DE4326007A1 - Method for coordinated digital transmission via a plurality of transmission paths - Google Patents

Abstract

A method for digital transmission via two or more transmission paths is indicated in which statistical dependencies between the interference which are superimposed on the individual received sub-signals are used to increase immunity to interference. This is done by forming estimated values for the interference in the next step using the interference determined in all transmission sub-systems for preceding steps and by subtracting these estimated interference values from the received signals. The method can be combined with known coordinated reception methods or can be used for direct minimisation of the residual interference. The coordinated transmission method can be implemented in a plurality of embodiments which are equivalent to one another, i.e. as interference prediction with cross-coupling between the transmission sub-systems, as a cross-coupled decorrelation of the interference with decision feedback equalisers and also as a generalised Tomlinson-Harashima precoding with cross-couplings at the transmitting and receiving ends. Further versions are represented by coordinated sequence estimation methods and status-reduced sequence estimation methods.

Description

1 Introduction, state of the art

The invention relates to digital message transmission methods, in which the Signal reaches the receiver via two or more transmission paths. An example for this purpose represents the digital multiple transmission with the conductor for the purpose a reduction in the signal bandwidth and thus also the signal attenuation of the data stream is split into two or more partial data streams, each via a line be transmitted to the receiver (examples: dual and triple duplex transmission via symmetrical lines for so-called HDSL or ADSL connections (HDSL: High Rate Digital subscriber line; ADSL: Asymmetric Digital Subscriber Line, see [1].) As further Examples can include directional radio connections split over several subchannels as well Diversity procedures and multipath propagation in digital radio transmission cited become. The method according to this invention uses statistical dependencies between the interference that is superimposed on the individual partial reception signals to increase the Reliability of the received digital message. This in turn becomes a Enlargement of the maximum bridgeable field length enables.

In [2, 3] methods for using a correlation of the disturbances in a dual Duplex transmission for HDSL presented. However, only statistical ties are used between the two faults at the same time, namely the time the decision about both currently available symbols. It is shown that through a linear transformation without memory on the receiving side a decorrelation of the simultaneous interference values can be achieved, whereby the signal to interference ratio in one subsystem is maximized and the other is minimized. By a corresponding transmission side inverse transformation eliminates their effect on the useful signals. On this way, two independent sub-transmission systems with maximum total capacity generated. This procedure can be used as a Karhunen-Loève transformation for correlated Random variables are identified and in this way to any number of partial signals be expanded immediately.

In the coordination procedure according to [2,3], however, there are no temporal statistical ties within the disturbance processes there are still links between the disturbance processes to different ones Points of time taken into account and exploited. Therefore, this is described below the term static coordination is used. The method according to the invention allows in contrast, the use of all mutual statistical dependencies between the individual faults, even at different times. That is why the term dynamic coordination was chosen for this. Statistical links to different For example, times arise from different damping and in particular phase responses of the individual transmission paths or the interference coupling mechanisms (Example: Different crosstalk frequency responses with symmetrical lines). In general, dynamic coordination according to the  Invention can achieve significantly higher signal-to-noise ratios than with the previously known static coordination.

2 Dynamic coordination 2.1 Specification of the procedure as a coordinated fault prediction

In the dynamic coordination according to the invention in the form of coordinated interference prediction, it is initially assumed that for all D sub-transmission systems in the receiver an equalization for the data signals which is free of impulse interference at the detection times is carried out, cf. [4, 5, 6]. If necessary, the freedom from impulse interference is enforced using common methods of adaptive equalization (zero forcing (IF) equalization, see eg [7, 12]). The statistical relationships within the D interference fractions n _0i (t) of the received signals e _0i (t) at the output of these equalizers are determined by their autocorrelation functions (AKF)

l _0ii (τ) = E {n _0i (t + τ) n _0i * (t)}; t, τ ∈ R i ∈ {1, 2,. . ., D}, (1)

their mutual ties through the cross-correlation functions (KKF)

l _0ii (τ) = E {n _0i (t + τ) n _0j * (t)}; i, j ∈ {1, 2,. . ., D}, (2)

described. (In the case of high-frequency, modulated signals, their equivalent complex low-pass signals are used for the description; z * denotes the conjugate complex number at z ∈ C. ) As long as the interference processes can be modeled as stationary Gaussian processes, the AKF and KKF will provide the entire statistical Bindings captured.

