JP2551171B2 - Echo canceller - Google Patents

Description

TECHNICAL FIELD The present invention relates to an echo canceller.

[Conventional technology]

First, the basic concept of the echo canceller will be described with reference to FIG.

Consider a system in which a hybrid transformer 11 converts a 4-wire line 12 and a 2-wire line 13, and a terminal 14 such as a telephone is connected to the line 13. Hybrid transformer
Let t (t) be the received waveform on the 4th line side of 12, be s (t) be the transmitted waveform, and u (t) be the reverberant waveform. The echo canceller makes a predicted waveform (t) (of the echo waveform u (t)) from the received waveform r (t) by the network 15, and the subtracter 16 transmits the transmitted waveform s on which the echo waveform (t) is superimposed.
The predicted waveform (t) is subtracted from (t) + n (t) to cancel u (t) in the echo waveform. The network 15 is usually realized by a transversal filter. How to reasonably estimate the filter coefficient of the transversal filter, that is, accurately, rapidly, and stably, is the key to realizing the echo canceller.

The conventional echo canceller estimates the filter coefficient by using the maximum gradient method such as the learning identification method.

[Problems to be Solved by the Invention]

However, the maximum gradient method such as the learning fixed method may be attracted to the local peak or the local minimum. In addition, since this method accurately estimates the filter coefficient while using a long-time waveform, the conventional echo canceller
There is no guarantee that the optimum filter coefficient can be obtained, and there is a drawback that it takes a long time to estimate the filter coefficient.

An object of the present invention is to provide an echo canceller that can always extract the optimum filter coefficient and can estimate the filter coefficient in a short time.

[Means for solving the problem]

The echo canceller of the present invention is a echo canceller that cancels an echo signal by the output of a transversal filter having a received signal as input, and a first means for calculating an autocorrelation coefficient sequence of a received signal waveform, and the echo signal. Second means for calculating a cross-correlation coefficient sequence between the transmission signal waveform and the reception signal waveform on which the above waveform is superimposed, the auto-correlation coefficient sequence and the cross-correlation coefficient sequence calculated by the first and second means. And a third means for configuring the transversal filter by using a coefficient derived from the above as a filter coefficient.

The third means of the echo canceller includes a fourth means for storing each value of the cross-correlation coefficient sequence and a fifth means for searching the maximum value of each value stored by the fourth means. Means and
Sixth means for determining the filter coefficient of the transversal filter from the position of the maximum value searched for by the fifth means and the maximum value; and the maximum value searched for by the fifth means and the self-phase relationship. The stored contents of the fourth means are compensated using a sequence of numbers, and the fourth value is compensated by the compensated values.
And a seventh means for updating the stored contents of the means.

〔Example〕

 Next, the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing a first embodiment of the present invention.

In the embodiment shown in FIG. 1, an autocorrelation coefficient calculation unit 1 for calculating an autocorrelation coefficient sequence of a received waveform r (t) and an echo waveform u.
Input from the cross-correlation coefficient calculation unit 2 for calculating a cross-correlation coefficient sequence of the transmission waveform s (t) + n (t) and the reception waveform r (t) on which (t) is superimposed, and the cross-correlation coefficient calculation unit 2. Storage unit 3 for storing each value of the cross-correlation coefficient sequence, and a maximum value search coefficient determination for searching the maximum value of each value stored in the storage unit 3 and determining the coefficient from the searched maximum value and its position Part 4 and
The stored content of the storage unit 3 is compensated by using the maximum value retrieved by the maximum value retrieval coefficient determination unit 4 and the autocorrelation coefficient sequence input from the autocorrelation coefficient calculation unit 1, and the stored values of the storage unit 3 are compensated by the compensated values. The compensation waveform 5 for updating the stored contents and the received waveform r using the coefficient determined by the maximum value search coefficient determination unit 4 as a filter coefficient
The transversal filter 6 receives (t) and the subtractor 7 subtracts the predicted waveform (t) output from the transversal filter 6 from the waveform s (t) + u (t). The transversal filter 6 has n
This is a transversal filter with n + 1 taps, consisting of unit delay elements, n + 1 multipliers and one adder.

