JPH0897753A - Echo canceler - Google Patents

Abstract

PURPOSE: To improve the arithmetic speed while keeping calculation accuracy by providing a means receiving a reference input signal generated by a reference input signal generating means and executing arithmetic processing based on a prescribed algorithm to generate a pseudo echo signal in the canceler. CONSTITUTION: The echo canceler 101 is made up of an arithmetic unit 105 calculating attenuation β of a transmission signal (x), a preamplifier 106 generating a reference input signal x' obtained by revising a signal level of the transmission signal (x) depending on the attenuation β calculated by the arithmetic unit 105, and a DSP 107 receiving the reference input signal x' generated by the preamplifier 106 to generate a pseudo echo signal y'. Then the reference input signal obtained by revising a level of the transmission signal is generated based on the transmission signal for output of a voice or the like to a destination, the received signal and a prescribed algorithm and the arithmetic processing based on the prescribed algorithm is executed by using the reference input signal x' to generate a pseudo echo signal y'.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo canceller for canceling a signal (echo) sneaking around from a transmitting side to a receiving side in a videophone system, a voice communication system or the like.

[0002]

2. Description of the Related Art In recent years, while the demand for communication using long-distance lines such as satellites and submarine cables has increased, signal processing devices that perform complicated processing have been introduced into communication systems in order to support various communication services. Therefore, in various communication systems, the signal propagation delay time tends to increase. When the delay time becomes large, it is indispensable to cancel the echo that goes around from the transmitting side to the receiving side. Therefore, an echo canceller that cancels this echo has been widely used in communication systems.

In general, this echo canceller adaptively estimates a frequency characteristic of an echo propagation path (hereinafter referred to as an echo path) peculiar to the line to which it is connected as a time series impulse response using a transversal filter. , A pseudo echo signal (pseudo echo) is generated to cancel the echo.

For example, voice communication in a video conference system is a conference in which a speaker / microphone is used to watch TV (partner) with high quality voice. However, since the speaker and the microphone are acoustically coupled to generate an echo in such a system, an echo canceller is also used in such a system.

FIG. 3 is a schematic block diagram showing a system configuration (video conference system) to which a conventional echo canceller is applied. Reference numeral 301 is an echo canceller, 302 is a speaker, and 303 is a microphone. The operation of the echo canceller 301 will be described with reference to FIG.

A transmission circuit (not shown) is connected to the terminal c, and the speaker 302 emits sound by inputting the transmission signal x output from the transmission circuit.
One of the microphones 303 outputs a reception signal obtained by picking up sound to the terminal d via the adder 304. A receiving circuit (not shown) for receiving a received signal is connected to the terminal d.

In this system, the sound emitted by the speaker 302 is picked up by the microphone 303 through the acoustic space, so that an echo is generated. The microphone 303 is
The echo signal y is output by collecting the sound emitted by the speaker 302.

The echo canceller 301 is a DSP (Digita).
Signal Processor) is used, and processing for eliminating the echo signal y is executed by executing arithmetic processing based on an algorithm such as the LMS algorithm. Specifically, first, the transmission signal x is input as the reference input signal x, and this reference input signal x and the impulse response h of the acoustic space (echo path) between the speaker 302 and the microphone 303 that are sequentially estimated are used. A pseudo echo signal y ′ is generated by performing convolutional integration, and this is output to the adder 304.

The impulse response h is estimated by using an adaptive transversal filter. This transversal filter generally obtains an optimum impulse response h that matches the characteristics of the echo path by adaptively adjusting the coefficients of the taps arranged at predetermined time intervals, using a shift register, a multiplier, D from adder
It is constructed in SP107.

The pseudo echo signal y'generated is inverted in its signal level from the echo canceller 301 and output to the adder 304. Upon receiving the pseudo echo signal y ′, the adder 304 adds the reception signal output from the microphone 303 and the pseudo echo signal y ′ to subtract the pseudo echo signal y ′ from the reception signal. As a result, the echo signal y included in this received signal is erased.