After sampling in the symbol cycle k · T with k ∈ Z (T: symbol interval), equivalent circuit diagram 1 results for the discrete-time received signals

e _i (k) = s _i (k) + n _i (k); i ∈ {1, 2,. . ., D}. (3)

The useful components s _i (k) of the time-discrete received signals pass undistorted from the transmitter to the decision-maker at the receiving end due to the pulse interference-free equalization. The discrete-time interfering signals n _i (k) are time-related and mutually correlated due to their process of origin and the equalization, which is due to their auto- or cross-correlation functions

l _ÿ (κ) = E {n _i (k + κ) n _j * (k)}; κ ∈ Z ; i, j, ∈ {1, 2,. . ., D} (4)

is expressed. (As long as the continuous-time disturbances n _0i (t) are at least weakly cyclostationary with the period T, the discrete-time disturbances can be represented as weakly stationary processes. Extending the representation to unsteady-state processes is easily possible.) The D parallel signals are used for simplification the representation of row vectors, e.g. B. (k) = (e₁ (k), e₂ (k),..., E _D (k)), summarized.

2.1.1 Coordinated disturbance prediction, combined with static coordination

After the decisions about the D partial signals on average κ, by subtracting the estimated useful signals _i (κ) from the received signals e _i (κ), measured values _i (κ) of the interference can be formed:

_i (κ) = e _i (κ) - _i (κ) (5)

(These measured values are correct in the case of error-free detections _i (κ) = s _i (κ). Since a very low probability of error is to be assumed for a usable transmission method, the measured values correspond to the actual interference values with very few exceptions.) measured values _j (κ) known from section k-1; κ ∈ {k-1, k-2,. . ., kp} are formed using linear predictor filters P _ji (z) prediction values ñ _j (k) for the disturbances n _i (k) for the subsequent step k and these are subtracted from the received signals e _i (k). When using non-recursive predictor filter P _ji (z) of degree p, the following applies, for example, to the prediction values:

A device according to block diagram 2 is therefore proposed for dynamic coordination in its form as a fault prediction, linked to static coordination. In contrast to the known interference prediction in digital transmission via linearly distorting transmission media (see, for example, [7, 8, 6]), not only previous interference measurement values _i (k-κ); κ ∈ N of the i-th partial transmission system is used to determine the interference prediction value ñ _i (k), but also the interference measurement values _j (k-κ); j ∈ {1, 2,. . ., D} of all other sub-transmission systems.

The predictor filters P _ÿ (z) should be selected so that the variances σ _ii ² of the residual disturbances n _i ′ (k) = n _i (k) - ñ _i (k) are minimized. As a result, after subtracting the prediction values ñ _i (k), only unpredictable, ie hardly correlated, disturbances n _i ′ (k) with lower power remain in the signals e _i ′ (k). The reliability of the digital transmission is thus increased. The filter coefficients can be set, for example, using the known methods for minimizing the mean square error, see e.g. B. [7, 12, 8]. (In particular, if the discrete-time disruptions are not stationary random processes, these methods can be used to adaptively adapt the predictor filters in a known manner.) If the AKF and KKF l _ÿ (κ) are known and time-discrete predictor filters P _ÿ (z) with FIR- If the structure of degree p is used, the filter coefficients p _ÿ (k) can be used as solutions of the systems of equations

determine (generalized Yule-Walker equations, [9]).  

With a sufficiently large predictor grade p, the coordinated perturbation prediction is used achieved according to this invention:

• • The autocorrelation within the residual perturbation processes n _i ′ (k) disappears with minimal residual variance because all predictable parts are subtracted.
• • The cross correlations between the residual disturbances n _i ′ (k) disappear at different times. The following applies: l _ÿ ′ (κ) = E {n _i ′ (k) n _j ′ * (k-κ)} = 0 for κ <0, (9) since all parts of the original disturbances n _i (k) that are predictable based on the previous interference measurement values _j (k-κ) using the predictor filter P _ÿ (z). The same reasoning applies to the residual disturbance n _j ′ (k): l _ji ′ (κ) = E {n _j ′ (k) n _i ′ * (k-κ)} = 0 for κ <0. (10) Because of the symmetry _relationship l _ÿ ′ (κ) = l _ji ′ * (- κ) (see e.g. [10]) the following applies:

This first version of the coordinated fault prediction thus achieves that

• The powers σ _ii ² of the residual disturbances are minimized within the framework of the given predictor filter,
• • the residual disturbances are only correlated at the same time.

