FR2480534A1 - METHOD AND DEVICE FOR CANCELING ECHO IN A DATA TRANSMISSION SYSTEM - Google Patents

Abstract

THE INVENTION RELATES TO DATA TRANSMISSIONS. A TWO-WIRE FULL-DUPLEX DATA TRANSMISSION SYSTEM INCLUDES AN ADAPTIVE 24 ECHO CANCELER AT EACH TERMINAL 10. DURING SIMULTANEOUS TRANSMISSION AND RECEPTION INTERVALS, THE ECHO COMPENSATED SIGNAL MAY INCLUDE NOT ONLY AN UNCancelled ECHO COMPONENT BUT ALSO ONE CORRESPONDING TO REMOTE TERMINAL DATA. AN ADAPTATION ERROR SIGNAL GENERATOR ESTIMATES AND DELETS THE REMOTE TERMINAL DATA COMPONENT IN THE ECHO COMPENSATED SIGNAL. APPLICATION TO DATA TRANSMISSIONS OVER THE COMMUTED TELEPHONE NETWORK.

Description

The present invention relates to the cancellation of

  echo signals in trans-

  his. The invention relates more particularly to the cancellation

  echo signals present in trans-

  put by two-way telephone transmission channels

two-wire

  In the field of data transmission, it is often advantageous to simultaneously route the traffic in two directions on a single transmission line (link, channel). Thus, the traffic is carried out in full duplex. A typical transmission medium consists of a single channel

  Two-wire telephone line belonging to the public

  switched long distance matter. The bandwidth of such

  Two-wire track is approximately 300 to 3000 Hz.

  For full-duplex data transmission, the available bandwidth can be divided into two halves,

  each half being assigned to a particular direction of transmission

  ticular. However, with this method error-free data transmission can only be performed at a rate equal to half that which could be achieved with unidirectional (half-duplex) transmission. A way

  to increase the data rate in full duplex

  to use two physically distinct two-wire lines

  each line carrying a unidirectional signal occupying the entire bandwidth in a respective transmission direction of the two. This is what

  which is called a four-wire path.

  Full-duplex fast transmission can also be achieved on a single two-wire path using hybrid coupling networks. These networks, placed at the so-called near end as at the so-called remote end of the two-wire path, accept a four-wire signal and convert it into a two-wire signal.

  to transmit it on a two-way telephone

  the two-wire. To ensure optimal suppression of spurious signals in the transmission, the impedance of the hybrid network access associated with the channel must be exactly matched to the impedance of the two-way channel.

  son. However, this is rarely possible in practice.

  Specifically, the switched nature of the toll network means that a large number of different impedance transmission paths are connected over time to the hybrid network. Because the hybrid network is designed to operate with the greatest possible number, in practice, of different transmission paths,

  there is usually a mismatch between the hybrid network and

  from and the way. Misadjustment causes a portion of the signal that has been transmitted from the near end to be reflected in the path from the point of connection between the pathway and the far end hybrid network. As in the voice transmission, this reflected distant signal is echoed. A data receiver is typically unable to distinguish between data that originates from the far end and echo data that originates from the near end. There is

  therefore a risk that the near-end receptor

  mistakenly mistaken as data from the far end

  the echo that is reflected at the far end.

  We can treat this problem by using

  echo cancellers. They produce a signal that is

  mentally an image of the echo component present in an incoming signal, i.e. the signal that the path

  two-wire application to the hybrid network of the

  che. More precisely, each symbol among a predetermined number

  terminated previous consecutive symbols of the transmitted signal is not only issued but also recorded in the echo canceller. Each of these symbols is multiplied

  in the canceller by a respective setting coefficient.

  The resulting products are summed to produce the image signal. A resultant substantially echo-free signal is obtained, hereinafter referred to as an echo-compensated signal, by subtracting the image signal from the incoming signal. The echo-compensated signal is applied to a data receiver which, after processing such as equalization and demodulation, makes decisions

  concerning the values of transmitted data symbols.

