DE3610383A1 - Adaptive decision feedback equaliser - Google Patents

Abstract

The invention relates to an adaptive decision feedback equaliser for devices in electrical communication systems with a first filter designed as an adaptive transverse filter (AF). In high-speed digital signal transmission, the line signal (xL) arriving in a device has so-called post-oscillations due to shortcomings in the transmission path. Such unwanted post-oscillations can be cancelled with an adaptive decision feedback equaliser. If relatively long post-oscillations are intended to be completely cancelled with an echo canceller of this type comprising a single adaptive transverse filter, this filter must have an economically unviable number of elements. According to the invention, a second non-adaptively operated filter is connected downstream of the first filter (AF) in such a way that the output of the last multiplier (Mn) of the first filter (AF) is connected to the input of the second filter (F) (Fig. 1). <IMAGE>

Description

The invention relates to an adaptive decision feedback Equalizer for electrical communications equipment according to the preambles of the claims 1 and 4.

Such an equalizer is from Figures 3 or 4 and that accompanying text of the German magazine "Archiv der electrical transmission "Volume 34 (1980) Issue 7/8, page 287 to 292 known.

In fast digital signal transmission, this indicates in line signal arriving at a device due to imperfections so-called post-oscillators on the transmission path. Such undesirable post-oscillation can be done with an adaptive decision feedback equalizers can be compensated. With such a, a single adaptive transversal filter contained equalizer for a relatively long time Post-oscillator must be completely compensated for, so this filter an economically unsustainable number of links exhibit. The delay elements, the associated multipliers and other associated assemblies, e.g. B. understood the coefficient calculation.

The invention is based on the task, a little complex adaptive decision feedback equalizer of the type mentioned at the beginning.

This task is characterized by the characteristics of Claims 1 and 4 solved. Give the subclaims advantageous further training.

From images 4 and 5 and the associated texts pages 302 to 309 of the German magazine "Frequency", Volume 36 (1982), Issue 11, it is related to a Echo canceller known per se, a first, as adaptive Transversal filter trained filter a second filter downstream, which is designed as a recursive filter is. This recursive filter also works adaptively.

However, this is associated with a great deal of effort because of the facilities and necessary for the adaptation because of the measures to make the known instability more adaptive to prevent recursive filter.

The invention is described with reference to FIGS. 1 to 6 and three exemplary embodiments, these being assigned to the claims according to the table.

The embodiment applies mutatis mutandis to claim 2 1.

Figs. 1, 2 and 4 enter the hold. Equalizer itself again, while their functions with reference to FIG. 3 and will be described. 5

Embodiment 1 according to FIG. 1 is first described. It means:
1 x _L : a line signal
x _k : an equalized signal
ES : a decision maker
x _e : a decision maker output signal
DG : a differential element
T ₁ to T _n : delay elements
M ₁ to M _n : multiplier
k ₁ to k _n : coefficients
KR : a coefficient calculator
y ₁ to y _n : multiplier output values
Su : a totalizer
d : a compensation signal
Sb : a subtractor
Q : a second filter
one
first,
adaptive
Transversal filter
AF

The line signal x _L is sent by a remote station, not shown, and reaches the plus input of the subtractor Sb in the equalizer shown here via a transmission path, also not shown. As a result of imperfections in the transmission path, the line signal x _{L has} so-called. Resonator on.

If one first thinks the second filter F away and connects the output of the last multiplier M _n directly to the summer Su , the result is the known equalizer constructed from a single adaptive transversal filter AF . Its function is as follows: The compensation input d is fed to the minus input of the subtractor Sb , by means of which the ringing oscillators are compensated. In this way, the equalized signal x _k , that is to say largely freed of post-oscillators, arises at the output of the subtractor Sb . This is fed to the decision maker ES , which delivers the decision maker output signal x _e at its output. In a manner not shown, this reaches any signal receiver for further processing.

The decision output signal x _e is fed to the multipliers M ₁ to M _n , specifically via the delay chain constructed from the delay elements T ₁ to T _n , all delay elements having the same delay time T. The delayed decision output signal x _e is multiplied in the multipliers by the associated coefficients k ₁ to k _n . The multiplier output values y ₁ to y _n thus formed are summed in the summer Su to the compensation signal d , which is fed to the minus input of the subtractor Sb .

The equalized signal x _k and the discriminator output signal x _e are supplied to the plus or minus input of the differential element DG, which gets from the error signal f. This is fed to the coefficient calculator KR . This determines the coefficients k ₁ to k _{n in} such a way that, in the parts of the transversal filter that can be compensated for on the basis of the number of its members, the compensation signal d is as close as possible to the relevant oscillation part. Then, the equalized signal x _{k is} largely freed of ringing, and the transversal filter is in the adapted state.

