RU2407148C1 - Adaptive echo-canceller on m-th order recursive filter - Google Patents

Abstract

FIELD: information technology.

SUBSTANCE: adaptive echo-canceller (AEC) consists of N≤2 channel filters (CF) (1), an N-tap delay line (DL) (2), and M-input adder (3), a subtracting unit (SU) (4) and an adaptive control unit (ACU) (5). Circuits for transmission of forward and reverse signals (echo) are created in the AEC. The predicted value of the signal level is calculated, from which by steepest descent, correcting weight coefficients are calculated for deeper suppression of echo signals.

EFFECT: deeper suppression of echo signals in the entire speech spectrum.

4 dwg

Description

The invention relates to the field of telecommunications and can be used in telecommunication lines of two-way communication.

Adaptive echo cancellers are known, described, for example, in EP No. 0412691 A2 of 02.13.91 with priority of 07.08.89; U.S. Patent No. 4,964,118 of October 16, 1990.

Adaptive echo canceller according to the application EP No. 0412691 consists of a first block for dividing the primary signal into a set of signals divided within the spectrum of the primary signal, and a second block that performs, similarly to the first block, a function with a secondary signal. The echo canceller also includes many adaptive filters designed for primary and secondary signals and providing the generation of an error signal.

The disadvantage of this filter is the relatively high level of uncompensated echo.

Adaptive echo canceller according to US Pat. US No. 4964118 consists of the first and second channel filters, providing suppression of the echo signal due to the generation of compensating signals. The echo canceller generates a compensating signal in accordance with a given function, which allows to reduce it exponentially.

The disadvantage of this analogue is the relatively high level of uncompensated echo.

The closest in technical essence to the claimed is the adaptive echo canceller according to ed. St. USSR No. 1446277, IPC Н04В 3/20, publ. 11/7/1988, bull. No. 41. The closest analogue consists of M≥2 channel filters (CF), N-tap delay line (NO LZ), adder, amplifier, subtraction unit. Each CF contains the first and second multipliers, the channel adder (CS), the channel integrator and the channel N-tap LZ.

After processing and comparing it with the residual signal in the closest analogue, the far-end subscriber generates an evaluation signal of the impulse response of the echo path and generates an estimate of the echo signal based on the current estimate of the cross-correlation function of the generated signals taking into account the values of the impulse response of the echo path.

However, the prototype has disadvantages:

insufficient level of suppression of the echo signal, which is due to the linear dependence of the feedback on the error signal and a fixed level of limitation of the amplitude of the generated signal.

The technical result from the use of the claimed device is the development of an adaptive echo canceller that provides deeper suppression of the echo signal in the entire range of the speech spectrum due to more accurate convergence of the iterative process of compensating the echo signal.

The claimed device extends the arsenal of funds for this purpose. This goal is achieved by the fact that in the known adaptive echo canceller (AEC) containing M≥2 KF, N-tap LZ, where N = M-1, the input of which is the input "distant subscriber signal" (SDA) AEC, M-input adder (MVkh.SUM), the output of which is connected to the output of the subtraction unit (BV), the second input and the output of which are respectively the input "reflected signal" (In. OTs) and the output "error signal" (output SO), n-th tap NO LZ, where n = 1, 2, ..., N, is connected to the signal input of the (n + 1) -th KF, and the signal input of the first KF is connected to the input NO LZ, the signal output of the m-th KF, where m = 1, 2, ..., M, is connected to the mth input of MVx. SUM, an adaptive control unit (BAU) is additionally introduced. The BAU is equipped with an M × M-bit signal input (M × MR SVx.), M-bit first output “signal of modeling coefficients” (1MP Output QMS), M × N-bit second output “signal of modeling coefficients” (2M × MR Output QMS), M-bit output "correction signal" (MR Output QMS), M-bit input "training signal" (MP In. ObS). The M-bit signal output (MR output) of the m-th KF is connected to the m-th M-bit signal input of the BAU, and the m-th N-bit output is 2M × NP Out. The QMS is connected to the second N-bit input “signal of modeling coefficients” (MR In. QMS) of the m-th KF. The output “training signal” (Output OBS), the input “correction signal” (In. SC) and the first input “signal of modeling coefficients” (1 In. QMS) of the m-th KF are connected to the m-th digits, respectively, MP In. OBS, MR Out. SC and 1MP Out. QMS BAU. The BAU is equipped with an additional input “error signal” (Vkh. СО), which is connected to the Vykh. WITH BV, which is the output of AEC.