If predictor filter P _ÿ (z) with FIR structure of degree p with the coefficients p _ÿ (k); k ∈ {1, 2,. . ., p} are used for the covariance σ _ÿ ² of the residual disturbances according to the coordinated disturbance prediction:

The following definition must be used to evaluate this relationship:

The definition (13) for (actually non-existent) 0th coefficients of the predictor filter corresponds to the direct paths of the received signals to the decision-makers, see Figure 2.

The covariances σ _ÿ ² of the residual perturbations n _i ′ (k) become the covariance matrix

N = (σ _ÿ ²) i, j ∈ {1, 2,. . ., D} (14)

summarized. The D² predictor filters also form a matrix:

P (z) = (P _ÿ (z)) i, j ∈ {1, 2,. . ., D}. (15)

For the received signals e '(k) after the subtraction of the interference prediction values, the time-discrete equivalent circuit 3 with (largely) white interference n _i ' (k) applies, which, however, are mutually correlated at the same times k. For the output signals e _i ′ (k) of the coordinated disturbance prediction, the requirements for the use of static coordination methods are met. The co-variances σ _ÿ ² of the simultaneous residual disturbances are mostly significantly increased by the coordinated disturbance prediction.

In addition to minimizing interference performance through coordinated interference prediction thus further gains in signal-to-noise ratio by linking them to static coordination processes can be achieved, which are often much higher than with a static Coordination without coordinated prediction of disturbance.

To carry out the static coordination, a Karhunen-Loève transformation for the received signals e _i ′ (k) according to the coordinated interference value prediction is proposed. This creates D partial replacement channels with mutually uncorrelated white interference. The hermetically symmetric covariance matrix N according to (14) has D real eigenvalues λ _i and D mutually orthogonal left eigenvectors _i , which are obtained by means of normalization

form an orthogonal base, they [11]. (With the notation = (Φ₁, Φ₂,..., Φ _D ), row vectors are represented.) The left-hand vectors become the modal matrix

summarized. The following applies to the modal matrix Φ ^{-1 of} the right-hand eigenvectors: Φ ^-1 = Φ * ^T. The covariance matrix can be broken down into

By means of a linear transformation of the received signals e _i ′ (k) according to the coordinated interference prediction into the signals

uncorrelated, white disturbances n " _i (k) with the variances λ _{i are} generated in all subchannels (Karhunen-Loève transformation).

In order to cancel this transformation with regard to the useful signals, a transmission side is inverse transformation

required. Figure 4 illustrates this procedure for generating uncorrelated sub-transmission systems. The different variances λ _{i of} the disturbances n " _i (k) can now be taken into account by an independent transmission with different partial rates R _i [bit / symbol]. The partial rates R _i for a given total rate R _G are preferably according to the following relationship

to be selected, while the partial transmission powers S _i = E {| s _i (k) | ²} should be approximately the same size, see [9].

The coordinated interference prediction was derived for the signals before the Karhunen-Loève transformation. For the z-transform of the received signals e " _i (k) after coordination

applies according to block diagrams 2 and 4 in vector notation

Figure 5 shows a modified block diagram of the coordinated transmission in the form of the coordinated interference prediction, linked to a static coordination using the Karhunen-Loève transformation. The signal paths each embody D parallel partial signals, the partial signals are linked to one another by the matrices Φ or Φ ^-1 and Y (z). As long as the decisions are made correctly, the transmission signals s " _i (k) and the reception signals e" _i (k) have distortion-free, independent partial replacement transmission channels with largely uncorrelated and white interference n " _i (k), the minimal but maximally different variances λ _i own.