  In general, the echo cancellation process is not perfect. The echo-compensated signal does indeed contain an echo component that is not canceled. he

  may also contain a data component of

  far away, as will be described in more detail below. In either case, the level of the un-canceled echo component indicates the current efficiency of the echo cancellation process. A non-canceled echo component of high value means that the image signal is an inaccurate image of the echo component desired to be canceled. In so-called adaptive echo cancellers, the echo-compensated signal is advantageously used as an error signal under which the values of all the tap coefficients are updated adaptive in order to minimize the component

  echo not canceled. This makes the image signal

  ge to reproduce continuously and, as far as possible, accurately, the echo component present in the incoming signal, even if the characteristics of the

track change.

  The configuration described in U.S.

  4,087,654 is an example of adaptive echo cancellers.

  tives operating at the baud rate. In these structures, sampling the incoming signal, developing an image of the echo component such as echo cancellation

  occur at the baud rate (symbol). Although these cancellations

  their structure is simple, they are highly sensitive

  the synchronism between the transmitted signal

  by the near end, which is used to define the

  of the echo signal, and the data received, the characteristics of which

  temporal characteristics are determined at the far end. In addition, the signal compensated for echo

  is available at the receiver only at the sampling rate.

  swimming baud. This severely restricts the ability of the receiver to accurately retrieve the synchronization

  from the signal from the far end.

  It was also suggested to make

  the canceller at the rate of Nyquist. Techniques to exchange

  at the Nyquist pace, which mitigate the problem

  the timing of the synchronization described above, are disclosed, for example, in U.S. Patent 4,131,767 and in US Pat.

  Articles: "Passband Data-Driven Echo Cancelbr for Full-

  Duplex Transmission on Two-Wire Circuits ", - IEEE Transactions on Communications, Vol COM-25, No. 7, July 1977, pages

  654-666, and "A New Digital Echo Canceller for Two-Wire Full-

  Duplex Transmission ", by KH Mueller, IEEE Transactions on Communications, Vol COM-24, No. 9, September 1976, pages 956- 962. Unlike cancellers operating at the baud rate, cancellers operating at Nyquist rate perform the following functions: sampling of the incoming signal, the generation of the echo image as the cancellation of

  the echo of Nyquist's cadence. The adaptation of the coefficients

  ing values in Nyquist configurations is satisfactory in full-duplex systems during

  unidirectional transmission intervals. How-

  However, adaptation is unreliable during double emission or bidirectional transmission intervals, ie in the case of simultaneous transmission of data.

  data from the far end and the near end.

  These problems come from the fact that the compensated signal

  vis-à-vis the echo that is applied to the adapta-

  during the double emission intervals contains

  not only the echo component not canceled, but also

  a data component of the far-end. Far-end data is not correlated with echo. The adaptation, and therefore the generation of the image of the echo from this signal, are thus either unreliable and inaccurate, or very slow. This can result in an erroneous reconstruction of the data. (Structures operating at baud rate are not affected by these problems because the error signal that is used to update the

  the echo canceller is taken at a different point.

  in the system, to which the data from the end

  were determined and were therefore subtracted.

  As a result, the error signal is not altered by the pre-

  endpoint data).

  The solutions of the prior art to the problems mentioned above concerning Nyquist cancellers

  include the use of a circuit for detecting

  emission to stop the adaptation and freeze the catch coefficients to the values they possessed before the double emission, to use these values during the

  double emission intervals. For example,

  As such, U.S. Patent 3,499,999. According to one variant, described by Weinstein in the aforementioned article.

  entitled "A Data-Driven Passband Echo CanceJJ-r for Full-

  Duplex Transmission on Two-Wire Circuits ", it is possible to use a rotating average of a predetermined number of anterior coefficient values for each take, instead of the adaptive coefficients during the double emission intervals, Although these solutions stabilize the operation of the system, echo cancellation during

  double emission intervals is potentially inaccurate

  you. This is because during the double emission intervals, the tap coefficients can not adapt to variations in the impulse response of

  the echo path that occur during these intervals.

  An object of the invention is therefore to improve the performance of echo cancellers during periods of

double emission.

  A more specific object of the invention is to improve the performance of sampling echo cancellers at the Nyquist rate during double emission periods. An even more particular object of the invention is to provide a sampling echo canceller at the Nyquist rate which is accurate and stable during double emission periods, even if the characteristics

of the echo path change.

  According to the invention, performance is achieved

  improved maneuvers during the double emission by suppressing

  virtually the data components of the far end

  present in the echo-compensated signal, prior to the application of this signal, as an error signal, to the adaptive echo cancellation structure. As an example, this function is accomplished by what one

  hereinafter refers to an adaptive reference trainer.