CF consists of the first and second M-input adders (MVx. SUM), the first, second and third multipliers (UM), the first and second adders (SUM), the first and second groups of multipliers (GUM) with N multipliers in each group, the first and a second N-tap delay line (NO LZ). The signal input of the first PA is the signal input KF. The first UM, the first and second MVh. SUM, the first and second SUM are cascaded along the signal outputs. The output of the second SUM is connected to the input of the first NO LZ and is the signal output of the CF.

The outputs of the n-th multipliers belonging to the first and second GUM, respectively, are connected to the m-th inputs of the first and second MVx, respectively. SUM. The outputs of the multipliers belonging to the first GUM are inverse. Moreover, the nth taps of the second NO LZ and the first NO LZ are connected to the first inputs of the n-th multipliers belonging respectively to the first GUM and the second GUM. The first inputs of the PA belonging to the first GUM, together with the input of the second NO LZ, are MRSv. CF The second inputs of the nth PA of the first GUM and the second GUM are combined and are the nth bit 2NPBx. QMS KF. The output of the second PA is connected to the first input of the third PA and to the second input of the second SMS. The second input of the first SUM is connected to the inverse output of the third PA, the second input of which is connected to the second input of the first PA and is 1 Vx. QMS KF. The output of the first NO LP is connected to the first input of the first PA, the second input of which is Bx. SC CF.

BAU consists of the first and second weight shapers (FVK), the first and second weight corrector (KVK), the feedback shaper (FSOS). M × M-bit FPS inputs are M × M CBx. BAU. M × M-bit input FPS is connected to the M × M-bit input of the first FVC. The FPS output is connected to the first input of the first FAC, the output of which is connected to the “correction” input of the first FVC. The first M-bit and second M × N-bit outputs of the "signal modeling coefficients" of the first FVC are respectively 1MP Output. QMS and 2M × NP Out. QMS BAU. MPOS FSOS input is connected to the M-bit input of the second FVK, the M-bit output of which is MR Output. SK BAU. The output of the FSOS is connected to the first input of the second KVK, the output of which is connected to the input "correction" of the second FVK. The second inputs of the first and second KVK are combined and are Vkh. SB BAU.

Thanks to this new set of essential features, the claimed device provides adaptive regulation of weight coefficients in real time, which in turn leads to a more accurate formation of the compensating signal and, therefore, a deeper suppression of the echo signal.

The claimed device is illustrated by drawings, which show:

figure 1 is a General block diagram of an adaptive echo canceller;

figure 2 is a structural diagram of a channel filter;

figure 3 is a structural diagram of an adaptive control unit;

figure 4 is a block diagram of the algorithm of the BAU.

AEC on the recursive filter of the Mth order, shown in figure 1, consists of M≥2 KF 1 ₁ , 1 ₂ , ..., 1 _m , NO LZ 2, MVx. SUM 3, BV 4 and BAU 5. The input NO LZ 2 is connected to the information input of the first (m = 1) KF ₁ and is Vkh. SDA device. The information input of the (n + 1) -th KF 1 _{n + 1 is} connected to the n-th tap of NO LZ 2. The signal output of the m-th KF 1 _{m is} connected to the m-th input of M In. SUM 3. Exit M In. SUM 3 is connected to the first input of the BV 4, the second input and output of which are respectively Bx. OTs and Vykh. SO AEC. The output of the BV 4 is also connected to the input. SB BAU 5. MR signal output of the m-th KF 1 _{m is} connected to the m-th MR input M × MR CBx. BAU 5. Exit. OBS, Vkh. SK and 1 Vkh. The QMS of the m-th CF 1 _{m are} connected to the m-th discharge, respectively, MR In. OBS, MR Out. SC and 1MP Out. QMS BAU 5. 2NP In QMS of the m-th KF 1 _m connected to the m-th NP Out. QMS 2M × NP Out. QMS BAU 5.

Channel filters 1 ₁ -1 _m are designed to simulate a pre-emptive speech echo signal.