2.1.2 Direct coordinated fault prediction

In a direct implementation of the coordinated fault prediction, the detour via static coordination (Karhunen-Loève transformation) is avoided. In addition to the measured values _i (k-κ); κ ∈ N of the disturbances for previous steps, certain current measured values _ν (k) are already used to predict other current disturbance values _µ (k). For this purpose, a suitable sequence ν (i) ∈ {1,. . ., D}; i ∈ {1,. . ., D} for simultaneous prediction. After determining an interference prediction value ñ _{ν (1)} (k) based on all previous interference measurement values _i (k-κ) using linear predictors Ξ _{i ν (1)} (z) = P _{i ν (1)} (z); i ∈ {1,. . ., D} according to section 2.1.1, the detection takes place via the partial signal e _{ν (1)} '(k), and the interference measurement value _{ν (1)} (k) according to (5) is formed. For the disturbance n (k) _{ν (2),} a prediction will now ñ _{ν (2)} (k) based on all previous Störmeßwerte _i (k-κ) and of the simultaneous Störmeßwertes _{ν (1)} (k) using a modified predictive filter Ξ _{i ν (2)} (z) determined. In this way it is achieved that the simultaneous residual disturbances n _{ν (1)} '(k) and n _{ν (2)} ' (k) are uncorrelated after the prediction of the disturbance. The simultaneous interference measurements _{ν (m)} (k), m ∈ {1,. . ., M-1} to determine the interference prediction value ñ _{ν (M)} (k). The coefficients ξ _ÿ (k) of the modified predictor filter

in FIR structure with degree p can be modified according to the following modified determine generalized Yule-Walker equations:

and the constraints according to the order ν (i)

ξ _{n (m) ν (i)} (0) = 0 for mi (25)

The predictor filters Ξ _ÿ (z) are combined to form the matrix Ξ (z). Since the current interference measurement values _i (k) are also coupled, in this version of the coordinated interference prediction the residual interference n _i ′ (k) are (largely) uncorrelated at the same time. The variances σ _ii ² of the residual perturbations can be calculated according to the relationships (12), using the values ξ _{μ i} (κ) according to (24) instead of the coefficients p _{μ i} (κ). The coefficients ξ _{μ i} (0); µ ≠ i are now largely different from zero according to (24). The values ξ _ii (0) = - 1 for the direct supply of the signals to the decisions are the same as for Eq. (12) to use. The covariance matrix N of the residual perturbations is essentially different from zero only in its main diagonal; Factoring and thus further static coordination are therefore not necessary.

In order to achieve high signal-to-noise ratio gains through coordination, the sequence ν (i) for the determination of the measured interference values should preferably be selected so that the geometric mean (or the product) of the variances σ _ii ² of the residual interference is minimal, cf. [9]:

Figure 6 shows the block diagram of the direct coordinated fault prediction. It is assumed that the decisions are made without delay and thus all decisions about the partial signals e _{ν (m)} ′ (k); m ∈ {1, 2,. . ., M-1} for the decisions about the partial signal e _{ν (M)} ′ (k) are already available. As long as the decisions on the D partial signals influence each other in a clear order ν (i); i = {1, 2,. . ., D} takes place, an unstable behavior of the decision feedback is excluded.

A further improvement is possible through a repeated, iterative calculation process of the current measured interference values. (Recalculation of _{ν (M)} (k) using the _{ν (m)} (k) already determined in previous iteration _loops ; also for m <M.)

The different interference variances in the uncorrelated replacement channels between transmitters and decision-makers can in turn preferably be taken into account by dividing the overall message flow into unequal partial rates R _i , cf. Section 2.1.1.

The transmission powers S _{i should} preferably be chosen approximately the same.

2.2 Coordinated equalization with feedback (CDFE)

Just like the usual interference prediction for digital transmission for one transmission path (D = 1) in an equivalent structure with noise whitening filter and subsequent Decision feedback equalization (DFE) can be transferred (see e.g. [6]) exist as further variants of the dynamic Coordination procedure with coordinated decision feedback equalization (Coordinated Decision-Feedback-Equalization: CDFE).