The above is the general operation of the echo canceller 301. Next, the operation of the echo canceller 301 described above will be described in detail with reference to FIG. FIG. 4 is a flowchart showing this conventional pseudo echo signal generation processing.

In this process, first, the adaptive transversal filter is initialized, that is, the initial value is set to the coefficient of each tap provided therein (S401). When the initialization of this filter coefficient is completed, the next transmission signal x (k)
Is input as the reference input signal x (k) (S402).
The input of this reference input signal x (k) is the input to the adaptive transversal filter, and
Indicates a reference input signal (transmission signal) x at time k (hereinafter, the same applies to other symbols).

When the input of the reference input signal x (k) is completed, the echo signal y (k), that is, the microphone 303, is next generated.
The received signal output by is input (S403). When the echo signal y is input, the already input reference input signal x (k)
And the impulse response h (n) (n is a number representing each tap) is used to execute a convolutional integration operation to obtain a pseudo echo signal y ′.
Is generated (S404). The generation of this pseudo echo signal y ′ is performed using, for example, the following Equation 1.

[0014]

[Equation 1]

When the pseudo echo signal y'is generated in this way, the generated pseudo echo signal y'is added to the adder 30.
The echo signal y is canceled by outputting the error e (k) to 4 and the error e (k) which is the difference between the echo signal y and the pseudo echo signal y'is calculated (S405). This error e (k)
Is calculated by the equation 2 when the noise is ignored (when considering the noise, the noise is calculated as Nz (k)
Then, this is an expression inserted by adding Nz (k) to the right side).

[0016]

## EQU00002 ## e (k) = y (k) -y '(k) When the process of step S405 is completed, the filter coefficient of the adaptive transversal filter, that is, the impulse response h (k) is calculated using, for example, Eq. Correct (S4
06), the process returns to step S402 and the next time (k + 1)
Move to the processing of. The correction of the impulse response h (k) is a correction for minimizing the above-mentioned error e (k), that is, for approaching to zero.

[0017]

[Equation 3]

[0018]

In the echo canceller, an LSI (Large Scale Integrat) is economically advantageous.
ed circuit), and general-purpose L as an LSI.
The application of SI to an echo canceller is drawing attention. This is because the general-purpose LSI does not newly develop itself, and the development cost of the software for executing the desired processing on the existing LSI is the main cost. This is because it has a great advantage over other types such as.

General-purpose LSI applied to echo canceller
Is MPU (microprocessor) or DS
P. Comparing the two, the MPU is superior to the DSP in the relatively complicated control functions such as memory management, judgment function, and interrupt function and the logical processing, but a powerful arithmetic unit such as a multiplier is built in like the DSP. Since it does not, the DSP is superior in the high speed of the arithmetic processing.

The echo canceller must perform processing in real time to cancel the echo signal.
However, in order to cover an echo delay amount of, for example, about 50 ms, the length of the adaptive transversal filter, that is, the total number of taps thereof is required to be about 400, and as shown in the above-mentioned formulas 1 and 3, Since there are many multiplications to generate a pseudo echo signal,
It is necessary to be able to perform arithmetic processing at high speed. For this reason,
As in the conventional example, a DSP is usually used in an echo canceller.

However, since the total number of taps of the adaptive transversal filter is large and many multiplications appear in the arithmetic expression, the echo canceller performs a huge amount of arithmetic processing at one time (each sample period). Therefore, research is being conducted to reduce the load of various processes including arithmetic processing.

In the echo canceller, an algorithm for generating a pseudo echo signal (L in the example of FIG. 4).
(Which is the MS algorithm) can be expressed by a mathematical formula, so that it is possible to grasp the approximate value of the amount of calculation executed at each sampling period. However, although this value is a powerful index, the time required to process the amount of calculation differs even if the amount of calculation is the same, depending on the basic hardware such as the arithmetic unit and memory provided in the DSP.

Recently, a DSP equipped with a floating-point type multiplier is also manufactured. However, since the hardware of this multiplier is heavier than that of the fixed-point type, a fixed-point type multiplier is used. It is expensive compared to the DSP equipped.