 The operation of the embodiment shown in FIG. 1 will be described.

Since u (t) is an echo waveform of the received waveform r (t), it is assumed that it can be expressed by a linear combination of r (t). This assumption may be correct from the standpoint of echo cancellation.
That is, u (t) is expressed by the following equation 1.

Waveform s that is an observed waveform in the cross-correlation coefficient calculation unit 2
Cross-correlation coefficient φ between (t) + u (t) and waveform r (t)
(J): j = 0,1,2, ... Is. Since the waveform r (t) and the waveform s (t) are uncorrelated, if m is set to be sufficiently large, it can be rewritten into the following three formulas.

If u in equation 3 is replaced by the expression in equation 1, However, R (j−k) is the received waveform r at the delay j−k.
It is an autocorrelation coefficient of (t).

When j = 0 to n Since R (l) = R (-1), equation 5 can be rewritten as equation 6.

Autocorrelation coefficients R (0) to R (n) and cross-correlation coefficient φ
Since (0) to φ (n) are obtained by the autocorrelation coefficient calculation unit 1 and the cross-correlation coefficient calculation unit 2, the six equations are b ₁ (i = 0
~ N), the transversal filter 6
The filter coefficient of the tap of delay i of is obtained as the solution b _i . However, if the connection time of the echo is long and n exceeds about 10, it becomes difficult to directly solve the simultaneous linear equations of equation 6, which is not realistic. The embodiment shown in FIG. 1 determines the filter coefficient based on the principle described below without directly solving equation (6).

When one filter coefficient is set at an arbitrary delay b of the transversal filter 6, the problem of determining the filter coefficient _h that is the optimum solution that minimizes the residual echo is as follows.
(S (t) and _h r (t−h) are uncorrelated, and s
Assuming that (t) and u (t) are uncorrelated, it results in the problem of minimizing the power p of the residual reverberation waveform u (t) -r (t-h).

As shown in Expression 9, the optimum filter coefficient _h at the delay h is proportional to the cross-correlation coefficient φ (h) at the delay h. Therefore, the most effective delay h ₀ for echo cancellation is
It is obtained as the delay h ₀ that maximizes | φ (h) |.

After determining the filter coefficients _h0 determine this delay h _0,
Next, the problem of determining the most effective one filter coefficient is h ₀ , _h0 as the problem of determining the most effective one filter coefficient for the residual echo of the echo cancellation by the filter coefficient _h0. You can treat it like any other problem you have done.

Waveform s obtained by adding the residual echo waveform to the transmission waveform s (t)
(T) + u (t) _−h0 r (t−h ₀ ) and received waveform r
The cross-correlation coefficient φ ₁ (j) with (t) is Residual echo waveform _{u (t) - h0 (t} -h 0) delay h ₁ the most effective in echo cancellation of | φ (j) | is obtained as j becomes maximum, even filter coefficients _h1 in the delay h ₁ It can be obtained by replacing φ (h) in Expression 9 with φ ₁ (h ₁ ).

By repeating the same procedure, the filter coefficient _h0 and _h1 are determined next, that is, the third effective filter coefficient _h2 and the fourth effective filter coefficient _h3 are determined one after another. can do.

Based on the principle described above, the embodiment shown in FIG. 1 determines the filter coefficient as described below.

Cross-correlation coefficient φ calculated by cross-correlation coefficient calculation unit 2
(0) to φ (n) are stored in the storage unit 3. The maximum value search coefficient determination unit 4 refers to the storage unit 3 to search for the maximum cross-correlation coefficient φ (h ₀ ) (in absolute value), determines the delay h ₀ from the storage position, and uses the formula 9 A calculation is performed to determine the filter coefficient _h0 . At this time, R obtained by the autocorrelation coefficient calculation unit 1
(0) is used.

