JPS6255739B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to an adaptive echo canceller that cancels echoes while estimating the transmission characteristics of an echo path using sequentially received signals and echo signals. Echo in long-distance telephone lines such as satellite lines causes a significant deterioration in call quality. Although echo suppressors currently in use can effectively suppress echoes, they have the disadvantage that they cannot in principle avoid deterioration in speech quality, such as speech disconnection and echo leakage during superimposed speech. For this reason, echo cancellers are attracting attention as a new echo control device. The principle of an echo canceller is to estimate the transmission characteristics of the echo path from the received signal and the echo signal, generate a pseudo echo signal based on the results, and cancel the echo by subtracting it from the true echo signal. be. Conventionally, the main algorithm for estimating the transmission characteristics of the echo path in an echo canceller has been based on the learning identification method, but when using audio signals as input signals, this method is based on the strong correlation of the audio signals. Therefore, the convergence time was longer and the amount of echo cancellation was insufficient compared to when white noise was input. In order to eliminate the above-mentioned drawbacks, the present inventors used received signals for each predetermined time length to find the optimal autoregressive coefficient in terms of the squared error when the received signal is used as the output of an autoregressive model. A difference signal between the predicted value of the received signal using the autoregressive coefficient and the received signal, and a difference signal between the predicted value of the echo signal using the autoregressive coefficient and the echo signal are calculated, and a difference signal between the received signals is calculated. The transmission characteristics of the echo path are sequentially estimated using the difference signal between the
It is configured so that the echo can be canceled by subtracting it from the true echo signal, and by appropriately selecting the order of the autoregressive model, it is possible to whiten the difference signal of the received signal, and the convergence We proposed an echo control method that takes less time and has an extremely large amount of echo cancellation compared to conventional methods. method”). These methods require inserting a register or regression filter into the main path, or newly convolving the received signal itself with estimated transmission characteristics in order to create a pseudo echo. The method of inserting a register into the main path has the disadvantage that delay time increases because the register is inserted into the main path. Another drawback was that when the echo canceller failed, the main path was disconnected, making communication impossible. Further, in the embodiment shown in FIG. 1 of Japanese Patent Application No. 53-165196 entitled "Echo Control System", a regression filter is required on the main path in order to prevent disturbances in the sound from the near end. This results in the same drawbacks as mentioned above. On the other hand, in the embodiment shown in FIG. 5 of Japanese Patent Application No. 53-165196 "Echo Control Method", a new method of creating a pseudo echo by convolution processing of the received signal itself and estimated transmission characteristics requires an increase in the amount of hardware. I had a lot of flaws. An object of the present invention is to provide an echo control method having characteristics comparable to those of the prior application method without significantly increasing the amount of hardware by using a regression filter inside the echo canceller instead of in the main path. . First, the principle of the present invention will be explained. Now the time jT
_{The estimated transmission characteristics} and _{received signal at} , x _jN )' (2) and the echo signal is y _j , the residual signals of the received signal and echo signal are each It can be expressed as Here, a ₁ , a ₂ , ..., a _M is the received signal x = (x ₁ ,
x ₂ , ..., x _L ). The algorithm is, for example, Durbin's method [Reference. Durbin, J.
(1960) The Fitting of time-series models,
According to Rev.Inst.Stat., 28, PP.233-244]. i.e. Then, the initial condition in a _j ⁽ⁱ⁾ = a _j ^(i-1) −a _i ⁽ⁱ⁾ a _ij ^(i-1) j=1, 2, ..., i-1 (8) E ⁽ⁱ⁾ = E ^{(i- 1)} (1-[a _i ⁽ⁱ⁾ ] ² ) i=2, 3, ..., M (9) It is found recursively. The algorithm for changing the estimated transmission characteristic 〓 _j is It is. The method for obtaining the pseudo echo ^y _j is ^y _j =〓 _j ′x _j (11) in the earlier application, Japanese Patent Application No. 165196/1983 titled ``Echo Control Method''. In contrast, in the present invention, the pseudo echo ^y _j is created by the following method. In other words, if we set the transmission characteristics of the echo path as 〓 _j = (h _1j , h _2j , ..., h _Nj )' (12), then from the linearity Therefore, the pseudo echo ^y _j is becomes. Here, if ＾y _j-1 〓y _j-1 , 〓 _j ′〓 _j
has already been created as shown in equation (10), and generally M is sufficiently smaller than N, so Therefore, a pseudo echo can be created with almost no increase in the amount of calculation. The above is the main principle of the present invention. In addition