Since the echo signal is usually much smaller than the transmitted signal, the impulse response will be even smaller than the echo signal. Therefore, for example, when the arithmetic operation of the equation 1 is executed by using a 16-bit fixed point multiplier, if the transmission signal x is represented by 16 bits, 16 or more bits are usually required to represent the impulse response h. .

In the echo canceller, it is important to maintain the calculation accuracy in addition to performing the arithmetic processing at high speed.
In a 16-bit fixed-point type multiplier, even if the impulse response h can be represented by 16 bits, if the number of significant digits is small, the calculation precision will be low, and therefore the calculation precision must be maintained. Therefore, in practice, it is necessary to perform a bit operation or the like on the data representing one of the signals, and the operation speed is reduced by executing this operation, which causes a problem that the required operation speed is difficult to obtain. It will be. This is especially true for RISC (Reduced Instrument
ction set computer) is applied to the echo canceller, the problem becomes large.

An object of the present invention is to improve calculation speed while maintaining calculation accuracy.

[0027]

In the echo canceller of the present invention, first, the reference input signal generating means generates a reference input signal in which the transmission signal is changed, based on the transmission signal, the reception signal, and a predetermined algorithm. The pseudo echo generation means receives the reference input signal generated by the reference input signal generation means and executes a calculation process based on a predetermined algorithm to generate a pseudo echo signal.

In the above structure, the pseudo echo generating means is provided with various arithmetic units for performing arithmetic processing at high speed, and the arithmetic units can execute arithmetic processing based on the predetermined algorithm. desirable.

Further, the reference signal generating means calculates the transmission signal and the reception signal according to a predetermined mathematical formula, and based on the mathematical formula expressing a predetermined algorithm, the transmission signal and the received signal are used to calculate the transmission signal. Calculating means for calculating the change amount of the signal, and signal output means for changing the transmission signal according to the change amount calculated by the calculating means and outputting the reference input signal,
It is desirable to have.

[0030]

The echo canceller of the present invention generates a reference input signal in which the signal level of the transmission signal is changed based on the transmission signal for outputting voice or the like to the other party, the received signal received, and a predetermined algorithm. A pseudo echo signal is generated by executing arithmetic processing based on a predetermined algorithm using this reference input signal.

The arithmetic processing can be performed at high speed by using an arithmetic unit such as a multiplier. For changing the signal level of the transmission signal, for example, the ratio between the multiplier and the value of the multiplication result is set to 1
The accuracy of all the data handled by the multiplier is always kept high by making it close to. As a result, in the fixed-point type multiplier, execution of processing associated with operation processing such as bit operation for maintaining calculation accuracy is avoided, and operation speed is improved.

[0032]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 1 is a schematic block diagram showing a system configuration using an echo canceller 101 to which the present invention is applied. 101 is an echo canceller according to the present embodiment, 102 is a speaker, 103 is a microphone, 10
4 is an adder. The schematic operation will be described with reference to the configuration of FIG.

A transmission circuit (not shown) is particularly connected to the terminal a, and the speaker 102 outputs a sound by inputting the transmission signal x output from the transmission circuit. One of the microphones 103 outputs a reception signal obtained by picking up sound to the terminal b via the adder 104. Although not shown in particular, a receiving circuit for receiving a received signal is connected to the terminal b.

In this system, the sound emitted by the speaker 102 is picked up by the microphone 103 through the acoustic space, so that an echo is generated. The microphone 103 is
The echo signal y is output by collecting the sound emitted by the speaker 102.

The echo canceller 101, as shown in FIG. 1, calculates the attenuation amount β of the transmission signal x as will be described later, and the transmission signal x according to the attenuation amount β calculated by this arithmetic device 105. The preamplifier 106 for generating the reference input signal x ′ whose signal level is changed and the DSP 107 for inputting the reference input signal x ′ generated by the preamplifier 106 and generating the pseudo echo signal y ′.