  More specifically, the reference trainer

  adaptive addresses a predetermined number of decisions

  receiver to generate, at the Nyquist rate, an estimate of the data component of the far end present in the compensated signal

  vis-à-vis the echo. This estimate is more particularly

  a linear combination of the receiver's previous decisions and is generated by multiplying each of the receiver's previous decisions by a

  taken, and combining the resulting products.

  This generated estimate of the far-end data is used to generate an adaptation error signal whose value is equal to the difference between the signal

  compensated for the echo and the estimation of the

  data from the far end. It is this error signal rather than the echo-compensated signal that is used as an error signal for the canceller

  adaptive echo. Consequently, the adaptation of the coefficients

  the echo canceller is taken only once-

  under the effect of un-canceled echo components.

  This makes it possible to achieve a uniformly accurate and stable echo cancellation both during the double emission

  during unidirectional transmission.

  To ensure that the estimate of the data from the remote end is continuously

  precise way the data component of the far-end

  If the signal is present in the incoming signal, the adaptation error signal may be returned to the adaptive reference formatter as an update error signal for use in the adaptive update.

  setting factors of the reference trainer adapt-

  tif. The invention will be better understood when reading

  the following description of an embodiment and in

2480534.

  Referring to the accompanying drawings in which: Figure 1 is a block diagram of a prior art digital data transmission system,

  of two-wire full duplex type, with a provision

  echo cancellation system with sampling at Nyquist rate; FIG. 2 is a block diagram of a data transmission terminal with sampling at the Nyquist rate corresponding to the invention; FIG. 3 is a block diagram of the circuits belonging to the terminal of FIG. 2 which generate the adaptation error signal of the invention; and Fig. 4 is a block diagram of the adaptive reference formatter which is used in

Figure 3.-

  Figure 1 shows a transmission system

  of full-duplex digital data of a known type.

  This system essentially comprises a transmission link

  two-wire mission, 5, linking two transmission terminals

  data transmission, namely a near end terminal 10 and a remote end terminal 13. For example, the transmission link 5 is part of the public switched automatic public network, although the invention is equally well applicable to other types of transmission links, such as private circuits. The communication link 5 is a bidirectional link, i.e. it carries data signals to

  from each of the two terminals to the other. Terminals

  10 and 13 are of the general type which is described and

  disclosed in U.S. Patent 4,131,767 previously mentioned.

  For example, the two terminals have an identical structure and they work in the same way. From this

  fact, the following description will be practically limited to

  data terminal 10 of the near end.

  The terminal 10 includes transmission circuits /

  reception 40 and a hybrid network 16. The circuits 40

  take a transmission part which contains a data source 11 and a transmitter 14. The data source produces

  a baseband train of data symbols from the

  close mite, bn, with n, 0,1,2 ... The index n progresses at the rate of baud. The symbols bn are applied to the transmitter

  14 to be shaped and modulated in a conventional manner.

  The transmitting part, as the reception part

  which will be described shortly, is associated with the hybrid network 16.

  The latter makes it possible to connect a pair of two-wire lines, that is to say a four-wire line, to a link

  bidirectional two-wire transmission, 5. More

  In brief, the hybrid network comprises three two-wire accesses 16a, 16b and 16c. The outgoing signal, that is to say the output signal of the transmitter 14, is applied to the two-wire access 16a via a two-wire line 18. The hybrid network 16 conveys this signal. signal to the link

  transmission 15 through the access 16c.

  On the other hand, an incoming signal coming from the far end, which appears on the transmission link 5 and reaches the access 16c, is routed by the hybrid network 16 to the access 16b. From there, the incoming signal r, representing a far-end data symbol stream, is conveyed by a separate two-wire line 19 to the receiving part of the near-end data transmission terminal 10. . Similarly, the

  hybrid network 15 of the far end connects

  transmission / reception cooking 17 (which are similar to

  baked 40) at port 15c and at the transmission link 5 via a pair of two-wire lines which

  are connected to the ports 15a and 15b.

  In order to optimally eliminate the disturbances likely to affect the transmission, the output impedance of the hybrid network 16 of the near end as well as the

  hybrid network 15 of the far-end must be

  adapted to the impedance of the transmission link 5.