When the determination of the filter coefficient _h0 is completed, the compensation unit 5
The determined filter coefficient b _h0 and autocorrelation coefficient calculation unit 1
Using the autocorrelation coefficients R (0) to R (n) obtained in
From the formula, the cross-correlation coefficient φ stored in the storage unit 3
(0) to φ (n) are compensated to obtain a cross-correlation coefficient φ ₁ (0) to
It is converted to φ ₁ (n), and the stored contents of the storage unit 3 are updated to the cross-correlation coefficients φ ₁ (0) to φ ₁ (n).

When the update of the storage unit 3 is completed, the maximum value search coefficient determination unit 4 determines a new filter coefficient by the same operation as the operation of determining the filter coefficient _h0 , and when this determination is completed, the compensation unit 5 causes the cross-correlation coefficient φ. The stored contents of the storage unit 3 are compensated and updated in the same manner as the compensation of (0) to φ (n). The same operation is repeated thereafter, and the maximum value search coefficient determination unit 4 determines filter coefficients one after another.

The transversal filter 6 is supplied with each filter coefficient from the maximum value search coefficient determination unit 4, and receives the received waveform r (t).
A predicted waveform (t) is created from Subtractor 16 has waveform s
The predicted waveform (t) is subtracted from (t) + u (t) to cancel the echo waveform u (t).

As described above, the embodiment shown in FIG. 1 operates in a stable manner because the filter coefficient is sequentially determined from the coefficient having the greatest effect of echo canceling.

In the case where the two-wire line 13 in FIG. 3 is short, the reverberation connection time may be short, and the number of taps n + 1 of the transversal filter 6 may be about 10 or less. In such a case, equation 6 can be solved directly.

FIG. 2 is a block diagram showing a reference example which adopts a method of directly solving equation 6.

The embodiment shown in FIG. 2 is a storage unit of the reference example shown in FIG.
3. The maximum value search coefficient determination unit 4 and the compensation unit 5 are replaced by the simultaneous equation solving unit 8.

The simultaneous equation solving unit 8 includes a matrix having the autocorrelation coefficients R (0) to R (n) input from the autocorrelation coefficient calculation unit 1 and the cross-correlation coefficient input from the cross-correlation coefficient calculation unit 2. Six equations, which are simultaneous linear equations having vectors having φ (0) to φ (n) as elements, are solved for b _{0 to} b _n , and the obtained solution is supplied to the transversal filter 6 as a filter coefficient.

Since the solution of Equation 6 is uniquely obtained (as long as Equation 6 can be solved), the reference example shown in FIG. 2 also operates stably.

Next, each block constituting FIGS. 1 and 2 will be described in detail with reference to the drawings. FIG. 4 is a block diagram for explaining the autocorrelation coefficient calculation unit 1 in detail. The autocorrelation coefficient calculator 1 shown in FIG. 4 includes a temporary memory 101, a window multiplier 102, a window coefficient memory 103, a blocked waveform memory 104, and a unit delay element 10.
5-0,105-1, ..., 105-m, unit delay element 106-0,106-1,
..., 106-m, multipliers 107-0, 107-1, ..., 107-m, accumulator
108, an autocorrelation memory 109, a timing generator 110, and switches 111A and 111B.

Received waveforms r (t), (t = ...,-1,0,1,2, ...) Are input to the temporary memory 101. The temporary memory 101 temporarily stores m + 1 consecutive waveform samples and outputs them for each frame. For convenience of explanation, it is assumed that the window hanging device 102 and the window coefficient memory 103 do not exist. At time t−h, m + 1 samples, that is, r (t−h−i), (i = −m, −m + 1,
..., -1, 0) is output to the blocked waveform memory 104.
The blocked waveform memory 104 stores r (t-h-i) and outputs it in the order of -m, -m + 1, ..., 0 in synchronization with the timing signal 110 supplied from the timing generator 110. . It is assumed that the switches 111A and B have the connections shown in the figure. The timing signal 1101 is composed of m + 1 pulses and is generated for each frame. The waveform r (t-h-i) output from the blocked waveform memory 104 is stored in the unit delay elements 105-0 to m, 106-0 to m. Of course r
(Th) is 105-0,106-0 and r (th-1) is 105.
−1, 106-1 and r (t−h−m) are stored in 105−m and 106−m.