The DSP 107 has an adaptive transversal filter (FIR filter) inside and has a fixed-point type multiplier (not shown) having a predetermined number of bits. Also, as will be described later in detail, the input reference input signal x ′ is given to this adaptive transversal filter, and convolutional integration is performed by using the impulse response h of the echo path between the speaker 102 and the microphone 103 that is updated sequentially. Generate an echo signal y '.

The generated pseudo echo signal y'has its signal level inverted and is output to the adder 104. When the pseudo echo signal y ′ is input, the adder 104 adds the pseudo echo signal y ′ to the reception signal output from the microphone 103, and subtracts the pseudo echo signal y ′ from the reception signal.

After outputting the pseudo echo signal y'to the adder 104, the DSP 107 calculates the signal level of the reception signal output to the terminal b, that is, the echo signal y (error e) remaining in the reception signal. . When this error e is calculated, the impulse response h is corrected by newly updating the tap coefficient from this error e.

The above is the general operation of the echo canceller 101. Next, the operation of the echo canceller 101 described above will be described in detail with reference to FIG. FIG. 2 is a flowchart showing the pseudo echo signal generation processing according to this embodiment.

When the power is turned on and the generation of the pseudo echo signal y'is instructed, first, the DSP 107 initializes the adaptive transversal filter, that is, sets the initial value to the coefficient of each tap provided therein (S201). ). When the initialization of this filter is completed, the reference input signal x ′ (k) is next input (S202), and the echo signal y is continuously input.
(K), that is, the received signal output from the microphone 103 is input (S203).

When the DSP 107 receives the echo signal y, the arithmetic unit 105 receives the transmission signal x as a reference input.
(K) and the received signal (echo signal y (k)) are input, and the energy values (signal level) of these two input signals
Are calculated (S204), and the attenuation amount β that attenuates the transmission signal x (k) is calculated (S205). The following Expression 4 is a calculation formula for calculating the above energy value, and Expression 5 is a calculation formula for calculating the attenuation amount β.

[0042]

## EQU00004 ## P1 (k) = (1-.rho.). Times.P1 (k-1) +. Rho..times.y (k) P2 (k) = (1-.rho.) * P2 (k-1) +. Rho.x (k) However, P1 (k): energy value of received signal (at time k) P2 (k): energy value of transmitted signal (at time k) ρ: predetermined constant

[0043]

## EQU5 ## β = P1 (k) / P2 (k) As can be seen from the equation 4, in this embodiment, the energy values P1 and P2 are set so as not to follow the abrupt changes of the input signals x and y.
Is calculated, it is possible to reduce the influence of noise and the like. When the calculation of the attenuation amount β is completed, the arithmetic unit 10
5 adjusts the preamplifier 106 according to the value of the attenuation amount β. As a result, the preamplifier 106 attenuates the transmission signal x according to this attenuation amount β and outputs it to the DSP 107.

By the way, the calculation (update) of the attenuation amount β is D
Since this is performed after the SP 107 has input the reference input signal x ′ (k), the reference input signal x ′ (k) input by the DSP 107 is the transmission signal x (k) according to the previously calculated attenuation amount β. ) Has been changed. The reason for doing this in this embodiment is that the adjustment of the preamplifier 106 takes a relatively long time and the sampling cycle (the time interval between the time k and the time k-1) is short. This is because the fluctuation of is usually small.

When the arithmetic unit 105 calculates the attenuation amount β,
Next, the DSP 107 applies the already-input reference input x ′ (k) to the adaptive transversal filter, and executes a convolution integration operation by using the reference input signal x ′ (k) and the impulse response h (n) to generate a pseudo echo. The signal y'is generated (S206). This pseudo echo signal y '
Is generated using Equation 6.

[0046]

[Equation 6]

When the pseudo echo signal y'is generated, the DSP
107 adds the generated pseudo echo signal y ′ to the adder 1
The echo signal y is canceled by outputting it to 04, and the error e (k), which is the difference between the echo signal y and the pseudo echo signal y ′, is calculated (S207). This error e
The calculation of (k) is performed by using the above-described Equation 2, the error e
When (k) is detected, the filter coefficient, that is, the impulse response h (k) is corrected using the error e (k) (S208), and the process returns to step S202 to return to the next time (k).
The process shifts to +1). This impulse response h (k)
Is corrected using the calculation formula shown in Expression 7.