2.2.1 CDFE, linked to static coordination

The relationship (23) for the signals e " _i (k) at the inputs of the decision-makers can easily be transformed into

I denote a D × D unit matrix. The coordinated transmission method according to the invention in its form as CDFE, combined with static coordination according to Eq. (28) is shown in Figure 7B. It was envisaged that the sampling was preceded by an interference-free equalization by means of the reception _filter H _0i (f). Since the time-discrete systems W _ii (z = e ^{j2 π fT} ) in the range of low frequencies f <1 / (2T) often have largely inverse frequency responses to the equalizer _filters H _0i (f) (cf. [6]), it is _relevant sometimes inexpensive, simpler time-continuous reception filter

H _Ei (f) = H _0i (f) W _ii (z = e ^{j2 π fT} ) (29)

to realize. The sampling is then followed by cross-coupling filters W _ÿ (z) / W _ii (z); i ≠ j, see Figure 8. Of course, a complete time-continuous implementation of the cross-coupling and all mixed forms of time-discrete and time-continuous implementations are also possible. In the following, a memory-related D-dimensional signal transformation WM (f), which corresponds to noise band limitation and interference decorrelation with cross-coupling, is referred to as a coordinated whitened-matched filter, since the interference samples n " _i (k) at the output of such a system are both temporal in themselves and are as uncorrelated as possible. It applies

For the inversion of the transformation Φ ^-1 contained in WM (f), the transformation Φ according to section 2.1.1 must be provided on the transmission side. Figure 7a shows the implementation of the transmission process with CDFE, combined with static coordination when using a coordinated whitened-matched filter.

2.2.2 CDFE, direct form

For the direct form of the coordinated fault prediction, the decision input _signals e _i ′ (k) apply, see Figure 6:

Elementary reshaping results in the coordinated decision feedback equalization (CDFE) in direct form as shown in Figure 9:

The decision must again be made without delay; due to the directional influencing of the decisions according to the order ν (i), however, the stability of the decision feedback is guaranteed. For the time-continuous, coordinated whitened-matched filter WM _d (f), the following applies to the direct form of the CDFE:

Of course, all mixed forms are time-continuous and time-discrete implementation this D × D filter possible. A transmission-side transformation of the partial signals is not required.

2.3 Coupled sequence estimation methods

The characteristics of dynamic coordination as coordinated disturbance prediction according to Sections 2.1.1 and 2.1.2 as well as coordinated decision-based equalization according to sections 2.2.1 and 2.2.2 as well as all mixed forms from this can serve as a basis can be used for coupled sequence estimation methods. Doing so will not be instant Decisions made to determine interference measurements or decision feedback to allow, but the variances must be measured Disturbances (metrics) for different symbol sequences assumed to be sent calculated and (in the longer term) decided in favor of the symbol sequences for which the variance of the necessary disturbances is the smallest. With coupled sequence estimation metrics for combinations of D sequences are calculated and preferably from them made common decisions across all sequences. A variant of the coupled Sequence estimation is represented by a Viterbi algorithm (cf. [6]), the states of which the possible combinations of the D sequences are formed. In condition-reduced sequence estimation methods can by summarizing states into hyper states There is an exchange between effort and resistance to interference, cf. [6]. One more way the so-called selection or M algorithm represents a reduction in effort, see. [6]. The procedures according to sections 2.1 and 2.2 set out the special cases represent only a reduced state reduced sequence estimate.

2.4 Coordinated Tomlinson-Harashima precoding (CTH)

The processes of coordinated fault prediction and decision feedback Equalization has the disadvantage that decision errors about its feedback Subsequent errors arise. In addition, the decisions must be returned immediately be, resulting in the use of channel coding techniques that delay decisions condition is very difficult. These disadvantages are apparent in the dynamic Coordination as coordinated Tomlinson-Harashima precoding (CTH) avoided. In Tomlinson-Harashima pre-coding, instead of a reception-side, decision feedback Equalization (DFE) a pre-distortion on the transmission side.  