The adaptive transversal filter AF described above is followed by the second filter F. The transversal filter AF is therefore the first filter. This is followed by the second filter F in such a way that the output of the last multiplier M _{n is} connected to the input of the second filter F. The output of the second filter F is connected to the summer Su . It is thereby achieved that the output values of the second filter F also contribute to the formation of the compensation signal d . Details will be described later using embodiment 2.

The second filter F has the transfer function

on. Here, a _μ , μ = 0.1. . . m ; c _λ , λ = 0.1. . . m fixed coefficients, ie the second filter F is not operated adaptively. The degree of the filter is specified with m .

A transversal filter results for H _N ( z ) = 1, and a purely recursive filter is available for H _P ( z ) = 1.

Claim 2 is directed to an equalizer in which the second filter F is a purely recursive filter of degree m . The transfer function is:

The claim 3 relates to an equalizer, in which the second filter is also a recursive filter, however  has grade 1. The transfer function is:

Such an equalizer is described with reference to FIGS. 2 and 3. The description given for FIG. 1 applies here analogously. The second filter comprises the adder Ad , the delay element T _R , the multiplier M _R and the memory Sp . It is connected downstream of the first filter in such a way that its adder Ad is looped into the connection from the output of the last multiplier M _{n of} the transversal filter to the summer Su . The recursive filter is a first degree filter because it has only one delay element and one multiplier. Its multiplier M _R is operated with a fixed decay factor c _k stored in the memory Sp , which is less than one.

The transversal filter and the recursive filter are a time-discrete system, ie the summer Su only outputs compensation values d _ν at a time interval T , these values being able to be maintained between the sampling times and thus resulting in a step-shaped course of the compensation signal d . The line signal x _L thus only experiences the required compensation at a time interval T. Only at these points in time does the decision maker ES evaluate the equalized signal x _e , and only at these points in time is the equality of the post-oscillation component and the compensation signal d sought.

The other functions are explained with reference to FIG. 3. In it, the time t is plotted on the abscissa and on the ordinate d _n, the normalized amplitudes of the compensation values of the line signal and the equalized signal x _L x _k applied. The solid curve shows a section of the line signal x _L , provided that a Dirac pulse was sent in the remote station. In this case, the line signal x _{L is} the impulse response of the transmission path, and its imperfections are shown by the fact that this impulse response has, among other things, a post-oscillation. From this post-oscillations a detail indicated by the labeled x _L curve.

This post-oscillator is so long in this case that it is from Transversal filter alone because of the limited number of its Limbs cannot be fully compensated. That is, the from Transversal filter formed compensation window includes only part of the post-oscillator.

From the compensation values d _ν, the transversal filter generates those for ν = 1 to ν = n . Since only a section is shown here, only those for ν = n - 1 and ν = n are shown. Provided that a Dirac burst was sent in the remote station, the compensation values provided by the transversal filter are equal to the corresponding multiplier output values y ₁ to y _n .

The compensation values d _ν for ν ≦ λτ n are obtained from the recursive filter according to the relationship

d _{n + i} = y _n · c _k ⁱ

generated because its adder Ad is inserted in the connection between the output of the last multiplier M _n and the summer Su . Since y _n = d _n applies in the special case shown here (Dirac impact), the following also applies here

d _{n + i}  =d _n  ·c _k ⁱ (i = 1, 2,. . .)

Based on this relationship, the compensation values d _{n +1} to d _{n +8 were} determined for FIG. 3, starting from d _n = 0.1, the decay factor being chosen to be exaggeratedly small with c _k = 0.6 for the sake of clarity. In reality, you choose it from the range between 0.75 and 0.95.

In the special case shown here (Dirac shock), the equalized signal x _k contains the residual post-oscillation that remains after the compensation, which is drawn as a dashed curve. It was assumed that the transversal filter is in the adapted state and that, as a special case, its compensation values d ₁ to d _n (only the last two of which are shown) are equal to the respective values of the line signal x _L. It follows that within the compensation window the residual post-oscillator is zero. In Fig. 3 only the end of this compensation window appears, which is at t = n · T. Outside this compensation window, that is, at t ≧ ( n + 1) T , the residual post-oscillator is only exceptionally zero, namely if the post-oscillator coincides with the compensation values of the recursive filter. As a rule, as shown in dashed lines in FIG. 3, a residual post-oscillation deviating from zero will result. It is no longer shown that in the range t ≦ λτ ( n + 8) T the post-oscillation and the compensation values d _ν (for ν ≦ λτ n + 8) decay to zero and that the residual post-oscillation also decays to zero.