  However, this is rarely possible in practice. For example

  In the switched automatic toll network, a large number of different transmission links are connected over time between the hybrid networks 15 and 16. Because the hybrid networks are designed to operate with a number Since transmission is as high as practically possible, there is generally impedance mismatch, for example between the hybrid network of the far end and the transmission link 5. This has the consequence that an appreciable fraction of the signal from the near end and reaching the hybrid network of the far end is reflected in the transmission link 5 in the form of an echo. At the end of a finite time interval, the echo appears at the access 16c of the hybrid network 16 of the near end. The receiving part of the near-end data terminal 10 is incapable of distinguishing between the incoming data and the echo. However, the presence of the echo canceller 24 (which will be considered shortly) in the receiving part prevents the echo disturbing the process of

data recovery.

  As indicated above, the signal

  incoming r which is received from the -transmis-

  5 is routed by the hybrid network 16 and the access 16b to the receiving part of the data transmission terminal 10. In the latter case, this signal is all

  first applied to a Nyquist 20 sampler.

  nier samples the signal r finely, at a rate at least equal to the rate of Nyquist, that is to say a rate at least equal to twice the highest frequency that can exist in the incoming signal. In one

  purpose which will appear later, we take as an example

  for the Nyquist rate, an integer multiple P of the baud rate. The sample rM is one of the resulting samples, that is, the M th sample of a sample stream of the incoming signal. The index M progresses at the rate of Nyquist. In the general case, the signal r comprises both data signals from the far end and echo signals. So, a fraction of the

  value of the sample rM is due to the data from the

  distant and another fraction is due to echo. These fractions of the sample rM will be referred to hereinafter as the data component of the far end and the echo component respectively. It should be noted that, for example,

  Nyquist sampling, intersymbol disturbances

  and other distortions, the value of the data component of the rM sample of the incoming signal does not correspond

  not to the value of a particular issued symbol.

  We will now consider the problem that the invention aims to solve, assuming first of all that

  at any moment there is only one unidirectional transmission

  recalibration on the transmission link 5, that is to say that the double emission is excluded. It will further be assumed that the hybrid network 16 is leak-free, i.e., the outbound transmitted signals that are applied to the access 16a do not cross the hybrid network to appear at the same time.

access 16b.

  Under these conditions, the sample rM is

  killed exclusively by a data component of the

  distant or exclusively by an echo component.

  During reception periods, for example, the sample

  lon rM consists solely of a data component of the far end, i.e. it is echo-free. The sample rM is applied to a combining circuit 22 in which it is subtractively combined with the echo image signal ZM ', the latter being the Mth signal of an echo image signal stream which is applied to the adaptive echo canceller 24 by a driver 21. Specifically, each image signal

  zM echo is an estimate by the echo canceller

  tif 24 of the echo component of the sample rM * Since this component is null by hypothesis, the

  Echo image signal ZM is also zero. Thus, the exchange

  tillon rM almost seamlessly crosses the circuit

  22. The output signal of the combination circuit

  season 22 is a compensated signal train vis-à-vis

  echo, and the SM signal is the Mth signal of this train.

  In this case, the echo-compensated signal SM is substantially equal to the data component of the far end of the sample rM. The echo-compensated signal is applied by a conductor 28 to a low-pass filter 30 which reconstructs a continuous wave from that signal. The output signal of the filter is in turn transmitted to the receiver 34 in which it can be sampled, for example at the baud rate, filtered again (equalized)

  to combat intersymbol disturbances, and quanti-

  to produce decisions an, n - 0,1,2, ..., concerning

Iémen

  the value of the n symbol emitted by the remote end

  year, year. The decision an is applied to a receiving element

data carrier 36.

  Each echo compensated signal, SM, is also returned by the driver 26, as an error signal, to the adaptive echo canceller 24,

  as will be discussed below in a more complete manner.

  However, as long as the source 11 does not provide data symbols, the echo canceller 24 maintains the value of the echo image signal M at zero. On the other hand, during the unidirectional transmission periods by the terminal 10 (assuming again that there is no double emission), the sample rM consists exclusively of the echo component produced by the signal emitted by the near end and reflected at the

  level of impedance mismatch at the point of conne-

  xion between the channel and the hybrid network of the far end. Each echo image signal zM is now different from zero. More specifically, the echo canceller 24 generates an echo image signal zM by working on a predetermined number of previous consecutive symbols in

  the data sequence that the data source produces 11.