[Equation] is realized by a regression filter, so in order to maintain stability, a _i ⁽ⁱ⁾ (where i = 1, 2, ..., M) in Equation (7) must be -1<a _i ⁽ⁱ⁾ <1
However, as is clear from equations (7) to (9), this can be detected during the process of calculating the prediction coefficient a _i ⁽ⁱ⁾ . Note that when j becomes sufficiently large and 〓 _j becomes sufficiently equal to 〓 _j , y _j-1 = y _j-1 approximately holds true, but when j is small, ^ y _j-1 = y _j-1 Since this does not hold true, the characteristics deteriorate. However, as shown in the example below, by setting the prediction coefficient a _i to zero every predetermined number of times, y _j =〓 _j ′〓 _j (16) from equation (14) in that part, Characteristics can be improved. In addition, the following methods are also available as other means. In general, audio signals are broadly divided into voiced and unvoiced sounds, and when unvoiced sounds are input, the estimated transmission characteristics improve rapidly, but when voiced sounds are input, the estimated transmission characteristics hardly improve. It has become clear [for example, literature, Yamamoto, S.,
Kitayama, S., Tamura, J. and Ishigami, H.:
“An Adaptive Echo Canceller with Linear
Predictor”, Trans.IECE Japan, E62.12,
pp.851-857 (Dec.1979)]. On the other hand, the power of voiced sounds is much larger than that of voiceless sounds. Therefore, in the case of unvoiced sound input, using the obtained prediction coefficients, In the case of voiced sound input, the prediction coefficient is set to zero and the pseudo echo is determined as ^y _j =〓′ _j ′〓 _j (18). In this case, voiced sounds and unvoiced sounds can be distinguished by comparing the value of a ₁ (1) with a predetermined value for every L samples in equation (5). Moreover, such a degree of differentiation shows sufficient characteristics. Hereinafter, the present invention will be described in detail with reference to the drawings, but for ease of explanation, the difference signals between various signals and their predicted values will be referred to as residual signals. FIG. 1 shows a block diagram of an embodiment of an echo canceller using the present invention. 1 is an echo canceller, 2 is a receiving side input terminal, 3 is a transmitting side output terminal, 4 is a receiving side output terminal, 5 is a transmitting side input terminal,
6 is a hybrid coil, 7 is a terminal device such as a telephone, 8 is a register, 9 is a regression coefficient calculator, 10 is a predictor, 11 is a subtracter, 12 is a register, 13 is a convolution operator, 14 is a predictor, 15 and 16 are subtracters, 17 is a corrector, 18 is a predictor, 19 is an adder, and 20 is a subtracter. In order to simplify the explanation, it is assumed that the signals within the echo canceller 1 are digitized, and that, although not shown in FIG. 1, a clock is naturally supplied to each section. To explain the operation with reference to FIG.
However, a part of the received signal passes through the hybrid coil 6 and enters the transmitting side input terminal 5 as an echo. On the other hand, inside the echo canceller 1, the received signal x _j is once accumulated in the register 8, sent to the regression coefficient calculator 9, and also sent to the predictor 10. The regression coefficient calculator 9 calculates the received signal X _j within a predetermined time = (x _j-1 , x _j-(L-1) , ..., x _j-
1 ) to find regression coefficients a=a ₁ , a ₂ , ..., a _M. The algorithm is as described above. Since the regression coefficient calculator 9 may have a slower processing speed than other parts, it can be configured mainly using a microprocessor. Note that the regression coefficients a ₁ , a ₂ ,...
..., a _M are a ₁ ^(M) , a ₂ ^(M) , ... in equations (7), (8), and (9), respectively.
..., corresponds to a _M ^(M) . Regression coefficient a= calculated by regression coefficient calculator 9
(a ₁ , a ₂ , . . . , a _M ) are transferred to predictors 10 and 14 . The predictor 10 uses the regression coefficient a and the received signal sent through the register 8 to predict the received signal x _j at time j, as follows:

Create [formula]. The configuration of the predictor 10 is as shown in FIG. 2, and FIG. 2 shows the case where M=5.
101, 102, 103, 104, 105 are delay elements, 111, 112, 113, 114, 115
is a multiplier, and 120 is an adder. The output x _j of the predictor 10 is transferred to the subtractor 11 . The subtracter 11 creates a residual signal x _j =x _j -^x _j from the received signal x _j transferred from the register 8 and the output ^x _j of the predictor 10, and transfers it to the convolution calculator 13. Ru. Residual signal sequence〓 _j = (^x _j-1 , x _j-
2 ,...,x _jN ) is the signal in register 12 ==
(h ₁ , h ₂ , ..., h _N ) and convolution operation in the convolution operator 13

[Formula] is performed and the calculation result

[Formula] is transferred to the subtracter 16. On the other hand, the echo signal y _j input from the transmission side input terminal 5 is transmitted to the predictor 14 and the subtracters 15 and 20.
sent to. The predictor 14 uses the regression coefficient 〓 transferred from the regression coefficient calculator 9 and the echo signal y _j as the predicted value ^y _j of the echo signal at time j.

The subtracter 15 creates a residual signal y _j =y _j -^y _j from the echo signal y _j and the output ^y _j of the predictor 14, and transfers it to the subtracter 16. The subtracter 16 generates an error signal e _j =y _j −y _j from the residual signal y _j from the subtracter 15 and the signal y _j from the convolution calculator 13, and transfers it to the corrector 17. In the corrector 17, the error signal e _j
=y _j −y _j =y _j −〓 _j ′〓 _j and the signal sequence from the subtracter 11〓 _j = (x _j-1 , x _j-2 , ..., x _jN ) using equation (10) The value 〓 _j = (h ₁ , h ₂ , . . . , h _N ) in the register 12 is modified according to the algorithm shown in FIG. In formula (10), 〓 _j represents the value in register 12 before being modified, and 〓 _j+1 represents the value in register 12 after being modified.
Shows the value within. On the other hand, the output of the convolution operator 13 transferred to the adder 19 is added to the output of the predictor 18,
As a result, a pseudo echo ^y _j is created, and the subtractor 20
and is transferred to the predictor 18. The predictor 18 uses the regression coefficient 〓 transferred from the regression coefficient calculator 9 and the signal ^y _j from the adder 19,

Create [formula]. On the other hand, the subtracter 20 subtracts the output y _j of the adder 19 from the echo signal y _j and sends out the error e _j =y _j -^y _j through the transmitter output terminal 3. In this case, if the value 〓 _j in the register 12 becomes the same as the transmission characteristic of the echo path, the error e becomes zero and the echo to the speaker disappears. In addition, at each predetermined number of times or a ₁ (1) of formula (7)
is not within a predetermined value range, it is extremely easy to set the regression coefficient a to zero, and in this case, the pseudo echo ^y _j becomes as shown in equation (16). Now that the embodiments of the present invention have been described above, one example of the regression coefficient calculator 9 will be described in detail. FIG. 3a is a diagram of a configuration example of the regression coefficient calculator 9. In Fig. 3, 201 is a serial-parallel converter, 2
02 and 203 are registers, 204 is a flip-flop, 205 is an adder, 206 is a multiplier, 207
is a gate, 208 is an accumulator, 209 is a constant multiplier,
210 is a register, 211 is a flip-flop,
212, 213, and 214 are counters, 216 is a gate, 217 is a memory, and 218 is a microprocessor. To explain the operation according to the illustration,
Received signal x _j input from register 8 in FIG.
is transferred to the serial-parallel converter 201,
When L signals are accumulated in serial-parallel converter 201, they are transferred to registers 202, 203 and flip-flop 204. register 20
2, 203 and flip-flop 204 circulate in synchronization, multiplier 206 calculates x _j x _j , and the obtained signal is transferred to accumulator 208.
In the accumulator 208, the signal x _j x _j from the multiplier 206
cumulative value of