The advantage of the invention is that the transversal filter does not have to have as many elements in order to fully compensate for the longest occurring post-oscillator, ie the compensation window can be smaller than the longest post-oscillator. Since, according to the invention, the first compensation value d _{n +1 of} the recursive filter is obtained from the last multiplier output value y _{n of} the transversal filter in the case of a post-oscillation which projects beyond the compensation window, only when the decay factor lies within the range specified above, only a slow and moderate one occurs , the increase in the residual post-oscillation, which does not yet impair the transmission quality. Nevertheless, it is advantageous to choose a "favorable" decay factor, ie one in which the residual post-oscillation is as small as possible.

For the determination and storage of such a favorable decay factor c _k , one of the possibilities is to determine one from examinations of the transmission paths in question and to set this permanently in the memory Sp during manufacture of the device. The other possibility is to make the memory Sp adjustable. Thus, the maintenance personnel can set a favorable decay factor during commissioning or later by setting the decay factor for as long as. B. changed in steps until the eye diagram shows the largest possible opening or the smallest possible bit error number on a bit error measuring device.

The above explanations apply mutatis mutandis to embodiment 1.

If the compensation values d _{ν of} the recursive filter are to be adapted more precisely to the post-oscillators which result from different transmission paths, an equalizer according to claim 4 is more favorable. An exemplary embodiment of this is described with reference to FIGS. 4 and 5, in which only the differences from exemplary embodiment 1 are referred to below.

Instead of the memory for the decay factor, a decay factor computer AR is provided, to which the two last coefficients k _{n -1} and k _{n of} the transversal filter are applied. From this, he calculates the variable decay factor c _v according to the relationship:

This relationship applies under the generally applicable assumption that the latter delay element T _{n of} the transversal filter and the delay element T _{R of} the recursive filter have the same delay times. The multiplier M _{R is} operated with the decay factor c _v obtained in this way. The transfer function then reads accordingly:

The effect of the above measure is explained with reference to FIG. 5. It agrees with FIG. 3 in the following details:

• a) in the amplitudes of the last two compensation values d _{n -1} and d _n . The following applies:
  d _{n -1} = 0.125 and d _n = 0.1.
• b) in the curve of the line signal x _L

The compensation values d _{n -1} and d _n have the same values as the coefficients k _{n -1} and k _n in the case of a Dirac shock as the transmission signal. It can thus be shown on the basis of the compensation values d _{n -1} and d _n that in this example the decay factor calculator AR calculates a decay factor c _v of 0.8. The compensation values of the recursive filter result from this decay factor and the last compensation value d _n , as described in exemplary embodiment 1.

Calculated by the decay factor according to the invention compared to embodiment 2 a better one Match the compensation values of the recursive filter with the actual course of the ringers and thus smaller residual post-oscillators.

Based onFig. 6 becomes an embodiment of the decay factor calculator AR described according to claim 5. It has two entrancesE 1 andE 2nd on over which the coefficients k _{n -1} respectively.k _n  a first or a second averager MW 1 respectively.MW 2nd are fed from it Averages _{n -1} respectively. _n  form. These averages will be a first input of a multiplierMU or the plus Input of a differential elementDG ′ fed. The exit of the multiplierMU is with the minus input of the Differential elementDG ′ connected. Its exit is with a first input of a comparatorV connected. One second input of this comparatorV becomes a threshold Sch fed. The output of the comparator is one first input of an AND circuit1 connected. On theirs the second input is a clockCL. The outcome of this and- circuit1 is with the control input of a decay factor memory AFS connected. In it are in a cyclical Store different decay factors from the in question upcoming value range saved. Here it is the values  0.75; 0.8; 0.85; 0.9; 0.95. Depending on the position of the cyclical One of these decay factors appears in memoryc _v  at the exit of the decay factor memoryAFS and therefore also on the second Input of the multiplierMU. About the exitA of Decay factor calculatorAR reaches this decay factorc _v   also, like in theFig. 4 is shown at the entrance of the multiplierM _R .

In the multiplierMU becomes the mean _{n -1} with the respective Decay factorc _v  multiplied. In the differential element DG ′ is the product obtained and the mean k _n  a differenced ′ won which in the comparatorV With the thresholdSch is compared. Is the differenced ′  less than the thresholdSch, takes the output of the comparator V the value logically "0" and none arrive Clock pulses of the clockCL about the AND circuit1 to the Control input of the decay factor memoryAFS. The cyclical Speicher maintains its position.

However, if the difference d 'is greater than the threshold value Sch , the output of the comparator V assumes the value logic "1", the AND circuit 1 switches the clock CL through to the control input of the decay factor memory AFS . With each clock pulse, the cyclic memory is advanced by one position, and the next stored decay factor appears at the output of the decay factor memory AFS . This switching continues until the difference d 'is smaller than the threshold value Sch and the AND circuit 1 blocks the clock CL . The last appearing decay factor then appears continuously.

