DE2427102B2 - ECHO COMPENSATOR - Google Patents

Description

a) the value (Σ ^Η Ό ^ ^he sum of squares

all output signals (Wj) of the branching device that are recorded in cycles are determined,

b) a correction factor (q) is obtained by dividing the residual echo signal (e _t ) produced by means of the available echo path model by the step after the preceding process

determined total value (] § £ w?).

c) the coefficients (c * ²¹ ) of the second echo path model are each determined by summing the respective coefficient (cf ^] ) of the first echo path model with the product (q · w,) of the correction factor (q) formed according to the preceding procedure and the associated output signal (w,) of the branching device specified,

d) a second residual echo signal (e ₂ ) (internally) is determined by means of the second echo path model,

e) the selection device (K) checks whether the absolute value of the second residual echo signal (e ₂ ) falls below a specified threshold value! /) and, if this condition (| f ₂ | </) is met, causes the second echo path model to become the new first echo path model ; if the aforementioned condition is not met (Ie ₂ > /), the previous first echo path model becomes the new first echo path model, either directly or in a modified form,

Γ) the new first echo path mode determined after the previous process step! is used to generate the new artificial echo signal and therefore to generate the outgoing signal (s) of the second transmission direction.

2. Echo canceller according to claim 1, characterized in that the coefficients (c) ³¹ ) of the The invention relates to an echo canceller for a first and a second transmission direction having transmission system, in which the signals of the first transmission direction, in particular at the end of the four-wire transmission system , are partially reflected and appear as echo signals in the signal of the second transmission direction. This echo canceller is equipped with a device for generating an artificial echo signal, which has two echo path models approximating the transmission properties of the echo path, a branching device common to the echo path models with a number of the adjustable coefficients of each of the two echo path models corresponding number of output signals, with means for subtracting the artificial echo signal from the signals of the second transmission direction and with a selection device for selecting the door, the generation of the outgoing signal of the second transmission direction used echo path model.

An echo canceller of the aforementioned type is known from DT-OS 22 24403. This well-known Echo canceller contains a comparator, which the residual echo signals generated by means of the two echo path models level with each other and cause the echo path model to approximate better to the actual echo path is used to generate the outgoing speech signals.

The object of the invention is to specify an echo canceller that has a higher setting speed works as known arrangements with comparable effort.

This object is given by the application of the characterizing part of claim 1 Measures resolved.

The application of the measures according to the invention includes a two-step process that with comparably little effort leads to such a high setting speed that the residual echo signal is eliminated except for a quantization noise even with any frequency rejection in the echo path

The invention is explained in more detail below with reference to the drawing. The drawing shows in detail in

Fig. 1 shows the basic mode of operation of a Echo canceller and its inclusion in a transmission system, which may contain a non-synchronous carrier frequency system,

F i g. 2 a scheme and the associated arrangement for determining the preliminary residual echo signal e and the correction factor q,

Fig. 3 is a scheme and the associated arrangement for determining the final Restechosi signals e.

F i g. 1 shows, as an excerpt from a telephone exchange, a fork junction 5 equipped with a line replica, a four-wire path with a delay connected to the fork junction 5 with a first, incoming transmission direction 1-2-4 and a second, outgoing transmission direction 7-8-11 as well as one Two-wire weave connected to the fork junction 5 to a subscriber 6. The signal χ on the incoming transmission direction is here as a rule, especially at the end of the four-wire system, partially reflected because of the imperfect line simulation of the fork 5 and appears as an echo signal y in the outgoing transmission path. The echo canceller contains a quadrupole 9 connected across the four-wire path, which generates the artificial echo signal j from the signal χ of the incoming transmission direction 1; generated by being a ~. the transmission properties of the echo path 2-4-5-7-8 automatically adapted. The artificial echo signal y is added to the signal y + η of the outgoing transmission direction 8 via the subtracting device 16 in the echo-suppressing sense, so that in the ideal case [y - y) the residual echo signal e at the output of the subtracting device 10 is only the signal from the nearby subscriber 6 contains η.

Many four-wire links do not require their own echo cancellers or echo blocks because of their short transit time. Echo cancellers, so that only long-distance connections with high delay times, such as. B. Submarine cable or satellite links, are equipped with these devices at their terminals. Between these connections, which are heavily affected by runtime, and the fork transitions to the two-wire connections but can then still further four-wire transmission systems such. B. a non-synchronous one Carrier frequency system 3 must be switched on, as indicated by dashed lines in FIG. 1.