[0048]

[Equation 7]

In this embodiment, as is apparent from the equations 4 and 5, the reference input signal x '(k), which is a signal after being attenuated by the attenuation amount β, is the echo signal y (k). And the same level (the ratio of the two signals is between 0 and 2). Therefore, when the arithmetic processing of the equation 6, that is, the processing of step S206 is executed, for example, the multiplier (fixed point system) is 16
When the bit is used, the impulse response h (n) can also be represented by 16 bits without performing processing such as bit manipulation, so that the calculation processing speed can be improved while maintaining the calculation accuracy.

In the flowchart shown in FIG. 2, the amount of calculation processing is concentrated on steps S206 and S208, that is, the generation of the pseudo echo signal y'and the filter coefficient h.
It takes time to correct (n). This is the total number of taps N
There is a tendency for it to increase.

However, in this embodiment, step S206
It is possible to shorten the processing time required for generating the pseudo echo signal y'of. Therefore, since one of the processes that requires a particularly long processing time is performed at high speed, the processing time can be greatly shortened, that is, the speed of the arithmetic processing can be greatly improved. This means that an inexpensive DSP equipped with a fixed-point type multiplier can be applied to the echo canceller, that is, the cost of the echo canceller can be suppressed. As a result of experiments conducted with the configuration and operation shown in this embodiment, it has been confirmed that good results can be obtained in improvement of calculation accuracy and calculation speed.

In this embodiment, the transmission signal x is changed according to the algorithm (mathematical expression) for generating the pseudo echo signal y ', but this change is simple as shown in FIG. Is done in. Therefore, in addition to the effect of reducing the cost of the echo canceller, the present invention can be easily applied to existing echo cancellers.

In this embodiment, the pseudo echo signal y '
Although the LMS algorithm is used for the generation of, the present invention is not limited to this algorithm and can be applied to various algorithms.

Although the present embodiment is applied to a system such as a videophone, the present invention is not limited to such a system and is similarly applied to other systems such as a telephone network. It is possible.

[0055]

As described above, the echo canceller of the present invention determines the signal level of the transmission signal based on the transmission signal for outputting voice or the like to the other party, the received reception signal, and a predetermined algorithm. Since the pseudo reference echo signal is generated by generating the modified reference input signal and executing the arithmetic processing based on the predetermined algorithm using this reference input signal, it is possible to improve the arithmetic speed while maintaining the calculation accuracy. it can.

[Brief description of drawings]

FIG. 1 is a schematic block diagram showing a system configuration to which an echo canceller according to this embodiment is applied.

FIG. 2 is a flowchart showing a pseudo echo signal generation process according to this embodiment.

FIG. 3 is a schematic block diagram showing a system configuration to which a conventional echo canceller is applied.

FIG. 4 is a flowchart showing a conventional pseudo echo signal generation process.

[Explanation of symbols]

 101 echo canceller 105 arithmetic unit 106 preamplifier 107 DSP

Claims (3)

[Claims] 1. A reference input signal generation means for generating a reference input signal obtained by changing the transmission signal based on a transmission signal, a reception signal, and a predetermined algorithm, and a reference input signal generated by the reference input signal generation means. And an echo canceller that generates a pseudo echo signal by executing a calculation process based on the predetermined algorithm. 2. The echo canceller according to claim 1, wherein the pseudo echo generating means includes various kinds of arithmetic units, and executes arithmetic processing based on the predetermined algorithm using the arithmetic units. 3. The reference signal generation means calculates the transmission signal and the reception signal according to a predetermined mathematical formula, and uses the calculated transmission signal and received signal based on the mathematical formula expressing the predetermined algorithm. Calculating means for calculating the change amount of the transmission signal, and signal output means for changing the transmission signal according to the change amount calculated by the calculating means and outputting the reference input signal. The echo canceller according to claim 1 or 2, which is characterized.