Each of the multipliers 107-0 to m has a unit delay element 105-0 to a unit delay element 105-0.
Outputs of m, 106-0 to m are input. Multiplier 107-0 to m
Outputs the product of each of the two inputs to accumulator 108. The accumulator 108 is the sum of these products And outputs the result to the autocorrelation memory 109. The switching signal 1103 is generated by the timing generator for each frame and switches the switches 111A and 111B. The timing generator 110 is
It also outputs a timing signal 1102 composed of n + 1 pulses for each frame. The autocorrelation memory 109 takes in R (0) in synchronization with the first pulse of the timing signal 1102. The unit delay elements 106-0 to m are timing signals 1102.
To shift the contents. 106-1 to r (t-h-0),
106-2 to r (t-h-1), 106-m to r (t-hm)
+1) is stored. Data "0" is stored in 106-0. Therefore, the output R (1) of the accumulator 108 is Then, it is taken into the autocorrelation memory 109 in synchronization with the second pulse of the timing signal 1102. Similarly, R (n) is taken into the autocorrelation memory 109 in synchronization with the nth pulse of the timing signal 1102. Contents of autocorrelation memory 109 R (0)
(N) are output for each frame as the output of the autocorrelation coefficient calculation unit 1.

FIG. 5 shows the time relationship between the timing signals 1101 and 1102 and the switching signal 1103.

Of course, when the autocorrelation coefficients R (0) to R (n) are calculated, the effect of discontinuity of the waveform due to frames is easily mitigated by windowing. In FIG. 4, the window bracket 102 is used for this purpose. The window device 102 is r (t-h-
i) and (i = -m, -m + 1, ..., 0) are multiplied by the window coefficient w (i) (i = -m, ..., 0) and output. The window coefficient is, for example, a Hamming window coefficient and is supplied from the window coefficient memory 103.

FIG. 6 is a block diagram for explaining the cross-correlation coefficient calculation unit 2 in detail. The cross-correlation coefficient calculator shown in FIG. 6 includes temporary memories 201 and 211, blocked waveform memories 202 and 212,
n + 1 stage shift register 203, unit delay elements 204-0, 20
4-1, ..., 204-m, 214-0, 214-1, ..., 214-m, multiplier 205
-0,205-1, ..., 205-m, accumulator 206, cross-correlation memory 20
It is configured to include a timing generator 208 and a switch 209.

The received waveforms r (t) and (t = ..., -1,0,1, ...) Are stored in the temporary memory 201, and the residual echo waveform u (t) is added to the transmitted waveform s (t).
Waveforms s (t) + u (t), (t = ...,-1,0,1,
...) are respectively input to the temporary memory 211. Temporary memory 201
Are temporarily stored in m + n + 2, and the temporary memory 211 temporarily stores m + 1 consecutive waveform samples, respectively, and each frame is similar to the unit delay elements 105-0 to m, 106-0 to m in FIG.
Shift registers 203, 213, unit delay elements 204-0 to m, 214-
Output m + n + 2 samples from 0 to m. Now, for each of the unit delay elements 204-0 to 204-m, the received waveform sample r
(T-m), r (t-m + 1), ..., r (t-1), r (t)
Shift shift register 203 receives the received waveform sample r (tm-
n ,, r (t-m-n + 1), ..., R (t-m-1) are stored, and S (t- is stored in each of the unit delay elements 214-0 to 214-m.
m) + u (t-m), s (t-m + 1) + u (t-m +
1), ..., S (t) + u (t) are stored, and indeterminate data is stored in the shift register 213. Multiplier 205
-0 to m, the value supplied to cross-correlation memory 207 via accumulator 206 is obviously Is. Based on the same principle as in FIG. 4, φ (0) is taken into the cross-correlation memory 207, and at the same time, the unit delay elements 204-0 to 204-m and the contents of the shift register 203 are shifted. Unit delay element 204−
R (tm-1), r (tm), r (tm +) in 0 to m
1), ..., R (t-1) are stored, and the cross-correlation memory 20
To 7 Is supplied. Similarly, φ (1) to φ (n) are stored in the cross-correlation memory. FIG. 7 shows the time relationship between the timing signals 2081 and 2082 and the switching signal 2083.