CF 1, shown in figure 2, consists of the first 1.3 and the second 1.4 MVX. SUM, the first 1.2, the second 1.10 and the third 1.9 SUM, the first 1.5 and the second 1.6 SUM, the first 1.7 and the second 1.8 GUM of M multipliers in each group, the first 1.11 and the second 1.1 NO LZ. The signal input of the first PA 1.2 is the signal input of CF 1. The first PA 1.2, the first 1.3 and the second 1.4 M Vh. SUM, the first 1.5 and the second 1.6 SUM are cascaded along the signal outputs. The output of the second SUM 1.6 is connected to the input of the first NO LZ 1.11 and is the signal output of CF 1. The first 1.3 and the second 1.4 M In. SUM me inputs are connected to the outputs of the n-th PAs belonging respectively to the first 1.7 and second 1.8 GUM. The nth taps of the second 1.1 and the first 1.11 NO LZ, respectively, are connected to the first inputs of the n-th PAs belonging to the first 1.7 and second 1.8 GUM, respectively. The first inputs of the first GUM 1.7 together with the input of the second NO LZ 1.1 are the M-bit signal output KF1. The second inputs of the MIND 1.7 ₁ -1.7 _N from the first GUM 1.7 and the corresponding second inputs of the MIND 1.8 ₁ -1.8 _N from the second GUM 1.8 are combined and are 2NP In. QMS KF 1. The output of the second UM 1.10 is connected to the first input of the third UM 1.9 and to the second input of the second SUM 1.6. The second input of the first SUM 1.5 is connected to the inverse output of the third UM 1.9. The second input of the third PA 1.9 is connected to the second input of the first PA 1.2 and is 1 Vx. QMS KF 1. The output of the first NO LZ 1.11 is connected to the first input of the second PA 1.10, the second input of which is Bx. SC CF.

BAU 5 is intended for the calculation and formation of weighting coefficients for each KF1 along the direct and feedback circuits.

BAU 5, shown in figure 3, consists of the first 5.1 and the second 5.2 FVK, the first 5.4 and the second 5.5. KVK, FPS 5.3 and FSOS 5.6. The M × MR inputs of FPS 5.3 are M × MRPC. BAU 5. M × M-bit input FPS 5.3 is connected to the M × M-bit input of the first FVC 5.1. The output of FPS 5.3 is connected to the first input of the first KVK 5.4, the output of which is connected to the input "correction" of the first FVK 5.1. The first M-bit output of the QMS and the second M × N-bit output of the QMS of the first FVC 5.1 are respectively 1MP Output. QMS and 2M × NP Out. QMS BAU 5. M-bit input FSOS 5.6 is connected to the M-bit input of the second FVC 5.2. The M-bit output of the second PCF 5.2 is MR Output. SC BAU 5. The output of FSOS 5.6 is connected to the first input of the second KVK 5.5, the output of which is connected to the “correction” input of the second FVK 5.2. The second inputs of the first 5.4 and second 5.5 KVK are combined and are Vkh. SB BAU 5.

The elements included in BAU 5 have the following purpose.

и

в цепях соответственно прямой и обратной связи для каждого m-го КФ 1_m, в период времени выборки Т.The first 5.1 and the second 5.2 FVK are designed to calculate the weight coefficients

and

in the chains of direct and feedback, respectively, for each m-th CF 1 _m , in the sampling time period T.

The first 5.4 and the second 5.5. KVK are designed to calculate in each sampling period T correction signals X _K , Y _K, respectively, for direct and feedback circuits, taking into account the residual level of the reflected signal (error signal) U _co .

FPS 5.3 and FSOS 5.6 are designed to calculate the total signals X _Σ , Y _Σ, respectively, along the direct and feedback circuits in each time period of sample T.

Other elements that are part of the general AEC scheme and the KF scheme, multipliers, adders, delay lines and a subtraction unit are known and described, for example, in the book by B. Widrow, S. Stearns, “Adaptive Signal Processing”: Per. from English // M .: "Radio and communications", 1989, 440 p. or Chao J., Shigeo T., A new IIR adaptive echo canceler: GIVE., IEEE journal on selected areas in communications. vol. 12, no. 9, December 1994.

BAU 5 can also be implemented as a processor, the algorithm of which is shown in Fig.4. Ports P 2 and P 1 correspond to M × MPCx. and MR. OBA BAU 5. Upon receipt of signals at these ports via direct and feedback circuits, respectively, their total values X _Σ and Y _Σ are calculated (blocks 2 and 3 of the algorithm).