  These symbols are recorded in a transverse structure.

  dirty inside the echo canceller. Physically, this structure can be for example a delay line

  analogue, a shift register or a RAM.

  The echo canceller generates a linear combination of symbols

  previous consecutives by multiplying each symbol

  individual by a respective catch coefficient. Professionals-

  The resultant bits are summed together to produce the echo image signal zM. Since the terminal 13 of the far end is not transmitting at this time, the signal

  echo compensated, SM, is exclusively

  by an un-canceled echo component. As we have

  indicated above, the signal SM is returned by the

  driver 26 to the echo canceller 24 as an error signal. As a result of this error signal, the values

  catch coefficients are updated appropriately

  to ensure that the echo image signal accurately reproduces, as far as possible, the echo component of the sample rM *. In this way, the un-canceled echo component that remains in the

  Echo-compensated signal is minimized.

  Although the configuration in Figure 1 is

  door satisfactorily during operation

  unidirectional, as has just been described, it presents

  serious disadvantages for the two-way operation

  rectional, that is, during the double emission intervals. More specifically, each time the terminals and 13 transmit simultaneously, the echo of the data transmitted

  by the near end is present on the transmission link

  mission 5 together with the data from the far end that is transmitted by the terminal of the far end. So, the error signal that is applied to

  the echo canceller 24 by the driver 26 contains not only

  an echo component not canceled, but also a data component of the far end. This error signal is thus altered by the data component of the far end. The echo canceller is unable to distinguish between the component

  echo and the data component of the far end.

  In addition, data from the far end is not correlated with echo. Thus, adaptation is either unreliable and inaccurate, or very slow. he can

  this results in an incorrect echo cancellation.

  The invention relates to a method and a device

  designed to provide accurate, stable and reliable echo adaptation and cancellation during the dual emission periods. In accordance with the invention, this is achieved by substantially eliminating the data disturbing component of the far end in the echo-compensated signal to generate an adaptation error signal. . This adaptation error signal is equal to a combination of the echo image signal ZM, the sample rM and an estimate (considered hereinafter) of the data component of the far end which is associated with sample rM. It is the adaptation error signal and not the echo-compensated signal, which is applied as an error signal to the adaptive structure

echo cancellation.

  Figure 2 shows a transmission terminal

  10 'data record containing an adaptive canceling device

  echo pattern corresponding to the invention. The difference is

  The transmission between the terminal 10 'and the terminal 10 of the prior art consists in that the transmit / receive circuits 40' of the first city include an adaptation error signal generator 80. The signal generator of adaptation error receives input and processes both the receiver's decisions appearing on the driver

  and the signal compensated for the echo, SM, appears

  on the driver 26. The output signal,

  on the conductor 27, is a train of adaptation error signals among which YEM-D is the (M-D) th signal of this train. The remaining elements of the terminal 10 'are similar to the corresponding elements of the terminal 10

  and they carry the same reference numbers.

  Figure 3 shows in detail the sub-

  constituent sets of the error signal generator

  80. More specifically, the decisions of the Receiving

  for example the decision an, are applied to a

  Adaptive benchmark 82. Decisions are processed

  in the adaptive reference formatter to produce on a conductor 83 a train of estimates, each of these estimates being an approximation of the data component of the far end of a particular one of the samples of the incoming signal being applied to the combination circuit 22. There is a delay in processing D

248Q534

  Nyquist intervals between the output of the combination circuit

  22 and the output of the receiver 34. Thus, at the instant at which the echo-compensated signal, SM, appears on the conductor 26 (FIG. 2), the signal which appears on the conductor 83 is an estimate XM D of the data component of the compensated signal with respect to the echo

  was generated at Nyquist intervals earlier, that is,

  say the SMD signal. The generator 80 comprises a delay circuit 85 which communicates a delay of D intervals of Nyquist to the compensated signals with respect to the echo which

  are applied by the driver 26. Thus, the signal

* thought about the echo and delayed S appears on the

  output lead 88 of the delay circuit 85 to the

  the estimate XMD appears on the conductor 83. This estimate is subtracted from the SM-D signal in the combining circuit 84 to produce an EM-D error signal.