FIG. 8 is a block diagram for explaining the maximum value search coefficient determination unit 4 and the compensation unit 5 in detail. The maximum value search coefficient determination unit 4 shown in FIG. 8 includes an address generator 401, a RAM 402, an absolute value calculator 403, an absolute value calculator 404, a comparator 405, a latch 406, a latch 407, a divider 408, an adder 409. , RAM410, address generator 4
11, timing generator 412, latch 413, unit delay element 414-0
-N, latches 415-0 to n, a switch 416, a switch 417, and a controller 418. The compensator 5 includes an address generator 501, a RAM 502, a microprocessor 503 and a controller 504.
Is configured.

The operations of the maximum value search coefficient determining unit 4 and the compensating unit 5 are controlled by the built-in controllers 418 and 504, respectively. The address generator 501 has its contents set to "2n + 1". Autocorrelation memory included in autocorrelation coefficient calculation unit 1
The data read from 109 in the order of R (n), R (n-1), ..., R (1), R (0) is decremented in synchronization with the reading of the data. Depending on the address data, the RAM 502 “via the microprocessor 503
It is written from address 2n + 1 "to address" n + 1 ". The autocorrelation memory 109 then outputs R (0) and continues to supply it to the divider 408. Both the address generator 501 and the address generator 401 have their contents. Is set to “n.” From the cross-correlation memory 207 included in the cross-correlation coefficient calculation unit 2, φ (n), φ (n
The data read in the order of -1), ..., φ (1), φ (0) are decremented in synchronization with the reading of the data.
It is written from "n" address of AM502, RAM402 to "0" address. The switches 416 and 417 are off. Data "-a" is supplied to one input terminal of the comparator 405. "A"
Is any positive value. Data “n” in address generator 401
Is loaded, and φ (n) is read from the RAM 402. The read φ (n) is supplied to the absolute value calculator 403 and the latch 406. Absolute value calculator 403 calculates | φ (n) | and outputs it to comparator 405. Since −a <| φ (n) |, the comparator 405 outputs a pulse to latch the input data of the latches 406 and 407. The latch 406 stores φ (n), and the latch 406 stores “n”. The switch 416 is turned on. Address generator 401 is decremented and RAM402
.Phi. (N-1) is read from. Absolute value calculator 403
φ (n−1) | is calculated and output to the comparator 405. Latch 4
The data φ (n) stored in 06 is supplied to the absolute value calculator 404. The absolute value calculator 404 calculates | φ (n) | and outputs it to the comparator 405 via the switch 416. The comparator 405 compares two inputs | φ (n) | and | φ (n-1) |
Only when | φ (n-1) |> | φ (n) |, a pulse is output to latch the input data of the latches 406 and 407. Therefore, max {| φ (n) |, | φ (n
-1) |} corresponds to the latch 407 by "n" or "n-"
1 "is stored. The address generator 401 is sequentially decremented to" 0 ", and the latch 406 stores max {| φ (n) |,
| φ (n−1) |, ..., | φ (0) |}, and one of the corresponding “n” to “0” is stored in the latch 407.

| Φ (h ₀ ) | = max {| φ (n) |, | φ (n-1) |, ..., | φ (0) |} (15) represents “h ₀ ” from the latch 407. , Φ (h ₀ ) is the latch
It will be output from 406. φ (h ₀ ) is input to the divider 408. The divider 408 calculates _h0 based on the equation (9) and outputs it to the adder 409 and the microprocessor 503.

The RAM 410 has addresses “0” to “n”, and data “0” is written in advance in all of these addresses. Latch 40
The data “0” stored in the RAM 410 is output by “h ₀ ” supplied from 7 and supplied to the latch 413. The latch 413 latches this data and outputs it to the adder 409. Adder 409
Is the addition result of these two inputs φ (h ₀ ) and “0”, φ
Output (h ₀ ) to RAM410. RAM410 is φ to "h _0" address (h
₀ ) is memorized.