Then, taking into account the error signal received at port P 3 from BV 4, the correction signals X _k and Y _k are calculated for the direct and feedback circuits, respectively (algorithm blocks 4, 5). Then, using the least squares method, we calculate the weighting coefficients for the direct V ^p coupling circuit for the first and second groups of N multipliers of each m-th CF 1 and the feedback V ^o coupling for each m-th CF 1, which are through processor ports P 6 and P 5 arrive respectively at 1 Vkh QMS and Vkh SK of each m-th KF 1. In addition, from port P 4, the calculated values of the weighting coefficients of the first and second groups of N multipliers of each m-th KF 1, which are temporary in relation to the corresponding weight coefficient of the port P6 a delay of one period T are 2M × NP You . QMS BAU 5 and enter 2 NP Vh QMS of each of CF 1 ₁ -1 _m .

Declared AEC works as follows.

The distant subscriber's speech signal x (t) is fed to the input of the N-tap LZ 2 and to the signal input of the first (m = 1) KF 1, as well as to the first discharge M × МРСВх. BAU 5 (figure 1). The signal inputs of other KF 1 are connected to the corresponding taps of the N-tap LZ 2, namely, the n-th tap of the N-tap LZ 2 is connected to the signal input of the (n + 1) -th KF 1 _{n + 1} . This achieves a delay of the signal x (t) between the nth and (n + 1) -th taps of the N-branch line for the time T.

To the second input of the first PA 1.2 with the corresponding discharge of 1 MR Out. QMS BAU 5 receives the calculated weight coefficient. At the signal output of the second SUM 1.6, a signal is generated for evaluating the impulse response of the echo path x '(t), which is fed to the corresponding input M-In. SUM 3 and to the input of the first N-tap LZ 1.11 of the corresponding KF 1. The signal from the output of the first N-tap LZ 1.11 goes to the first input of the second PA 1.10, and from the corresponding taps to the first inputs of the second group of multipliers 1.8 and then to the corresponding bit MP Vh. OS BAU 5. The second input of the second PA 1.10 receives the calculated coefficients from the corresponding discharge MR Out. SK BAU 5. From the output of the second multiplier 1.10, the converted signal of the impulse response is supplied to the first input of the third PA 1.9 and to the second input of the second SMS 1.6. To the second input of the third UM 1.9 from the corresponding discharge 1 MR Output. QMS BAU 5 received calculated weighting coefficients.

As a result of the secondary conversion of the pulse sequence in the third UM 1.9, the generated signal through the inverse output is fed to the second input of the first SUM 1.5. The second inputs of the multipliers of the second group of multipliers 1.8 and the second inputs of the multipliers of the first group of multipliers 1.7 receive N signals from the second NP-input “signal of modeling coefficients” for each CF 1 from the corresponding discharge 2M × NP Out. QMS BAU 5. These signals provide the conversion of the impulse response received from the output of the first N-tap LZ 1.11 and its further transmission to the corresponding inputs of the second M-input adder 1.4. The pulse sequence signals from the N-inverse outputs of the first group of multipliers 1.7 are fed to the corresponding inputs of the first M-input adder 1.3. These inverse N-signals are the result of multiplication in the first group of multipliers 1.7 of the impulse response received from the outputs of the second N-tap LZ 1.1 to the first inputs of the first group of multipliers 1.7 by the gains received at the second inputs of the first group of multipliers 1.7. From the signal outputs of all KF 1 signals are fed to the corresponding inputs of the MVx. SUM 3, the output of which forms a signal corresponding to the level of the echo signal y '(t), and then goes to the first input of the BV 4. The second input of the BV 4 receives the echo y (t) to be compensated. The residual signal e (t) at the output of the BV 4 is supplied to Bx. OSU BAU 5 and the output device.

Thus, in the AEC, the signal for evaluating the impulse response of the echo path is generated not only based on current estimates of the cross-correlation function of the signals x (t) and e (t), but also taking into account the values of the estimates of the impulse response of the echo path obtained at the M previous steps settings. This provides a "prediction" of the echo signal estimation values at time t based on its known values in M previous time instants x (t-T), x (t-2T), ..., x (t-M · T).