Find [formula]. Counter 212 outputs a signal every L clocks and sends the contents of accumulator 208 to constant multiplier 209. In the constant multiplier 209, the value from the accumulator 208

[Formula] is multiplied by a constant (1/L) and sent to the register 210.
The signal from counter 212 is passed through the interrupt line of microprocessor 216 to instruct microprocessor 216 to read the signal in register 210. On the other hand, the microprocessor 216 uses the register 21 in normal computer operation.
0 to a specific address in the signal memory 217. On the other hand, since the counter 213 outputs a signal every (L+1) clocks, the flip-flop 211
remains in the state "1" for every L clocks 1, 2, 3, . . . , as shown in FIG. 3b, during which time the gate 207 is closed. In FIG. 3B, 300 is a clock, 301 is the output of the counter 212, 302 is the output of the counter 213, and 303 is the state of the flip-flop 211. Therefore, the signal in register 203 is displaced by one with respect to the signal in register 202 every L clocks, as shown in FIG. 3c. 400 in Figure 3c
is the signal in register 202, 401, 402, 4
03 are the signals in register 203 and flip-flop 204 after the L, 2L, and 3L clocks, respectively. As a result, as before, register 210
sequentially, the signal

【formula】

[Expression] is transferred. When counter 2 14 counts M signals from counter 212,
Gate 216 is closed so that the M signals in register 210 are captured by microprocessor 216 into memory 217. Microprocessor 216 registers 210
Using the data in the memory 217 taken in from a ₁ ^(M) , a ₂ ^(M) ,…
..., a _M ^(M) and transfer it to the predictors 10, 14 and 16. As described above, the present invention can greatly improve the convergence characteristics of the echo canceller compared to conventional systems. Further, unlike the embodiment shown in FIG. 5 of the earlier application, Japanese Patent Application No. 53-165196 (echo control method), there is an advantage that there is almost no increase in the amount of calculation. Further, unlike the embodiment shown in FIG. 1 and Japanese Patent Application No. 53-57129 (echo control system), nothing is inserted into the main path, so there is an advantage that no delay or failure problems occur.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an echo canceller according to the present invention, FIG. 2 is a detailed configuration example diagram of a predictor used in the present invention, FIG. 3a is a detailed configuration example diagram of a regression coefficient calculator used in the present invention, Figure 3b,c is Figure 3a
FIG. 2 is a diagram for explaining the operation of a predictor.

Claims (1)

[Claims] 1. In an adaptive echo canceller that cancels echo while estimating the transmission characteristics of an echo path using successive received signals and echo signals, the received signal at each predetermined time length is processed by autoregressive processing having a constant autoregressive coefficient. a device for determining the autoregressive coefficient using a predetermined number of received signals within the time length, which are regarded as output signals of the model; a device for determining a difference signal between the predicted value of the echo signal using the autoregressive coefficient and a second difference signal between the echo signal;
a storage device for storing estimated transmission characteristics of an echo path; a corrector for modifying the contents of the storage device; a device for convoluting the output of the storage device and the first difference signal; a device for calculating a predicted value of the pseudo echo signal from the echo signal, using a third difference signal between the second difference signal and the output of the convolution calculation device and the first difference signal. sequentially correcting the contents of the storage device by the corrector, and creating the pseudo echo signal from the third difference signal corrected using the sequentially corrected estimated transmission characteristics and the predicted value of the pseudo echo signal; An echo control method characterized by canceling echo by subtracting it from the true echo signal.