  Of course, the error signal EM D does not represent

  not feel the current echo cancellation error, but the one that existed at Nyquist intervals earlier. We can nevertheless rely on this signal for the update

  setting factors that are used in the year

  24. More precisely, the EM-D error signal is multiplied by a parameter y in the multiplier

  86 to produce a YEMD adaptation error signal.

  The parameter Y, which is much less than unity, is chosen to ensure a gradual and stable convergence, that is to say a minimum of overshoot by higher value and by lower value in the response of

  the adaptive echo canceller 24 to echelon variations

  one of the characteristics of the transmission link 5.

  (In the exemplary embodiment, the value of parameter Y is fixed, but in other embodiments it may be advantageous to dynamically adjust the value of Y so that it is equal to the inverse of the mean squared value of all data symbols that are stored in the adaptive reference formatter.)

  As described below in more detail,

  te, the adaptive reference formatter 82 receives an update error signal which is for updating the tap coefficients used in the formatter. Like the adaptation error signal, the update error signal is equal to the product of the EMD error signal by a

  predetermined parameter. Thus, the error signals of adap-

  tation and update are mutually proportional.

  More particularly, in this embodiment, both are equal. Thus, as shown in Figure 3, the

  YEM-D adaptation error signal is not only

  to the adaptive echo canceller 24 by the driver 27,

  but still referred to the appropriate referee trainer

82 by the driver 89.

  It may be desired, however, that the update error signal that is applied to the adaptive reference formatter 82 be different from the adaptation error signal, which gives different error sensitivities to the reference formatter and echo canceller. This could be accomplished, for example, by passing the driver 89 out of the output of a second multiplier (not shown) instead of from the output of the multiplier 86. Like the multiplier 86, this second multiplier would receive a signal from the multiplier 86. input from the output of the combination circuit 84, but it would multiply this signal

  output by a different parameter.

  As shown in FIG. 4, the adaptive reference formatter 82 includes an offset portion 82a, a far end data estimation portion 82b, and a recording and update portion of the data.

  catch coefficients, 82c. In each part, the function

  during a treatment cycle

  whose duration does not exceed a Nyquist interval.

  This makes it possible to produce a new estimate of the

  data endpoint for each

  output signal from the Nyquist sampler 20.

  For example, the rate of Nyquist is equal to P times the baud rate, denoting by P an integer. Thus, estimates of the data component of the far-end must be generated during any baud interval. A single receiver decision is applied to the adaptive reference formatter during each baud interval. However, in accordance with one aspect of the invention, the adaptive reference trainer 82 forms each of the P estimates in the form of a combination

  respective linear of a common set of N decisions pre-

  of the receiver, each linear combination being

  formed using a particular set of P together

NOT

  coefficients of catch, and N being a selective number

  which is equal to the number of Nyquist intervals over which the decisions that are recorded in

  the adaptive reference trainer 82.

  More specifically, the offset portion 82a includes

  takes a selection logic circuit 114 and a shift register with p positions, 116. These two elements fill

  jointly the function of a circular shift register

  to the right of the length Np. At the appearance of each decision of the receiver, for example an, on the driver, a selection signal from a clock circuit

  (not shown) which is part of the terminal 10 'is

  As a result of this signal, the decision of the receiver, an, is transmitted by the selection logic circuit 114 to the input of the register.

  116. However, at all other times, the cir-

  The selection logic 114 applies the output signal of the shift register 116, appearing on the conductor 119, to the input of this same shift register. Thus, the shift register 116 continuously contains the most

  recent decision of the receiver, an, and N -1 previous decisions

  dentes of the receiver. Inside the shift register

  116, these decisions are ordered by rank of appearance.

  For example, just after the logic circuit has admitted a

  new decision, the "oldest" decision, that is to say

  say, when it is recorded at the far right position (exit), and the decision immediately preceding the oldest one is recorded in a position furthest to the left, and so on. offset 116 is shifted N times during each Nyquist interval (processing cycle), so that during each processing cycle a sequence of the N decisions of the receiver that are recorded appears at the connection point 1180 Thus, during each baud interval, this sequence of N past and ordered decisions P of the receiver is applied P times to the connection point 118. In addition, the register 116 is shifted one position to the right just before each new decision of the

  receiver is applied to it. From the point of contact

  xion 118, the complete sequence is applied to the parts

  82b and 82c by respective conductors 121 and 120.