The gain coefficients obtained from the outputs of the BAU 5 to the inputs of the PA are prediction factors, the values of which are determined by the least squares method in real time.

Moreover, an accurate determination of the signal energy at each iteration ensures both the stable operation of the echo canceller filter and deeper suppression of the echo signal, i.e. the achievement of the formulated technical result is ensured.

Claims (1)

Adaptive echo canceller on an M-order recursive filter, containing M≥2 channel filters, N-tap delay line, where N = M-1, whose input is the “far-end signal” of the adaptive echo canceller, M-input adder, the output of which is connected to the input of the subtraction unit, the second input and output of which are the reflected signal input and the error signal output of the adaptive echo canceller, the nth tap of the N-tap delay line, where n = 1, 2, ..., N, is connected to the signal the input of the (n + 1) -th channel filter, and the signal the input of the first channel filter is connected to the input of the N-branch delay line, the signal output of the m-th channel filter is connected to the m-th input of the M-input adder, where m = 1, 2, ..., M, characterized in that the unit is additionally introduced adaptive control, equipped with M × M-bit signal input, M-bit first output “signal of modeling coefficients”, M × N-bit second output “signal of modeling coefficients”, M-bit output “correction signal” and M-bit input “ training signal ”, m-th M-bit signal input of block a active control is connected to the M-bit signal output of the m-th channel filter, the m-th second N-bit output "signal of modeling coefficients" is connected to the second N-bit input "signal of the modeling coefficients" of the m-th channel filter, m-th bits M-bit first output “signal of modeling coefficients”, output “correction signal” and input “training signal” of the adaptive control unit are connected respectively to the first input “signal of modeling coefficients”, input “correction signal” and output “training with drove "the m-th channel filter, and the adaptive control unit is equipped with an additional input" error signal "connected to the output of the subtraction unit, and the channel filter consists of the first and second M-input adders, the first, and second, and third multipliers, the first and the second adders, the first and second groups of multipliers of N multipliers in each group, the first and second N-tap delay lines, the signal input of the first multiplier is the signal input of the channel filter, the first multiplier, the first and second M-input adder s, the first and second adders are cascaded along the signal outputs, and the output of the second adder is connected to the input of the first N-tap delay line and is the signal output of the channel filter, the nth inputs of the first and second M-input adders are connected to the outputs of n-multipliers belonging, respectively, to the first and second groups of multipliers, and the outputs of the multipliers belonging to the first group of multipliers are inverse, and the nth taps of the second and first N-tap delay lines are connected to the first inputs of the n-th multipliers, respectively, the first inputs of the first group of N multipliers together with the input of the second N-tap delay line are the M-bit signal output of the channel filter, the second inputs of the n-multipliers of the first and second groups of multipliers are combined and are N-bit the second input is a signal of modeling coefficients of the channel filter, the output of the second multiplier is connected to the first input of the third multiplier and to the second input of the second adder, the second input of the first adder is connected to inv the output of the third multiplier, the second input of which is connected to the second input of the first multiplier and is the first input of the channel filter modeling coefficients signal, the output of the first N-tap delay line is connected to the first input of the second multiplier, the second input of which is the channel correction input filter, and the adaptive control unit consists of the first and second formers of weighting factors, from the first and second correctors of weighting factors, former of a direct signal and shapers feedback signal generator, the M × M-bit inputs of the direct signal driver are the M × M-bit signal input of the adaptive control unit, the M × M-bit input of the direct signal driver is connected to the M × M-bit input of the first weight generator, the output of the driver direct signal is connected to the first input of the first corrector of weights, the output of which is connected to the input "correction" of the first driver of weights, the first M-bit output "signal of modeling factors" the second M × N-bit output "signal of modeling coefficients" which are respectively the first M-bit output "signal of modeling coefficients" and the second M × N-bit output "signal of modeling coefficients" of the adaptive control unit, M-bit input of the driver of the feedback signal connected to the M-bit input of the second shaper of weights, the M-bit output of which is the M-bit output of the "correction signal" of the adaptive control unit, the output of the feedback signal shaper and connected to the first input of the second weight corrector, the output of which is connected to the “correction” input of the second weight former, and the second inputs of the first and second weight corrector are combined and are the input “error signal” of the adaptive control unit.

Images