Due to the adaptation process of the adaptive transversal filter, the coefficients k _{n -1} and k _n fluctuate around their mean values in the adapted state. These fluctuations are suppressed by the mean value formers MW 1 and MW 2 and therefore do not influence the decay factor.

Another advantage of this decay factor calculator is that one required according to claim 4 for determining the decay factor and division that can only be carried out with great effort is avoided.

When using the equalizer according to the invention, the number of elements of the adaptive transversal filter is chosen so large that after the end of its compensation window the amplitude of the post-oscillation decreases, as shown in FIGS. 3 and 5. That is, the last coefficient k _n is less than the penultimate coefficient k _{n -1} , the division according to claim 4 leads to a decay factor which is less than 1, and the compensation values d _{ν of} the recursive filter sound asymptotic in the desired manner to zero. However, if, contrary to expectations and perhaps only briefly, the last coefficient is greater than the penultimate one, and the division gives a "decay factor" greater than one, then the execution of the decay factor calculator according to claim 5 also ensures the desired decay of the compensation values of the recursive in this case Filters to zero, because the decay factor memory AFS only stores decay factors that are less than one, and therefore only those can reach the multiplier M _R.

The invention is well suited for use on the subscriber side and exchange-side transmission modules the future so-called integrated digital telecommunications network, also known by the abbreviation "ISDN". In this new network is said to be that of the conventional telephone network existing subscriber cables can be used around the Avoid laying out new cables. The one about the so obtained  However, signals transmitted in transmission paths are proportionate long ringers. By applying this The invention achieves a satisfactory compensation of the post-oscillators with economic effort, so that these transmission routes for the digital signal transmission provided in the new network become usable with high bit rate.

The small number of elements of the adaptive transversal filter also enables a high rate of convergence.

Claims (5)

1. Adaptive decision feedback equalizer for devices of electrical communication with a first filter, designed as an adaptive transversal filter ( AF ), characterized in that the first filter has a second filter ( F ) with the transfer function is connected in such a way that the output of the last multiplier ( M _n ) of the first filter ( AF ) is connected to the input of the second filter ( F ), and that the coefficients ( a _μ , μ = 0, 1. . m ; c _λ , λ = 1, 2 ... m ) have fixed values. 2. Adaptive decision feedback equalizer according to claim 1, characterized in that the second filter ( F ) the transfer function having. 3. Adaptive decision feedback equalizer according to claim 2, characterized in that the second filter has the transfer function and is connected downstream of the first filter in such a way that the output of the last multiplier ( M _n ) of the first filter is connected to an input (+) of the adder ( Ad ) of the second filter, and that the decay factor ( c _k ) has a fixed value. 4. Adaptive decision feedback equalizer for devices for electrical communication with a first filter, designed as an adaptive transversal filter, characterized in that the first filter is followed by a second filter, designed as a recursive filter of the first degree, that the second filter performs the transfer function and is connected downstream of the first filter in such a way that the output of the last multiplier ( M _n ) of the first filter is connected to an input (+) of the adder ( AD ) of the second filter in such a way that a decay factor calculator ( AR ) is provided , which is loaded with the last two coefficients ( k _{n -1} , k _n ) of the first filter and from which the decay factor ( c _v ) of the second filter is calculated, where: 5. Adaptive decision feedback equalizer according to claim 4, characterized in that in the decay factor calculator ( AR ) the penultimate and last coefficient ( k _{n -1} and k _n ) a first and second averager ( MW 1 and MW 2 ) are supplied, the outputs are connected to an input of a multiplier ( MU ) or to the plus input of a differential element ( DG ′ ), that the output of the multiplier ( MU ) is connected to the negative input of the differential element ( DG ′ ), that its output is connected to a first input of a comparator ( V ) is connected, that its second input is supplied with a threshold value ( Sch ) and its output is connected to a first input of an AND circuit ( 1 ) that its second input is supplied with a clock ( CL ) and whose output is connected to a control input of a decay factor memory ( AFS ), that its output with the second input of the multiplier ( MU ) and with the multiplier of the recu rsive filter ( M _R ) is connected that the comparator ( V ) emits a logical "yes" signal at its output if the difference ( d ′ ) at the output of the differential element ( DG ′ ) is greater than the threshold value ( Sch ) that different decay factors are stored in the decay factor memory ( AFS ), that these decay factors appear in succession at the output of the decay factor memory ( AFS ), as long as the clock ( CL ) is present at the control input, that the last decay factor appears continuously, as long as the clock ( CL ) is blocked by the AND circuit ( 1 ), and that the stored decay factors are less than one ( Fig. 6).