The present invention is based on an echo canceller as it is, for. B. in the article "An adaptive echo cancellor", by MM Sondhi, is described, which is printed in "The Bell System Technical Journal", March 1967. Pages 497-511. In the case of the Sondhi echo canceller, the signal χ of the incoming direction is fed to a branching device, which in the arrangement according to FIG. 1 as a delay line and in the arrangement according to FIG. 3 is designed as a series of filters connected in parallel. In any case, the branching network has a number of N outputs, which correspond to systems with impulse responses that are linearly independent from one another. and the corresponding output signals χ are each passed through one of the setting elements M ₂₁ ... M _2n and then added up to form the artificial echo ζ, which in the present invention is denoted by jp. The coefficients g, ... g _N , which in the present invention are denoted by c. .. c _N or briefly c ₍ denoted, characterize the transmission properties of the four-pole (transversal filter) generating the artificial echo signal y at every moment and must be continuously readjusted if a frequency offset occurs in the echo path Convergence must be narrowly limited, these known eochocompensators can only very imperfectly from the frequency offset-the echo canceller according to the invention is described in more detail below with reference to FIGS

s accordingly has an analog-to-digital converter 12 for the signal χ, an analog-to-digital converter 13 for the signal y + η and a digital-to-analog converter 14 for the residual echo signal e leaving the echo canceller,

ίο If the 4 kHz band-limited signal χ or x (t) the incoming direction of the four-wire path is sampled at 8 kHz, so the sample values

x (t - (i - 1) Γ) = W ₁ -, i = 1, 2 ... W, T = 125 μβ,

15N = number of outputs of the branching network, the echo signal y {t) in the form

yit) =

i = I

show.

The coefficients c _f reflect the transmission properties of the echo path 2-4-5-7-8 , in which the fork 5 and, under certain circumstances, the non-synchronous carrier frequency system 3 are contained. If the frequency offset disappears, the coefficients c, are constant, otherwise they are time-dependent.
If the echo canceller is equipped with a transversal filter, the artificial echo has the shape

C ₁ W,

where the coefficients <"· characterize the setting of the quadrupole 9 (FIG. 1) of the echo canceller, which is designed as a transversal filter. Equations (1), (2) result in the residual echo signal e or e {t leaving the echo canceller ) the expression

eit) = yU) - yit) =

In order to achieve the smallest possible residual echo signal, the aim of the known adjustment method is to match the coefficients c of the model contained in the transversal filter 9 to the coefficients c of the actual echo path. It can now be shown that for any sat; from sample values w _f (i = 1 ... N) and from coefficients C ₁ a coefficient set r can be found in a simple manner, which does not necessarily _{satisfy the condition c (} == c ₍ , but which in the absence of interfering signals (n = 0) the residual echo signal e (i disappears. If there is no frequency offset, the coefficients c \ converge to the coefficient c, with continued application of the method according to the invention, which is only not directly applicable when a signal η of the nearby participant occurs, since in this case the residual echo signal e (t) does not disappear. On the other hand, a non-disappearing residual echo signal is an indication of the presence of an opponent

Speech signal η and, according to the invention, triggers a specific step to be described later. Let cj ″ be the setting of the echo canceller at the previous sampling time f - T, where T represents the sampling period. With this set of coefficients corresponding to a first echo path model, a residual echo signal results at time t according to equation (3)

(4)

Output signals \ t>, the branching device is determined

c) the preliminary residual echo signal e,

ejt) =

is formed and stored internally;

d) a correction factor q is obtained by dividing the provisional residual echo signal e _x that has arisen by means of the available echo path model by the sum value that has already been determined

With this first, preliminary residual echo signal C ₁ ^{, new coefficients cj 2} 'corresponding to a second echo path model can be formed according to the relationship

i = I

_ S (I)

= cj "+

Σ »ί

- P ₁ W ₁ . ι = 1 --Ν (5)

calculate a new residual echo signal

i - I

produce. Inserting (5) into (6)

(6)

= Σ

c, - c! "

C ₁ W ₁ -)

20th

e) the coefficients c \ ^{1] of} the second echo path model are each determined by summing the respective coefficient cj "of the first echo path model with the product q ■ w, of the correction factor q formed according to the previous method step with the associated output signal w, of the branching device:

cj ²⁾ = c! "+ q * w,;

0 by means of the second echo path model becomes a second artificial echo signal

35

= (21.

= e _x - e _x

= 0

= I.

i = I

generated;

g) a second residual echo signal e _{2 is} determined internally by means of the second echo path model using the generated second artificial echo signal _p ^<2):

shows that the residual echo signal e ₂ disappears.

The two-step process defined by equations (4) to (6) is now implemented using the following process steps:

a) the coefficients c} ¹ 'of the first echo path model and the artificial echo signal y ° * formed therefrom

i = I

are with a clock I ₁ , the z. B. corresponds to a clock toe of T = 125 μβ, determined and stored in a known manner, where N is the number of outputs of the branching network or the number of samples w of a shift register realizing the branching network; b) the value

the sum of the squares of all the bars recorded

45

h) the selection device K checks whether the absolute value of the second residual echo signal e ₂ falls below a specified threshold value /, ie whether there is no interference, in particular from the nearby subscriber 6.

so signal η occurs in signal e ₂ ; if this condition \ e ₂ \ <f is met , the selection device causes the second echo path model to become the new first echo path model; in the event of non-fulfillment (\ e ₂ \ ^ / of the aforementioned condition, the previous one will only «

Echo path model directly or modified for the new echo path model;

i) that according to the preceding procedural step defined new first echo path model is used to generate the new artificial echo signal and with

towards the generation of the outgoing signal e dei second transmission direction 11 is used.

