthermore, by using software that has multiple built-in operating modes and switches the operating modes using a switch, etc., it can be made to look like it is composed of multiple types of adaptive filters. However, in this case, D requires the most processing time.
SP becomes a bottleneck, reducing overall processing power, and DSP
unable to use their full potential.

Therefore, this embodiment, which equally divides the processing to each DSP, has the effect of maximizing the operating efficiency of the DSPs.

〔Effect of the invention〕

According to the present invention, by connecting one or more adaptive filters made of exactly the same hardware, it is possible to realize an echo canceller with an arbitrary echo compensation time depending on the number of adaptive filters connected. . That is, by simply converting one type of adaptive filter into an LSI, it is possible to configure many types of echo cancellers depending on the purpose.

Furthermore, even when the adaptive filter is constructed using a DSP, it can be constructed using one type of software, and its operations are the same, resulting in good operational efficiency of the DSP.

As a result, it is possible to construct an echo canceller suited to the purpose at a low cost.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing a first embodiment of the present invention, Fig. 2 is a block diagram showing an example in which a conventional echo signal canceling circuit is applied to howling prevention, and Fig. 3 is a block diagram showing the first embodiment. A block diagram showing the internal functions of the adaptive filter shown in Figure 4.
The figure is a block diagram showing a second embodiment of the invention, FIG. 5 is a block diagram showing a third embodiment of the invention, FIG. 6 is a block diagram showing a fourth embodiment, and FIG. FIG. 7 is a block diagram showing the fifth embodiment. Explanation of symbols 18... E buffer, 30. 30A-3ON... adaptive filter, 31-33... input terminal, 34-36.
...Output terminal, 37...X register, 38...Tap coefficient register, 39...Convolution operator, 40...
Subtractor, 41... E register, 42... Correction amount calculator, 43... Adder, 44... Square sum calculator. Agent Patent Attorney Akio Namiki 12 Seal 3!1 lI41! l Filling 5 Figure 16 Figure

Claims (1)

[Claims] 1. A telephone including a speaker that converts a reception signal from a reception line into voice and outputs it, and a microphone that converts the voice to be transmitted into a transmission signal and outputs it to the transmission line. , an echo canceller that performs echo cancellation by creating a corresponding pseudo echo from the received signal and canceling the echo caused by the sound output from the speaker going around to the microphone; an input terminal (31), a first register (37) that sequentially stores a first signal input from the first input terminal (31) over a plurality of samples over time; ), a second register (38) that stores tap coefficients, and a first output terminal (34) that sequentially outputs each sample from the first register (37);
a convolution operator (39) that performs a convolution operation using tap coefficients taken out from a register (38); a second input terminal (32) into which a second signal is input; ); a subtracter (40) that subtracts the output of the convolution operator (39) from a second signal input from the subtractor (40); and a second output terminal (35) that outputs the subtraction result in the subtracter (40). , a third input terminal (33) for inputting a third signal, a third register (41) for storing the third signal input from the third input terminal (33), and a third register (41) for storing the third signal input from the third input terminal (33); A third output terminal (36) that outputs the third signal input from the input terminal (33), and sample data extracted from the output of the third register (41) and the first register (37). and from the second register (38
), and a correction unit for correcting the tap coefficient stored in the second register (38) according to the calculated correction amount. an adder (43), and N adaptive filters from the first to Nth (however,
is an arbitrary integer), and in the first adaptive filter, the first input terminal (31
) and the output signal of the microphone from the second input terminal (32).
In the N-th adaptive filter, the output signal from the second output terminal (35) is input to the third input terminal (33) as the third input signal in the first adaptive filter. , for each of the second to N-1th adaptive filters, each of the first to third input terminals (3
1, 32, 33) as the first to third adaptive filters in the previous stage.
The first to third output terminals (34, 35, 36) are connected to the first to third input terminals (31, 36) of the subsequent adaptive filter.
2, 33), and extracts an output signal from the second output terminal (35) or the third output terminal (36) of the Nth adaptive filter and sends it to the transmission line. echo canceller.