Based on the data "h ₀ " input to the microprocessor 503, the controller 504 sets the address generator 501 to "h ₀ ". Data φ stored at address “h ₀ ” of RAM502
(H ₀ ) is output to the microprocessor 503. Controller 5
04 sets the address generator 501 to "n + 1". RAM502
The data R (0) stored in the address "n + 1" is output to the microprocessor 503. Microprocessor 5
03 assignment expression _{φ (h 0) = φ (} h 0) - h0 * Run the R (0) (16), operates to write the result to the address "h _0" of RAM502. Next, data φ (h ₀ +1), φ (h ₀ −
1) and R (1) are read and the microprocessor 503
Supplied to Microprocessor 503 is assignment type And executes the operation to write the result to the addresses “h ₀ +1” and “h ₀ −1” of the RAM 502. The same is performed thereafter. This makes RA
It is self-evident that the contents of M502 are replaced with the result of Equation 10 above.

Subsequently, the data in the addresses "0" to "n" of the RAM 502 are transferred to the addresses "0" to "n" of the RAM 402. The above operation is repeated, and the maximum value search is performed up to the desired number of times. This allows RAM410
The data stored in the addresses "0" to "n" of the above means b _{0 to} b _n . The switch 417 is turned on. The address generator 411 outputs address data “0” to “n” to the RAM 410 based on the timing signal generated by the timing generator 412. The data b ₀ , b ₁ , ..., B _n read from the RAM 410 are supplied to the unit delay elements 414- _{n- 0} and shifted. After completion of the shift, b ₀ , b ₁ , ..., B _n latched by the latches 415-0 to _n are output to the transversal filter 6.

It is obvious that the simultaneous equation solving section 8 shown in FIG. 2 can be realized by a microprocessor.

〔The invention's effect〕

As described above, the present invention derives the filter coefficient of the transversal fill from the cross-correlation coefficient sequence of the transmission signal waveform and the reception signal waveform on which the echo signal waveform is superimposed and the autocorrelation coefficient sequence of the reception signal waveform, There is an effect that it is possible to provide an echo canceller that operates stably, and there is an effect that the filter coefficients can be estimated in a short time because all the filter coefficients can be estimated by the waveform of the block of one window.

[Brief description of drawings]

FIG. 1 is a block diagram showing the first embodiment of the present invention, and FIG.
FIG. 4 is a block diagram showing a reference example, FIG. 3 is an explanatory diagram for explaining the basic idea of the echo canceller, FIG. 4 is a detailed block diagram of an autocorrelation coefficient calculation unit, and FIG. FIG. 6 is a waveform diagram for explaining the operation of the configuration of FIG. 6, FIG. 6 is a detailed block diagram of the cross-correlation coefficient calculation unit, FIG. 7 is a waveform diagram for explaining the operation of the configuration of FIG. 6, and FIG. FIG. 3 is a detailed block diagram of a maximum value search coefficient determination unit and a compensation unit. 1 ... Auto-correlation coefficient calculation unit, 2 ... Cross-correlation coefficient calculation unit, 3 ... Storage unit, 4 ... Maximum value search coefficient determination unit, 5 ...
... Compensator, 6 ... Transversal filter, 7 ... Subtractor, 8 ... Simultaneous equation solver.

Claims (1)

(57) [Claims] 1. An echo canceller for canceling an echo signal by the output of a transversal filter having a received signal as an input, which calculates an autocorrelation coefficient sequence of a received signal waveform.
Second means for calculating a cross-correlation coefficient sequence between a transmission signal waveform on which the echo signal waveform is superimposed and the reception signal waveform
Means, and third means for configuring the transversal filter by using coefficients derived from the autocorrelation coefficient sequence and the cross-correlation coefficient sequence calculated by the first and second means as filter coefficients. And a fourth means for storing each value of the cross-correlation coefficient sequence, and a fifth means for searching the maximum value of each value stored by the fourth means. Means for determining the filter coefficient of the transversal filter from the position of the maximum value and the maximum value retrieved by the fifth means, and the maximum value and the self-phase retrieved by the fifth means And a seventh means for compensating the stored content of the fourth means by using a relational sequence and updating the fourth stored content with each compensated value.