  The far end data estimating portion, 82b, uses the rank sequence i among the P tap coefficient sequences during the rank i processing cycle to produce the XM far end data estimate. ^ D The following equation governs the operation of this part N / PK = 'Wn (i) anK + 1 for i - mod (M) (1) K = .1 In this equation, WK (i) represents the current value n of the rank coefficient K in the sequence of rank coefficients i, among the P coefficient sequences. The value of the module function mod p (M) is equal to the remainder of the quotient p, for example i, 3 for M, 11, P = 4. It can thus be seen that each estimate of the far-end data that is formed during a given baud interval is a linear combination of a common set, i.e., of the same set, of past decisions of the receiver, this linear combination being formed with the respective one of the P sets of N0 gain coefficients P

of taking.

  More specifically, during the first Nyquist interval during the n-rank baud interval, the common set of N previous decisions of the receiver an-K + 1 'K, 1,2, ..., N is combined linearly using the first sequence of the gain coefficients of

K N

  taken, that is, W (1) K = 1,2, ... ,, to produce the first of the P estimates of the data component of the far end. During all successive Nyquist intervals during the same baud interval, the same set of N decisions of the receiver is combined with p N a different set of N tap gain coefficients to produce further estimates of the end data. remote. This process continues until the common set of decisions of the receiver has been processed with p N

  all P sets of - tap gain coefficients.

  The far end data estimation part, 82b, comprises a function selection circuit 134, an arithmetic unit 136 and a register 142. In the embodiment which is considered by way of example, each decision in year of the receiver is ternary, that is to say

  it can take one of three values: +1, -1 or 0.

  To calculate each term of the summation defined by

  Equation (1) above, the function selection circuit

  134 determines the value of each decision of the

  present on the conductor 121 and, depending on the particular value, it orders the arithmetic unit 136 to perform a particular operation on the signals which

  are shown on entries A and B of the latter.

  More precisely, sZ the decision of the receiver is +1, the arithmetic unit 136 receives the instruction to add

  -the values presented on entries A and B. If the decision

  receiver is -1, the arithmetic unit 136 receives the instruction to subtract the value presented on the input A with respect to that presented on its input B. In either case, the result is applied to exit 0. Thirdly, if the decision of the

  receiver is zero, the arithmetic unit 136 simply applies

  0 The signal that is applied to the input B is the output signal of the register 142, and that which is presented on the input A is the sequence of setting coefficients.

K N

  Wn (i) K 1,2, .... -; The register 142 is used to temporarily record the result produced by the arithmetic unit 136 and to apply it to the input B of the latter, for use in the following calculation. The contents of the register 142 are set to zero at the beginning of each processing cycle by the application of a reset signal

  (generated by circuits that are not represented).

  Thus, through the temporary recording of successive results, the register 142 contains a running total of the results of all previous operations performed during a Nyquist interval during the calculation of equation (1). At the end of each processing cycle, in which N decisions and tap coefficients have been processed, the output signal of the arithmetic unit 136 which appears at the connection point 138 and on the conductor 84 is the XM-estimate. D of the data component of

the far end.

  The recording and updating portion of the tap coefficients, 82c, provides and updates

  adaptive P sequences of tap gain coefficients.

  In particular, a coefficient is updated by modifying its value by means of a correction factor equal to the product of an update error signal (which in this embodiment is equal to error signal

  adaptation) by a decision of the receiver.

  Specifically, the coefficients of a particular sequence must be updated under the effect of the particular update error signal that has been formed as a result of the use of that sequence in part 82b. Thus, the coefficients of the sequence of rank i could be updated under the effect of the signal fE M-D with i = modp (M), as in equation (1). However,

  as will be shown in the following review, the values of the coefficients

  in the sequence of rank i are updated in this embodiment before the estimate XM-D, and thus the

signal 'EM-D, be formed.

  As a result, Part 82c updates the coeffi-

  The latter is equal to that of the P adaptation error signals formed in the preceding baud interval, that is to say the interval. of rank (n-1), which corresponds to

  the sequence of rank coefficients i.