In a further development of the invention, the selection device K is designed such that the coefficient c} ³ 'of the modified first echo path model!

Assume values that lie between the train-related value cj ^{11 of} the first and the associated value cj ² 'of the second EchopfedmodeHs. The selection device K controls the coefficients

cf ^] such that the greater the second residual echo signal | e, |, the closer they are to the coefficients cf> of the first echo path model, which can be expressed as follows: c \ '= cf ^] + α q · W ₁ -, where «is a modification factor between 0 and 1 (0 < a <V) . This modification factor is cyclically controlled in such a way that it is smaller, the larger the second residual echo signal | e ₂ l.

Figures 2 and 3 show one with a transversal filter built-up digital echo canceller, which carries out method steps a) to i): In detail gives F i g. 2 process steps a) to d) again.

The samples W ₁ - (/ = 1 ... N) are stored in the shift register S1 and the transversal filter coefficients cf are stored in the shift registers S 2 and S3. Which of the two memories S2 or S3 is used to carry out method step a) depends on the switch controlled by tj-. The function of I ₁ is explained below. By a multiplexing technique, the functions a) and b) are performed by the multiplier M and the adder Al, and the intermediate results y ^il) or plexten multiplier M of the signal e, the quotient

ι = I

stored in the memories S4 and S5. Here, the multiplex clock f controls the respective inputs of the multiplier M and the adder Al in such a way. that in one switch position the product w, c \

the correction factor q arises.

In Fig. 3 that part is the echo canceller shown, which carries out process steps e) to i).

Via the Mulliplextakl I ₂ , the product q ■ μ ·; formed, which is stored in the memory Sl and the respective coefficients cj "is added. At the output of the adder A 1, the coefficient arises c ^'21st After switching of the clock I ₂ is supplied via the adder Al and the memory S4, the second artificial echo signal y . ^a) formed in the absence of interfering signals (n = 0) would now be the second echo residual signal e ₂ at the output of the memory disappear S4 according to equation (7) by resulting from the digitization quantization error of the absolute value may |. e ₂ | of the second echo residual signal, only a If this absolute value | e ₂ | is greater than /, the state of double-talk is present, ie a double-talk signal η is superimposed on the echo signal y . In this case, the direct application of the algorithm according to equations (4) to (6) leads to an incorrect setting of the coefficients c | ²¹ and thus to an incorrect calculation of the echo signal y ^a> , the preliminary restech remains osignal

and obtain the original setting of the transversal filter coefficients cf ^1. For this purpose, the clock t _f , which is generated from the selection device K containing a comparator 35, controls the memories S2 and S3 and the input of the digital-to-analog converter 14 in such a way that, in the event of a contradiction, the preliminary residual echo signal C ₁ becomes the final residual echo signal t - and is used in the digital-to-analog converter 14 as an outgoing signal in the outgoing direction 11 ^ ₁ "'. For the next sampling time:

2j ^M ~ t + T the content of the memory S3 becomes the formation

of the new preliminary artificial echo signal y ^{01 is} formed. Used by multiplying by the multi- (r + T).

and in the other position the product wf is formed and accumulated in the corresponding memories S4 or S5. The first residual echo signal p produced by a subtracter A3 is then stored in the memory S6 and the quotient is stored by a divider D.

For this purpose 3 sheets of drawings

609528/3

Claims (3)

Patent claims: 1. Echo canceller for a transmission system having a first and a second transmission direction, in which the signals of the first transmission direction, in particular at the end of the four-wire transmission system, are partially reflected and appear as echo signals in the signal of the second transmission direction, with a w Device for generating an artificial echo signal, which has two echo path models approaching the transmission properties of the echo path, a branching device common to the echo path models with a number of output signals corresponding to the number of adjustable coefficients of each of the two echo path models, with a center for subtracting the artificial echo signal from the Signals of the second transmission direction and with a selection device for selecting the echo path model used for generating the outgoing signal of the second transmission direction, characterized by means for implementation fo The following process steps after setting the coefficients (f {") of the first echo path model in cycles: modified first echo path model assume values which are each between the associated value of the first uad and the associated value of the second echo path model.
" 3. Echo path model according to claim 2, characterized in that the selection device (K) controls the coefficients <cf ') in such a way that the greater the second residual echo signal (| t), the closer they are to the coefficients (cf) of the first echo path model - ₂ | J is.