  As previously indicated, the updating process includes multiplying the update error signal by a decision of the receiver. The latter is the decision by which the updated coefficient has been multiplied in the baud interval during which the

  error signal has been formed. Thus, in this mode of

  the appropriate decision is the decision in Kil K 1,2, - N

  It follows from the foregoing that the functioning

  Part 82c, that is the adaptation of the catch coefficients, is governed by the following equation

KK J 1,2, ..., N

  WK n (i) Wn K (i) + YE (pDoP) anK for {, modp (M; (2) This equation shows that a different sequence among the P tap coefficient sequences is updated during each processing cycle, so that each of the P catch coefficient sequences is updated to

during each baud interval.

  The recording and updating portion of the tap coefficients, 82c, includes one baud delay circuits, 90 and 117, a function select circuit 122, an arithmetic unit 124, a demultiplexer 123,

  shift registers 126a to 126p and a multiplexer 125.

  The function selection circuit 122 operates in a manner similar to the function selection circuit 134 of the portion 82b. In particular, the selection circuit

  function 122 instructs the arithmetic unit 124 to perform

  kill one of three operations on the signals applied to

  its inputs A and B, in order to calculate the coefficient of

  updates W K (i) according to equation (2). More precisely n + 1, depending on the value, i.e. +1, -1 or 0, of each of the decisions of the receiver provided by the delay circuit of a baud 117, the arithmetic unit 124 supplies respectively as the value of the updated coefficient the sum of the values-of the signals presented on its inputs A and B, the difference between the values of these signals or the value of the signal presented on the input B. The signal which is present on the input A by the driver 91 is the signal YEM DraP 'provided by the delay circuit of a baud 90. The signal which is presented on the input B by the conductor 127 and the connection point 128 is the value of the tap coefficient Wn (i). NOT

  In Part 82c, each set of coefficients

  The capture values are stored in a respective one of the shift registers 126a, 126b, ... 126p. The value of the signal PH applied to the conductor 115 indicates which of the sets of tap coefficients to be conveyed by the demultiplexer 123 and the multiplexer 125. The value of the signal PH is incremented during each successive processing cycle and is restored at the beginning of each baud interval. In this way, a

  different set of N tap coefficients is selected

  updated for each treatment cycle.

  ment, and all the P sets are selected

at each baud interval.

  It goes without saying that many modifications can be made to the described method and device.

  and represented without departing from the scope of the invention.

Claims (8)

  A method of processing samples of an incoming signal, said signal representing a train of data symbols and said samples comprising respective data components and echo components, which method comprises operations which consist in generating an image of each of the echo components, under the effect of an adaptation error signal train, to combine each of the samples with the image of its echo component to produce a compensated signal train the echo,   and to form decisions containing the values of the symbols   data from the echo-compensated signals; characterized by forming an estimate of the data component of an individual sample from individual decisions, and generating an individual adaptation error signal under the effect of that estimate of the individual sample and the image of his echo component.   Apparatus for carrying out the method of claim 1, said apparatus comprising echo canceling means (24) which operates under the control of an adaptation error signal stream so as to generating an image of each of the echo components, and signal processing means (22, 30, 34) for combining each sample with the image of its   echo component to produce a compensated signal stream   the echo and to formulate decisions concerning the   the values of the data symbols under the effect of echo-compensated signals; characterized in that it comprises reference training means (82)   which depend on individual decisions   formed by the signal processing means so as to form an estimation signal which is substantially   equal to the data component of an individual sample   dual mode, and means (22, 84, 85, 86) for generating an individual adaptation error signal under the effect of the estimation signal, the individual sample, and the image of its echo component.   3. Device according to claim 2, characterized   in that the reference training means comprise   estimation means (82b) for linearly combining the individual decisions to form the estimation signal.   4. Device according to claim 3, characterized   in that the estimation means perform the combination   linear multiplication by multiplying the individual decisions by respective coefficients and summing the resulting products.   5. Device according to claim 4, characterized   in that said adaptation error signal is practically   equal to an error signal multiplied by a para-   predetermined meter, this error signal being equal to the   combination of the estimation signal, the indi-   vidual and image of its echo component.   6. Device according to claim 5, characterized   in that the reference training means include   In addition, update means (86, 90, 117, 122, 124) are provided for updating the value of each   coefficients under the effect of the error signal.   7. Device according to claim 6, characterized   in that, in the updating means, the value of each of the coefficients is updated under the effect of   generates an error signal by a predetermined parameter.   8. Device according to claim 6, characterized   in that the updating means updates the value of each of the coefficients under the effect of the error signal product by that of the decisions with which   each coefficient has been multiplied.