A disproportionate increase in the transmission power is avoided by a non-linear limitation on the transmitter, see e.g. B. [13, 6]. This non-linear limitation is equivalent to a periodic continuation of the signal constellation with multiple representation of the digital symbols by signal values. The representative is selected for which the transmission signal amplitude is smallest after the predistortion. In the expansion of the Tomlinson-Harashima precoding to the trellis precoding, the representatives are chosen such that long sequences of transmission signal values s _i (k) have desired properties (e.g. minimum mean power, spectral zeros, etc., see [14, 15]).

The CDFE according to this invention can be transformed in the same way as ordinary DFE (D = 1) into equivalent structures with Tomlinson-Harashima precoding or trellis precoding. Figure 10 shows the forms of the invention as CTH, linked with static coordination and in direct form. Although there are undelayed feedback paths in the direct form, the stability of the feedback is ensured due to the directional dependency in the selection of the optimal signal representatives. The selection of the cheapest signal representative can be made individually for each subsystem as well as connected for all subsystems (multi-dimensional pre-coding or signal shaping). The coordinated whitened-matched filters are to be implemented continuously or discrete-time, preferably in accordance with sections 2.2.1 and 2.2.2.

3 Bibliography

[1] IEEE Journal on Selected Areas in Communications, special issue August 1991 for Topic "High Rate Digital Subscriber Line", Vol. JSAC-9, Aug. 1991

[2] J. W. Lechleider: Exploiting crosstalk pair-to-pair correlation in two-pair digital subscriber lines. Proceedings of the IEEE Int. Commun. Conference, Atlanta, GA, 1990, paper 348.2

[3] J. W. Lechleider: Coordinated Transmission for Two-Pair Digital Subscriber Lines. IEEE J. Select. Areas in Commun., Vol. JSAC-9, pp. 920-930, Aug. 1991

[4] R. W. Lucky, J. Salz, E. J. Weldon: Principles of Data Communication. McGraw-Hill, New York, 1968

[5] G. Söder, K. Tröndle: Digital transmission systems. Springer Verlag, Berlin, Heidelberg, 1985

[6] J. Huber: Trellis coding. Springer Verlag, Berlin, Heidelberg, 1992

[7] M. L. Honig, D. G. Messerschmidt: Adaptive filters: structures, algorithms, and applications. Kluwer Academic Publishers, Boston, 1984

[8] J.G. Proakis: Digital Communications. McGraw-Hill, New York, 1989

[9] J. Huber, R. Fischer: Coordinated digital transmission, in preparation, 1993  

[10] A. Papoulis: Probability, Random Variables and Stochastic Processes. McGraw-Hill, Tokyo, 1965

[11] K. Burg, H. Haf, F. Wille: Higher Mathematics for Engineers, Volume II: Linear Algebra. Teubner publishing house, Stuttgart, 1987

[12] B. Widrow, S.D. Stearns: Adaptive Signal Processing, Prentice-Hall, Englewood Cliffs, N.J., 1985

[13] G. D. Forney, M.V. Eyuboglu: Combined equalization and coding using precoding. IEEE Communications Magazine, Vol. 29, Dec. 1991

[14] G.D. Forney, M.V. Eyuboglu: Trellis Precoding: Combined Coding, Precoding and Shaping for Intersymbol Interference Channels. IEEE Trans. Inform. Theory, Vol. IT- 38, March 1992

[15] J. Huber, R. Fischer: Signal shaping in digital message transmission. German Patent application P 43 16 547.8 dated May 18, 1993

4 List of figures

Figure 1: Time-discrete equivalent circuit diagram for parallel transmission via D signal paths with equalization and sampling at symbol clock.

Figure 2: Coordinated disturbance prediction, combined with static coordination.

Figure 3: Discrete-time equivalent circuit diagram for the received signals after subtraction of the interference estimates.

Figure 4: Generation of independent subchannels with uncorrelated, white interference through Karhunen-Loève transformation.

Figure 5: Modified form of coordinated disturbance prediction, combined with static coordination.

Figure 6: Direct form of the coordinated disturbance prediction.

Figure 7: Coordinated, feedback-based equalization with an upstream coordinated whitening filter, combined with static coordination.

Figure 8: Coordinated (noise) whitening filter.

Figure 9: Coordinated decision feedback equalization with an upstream coordinated whitening filter in the direct form.

Figure 10: Dynamic coordination, implemented as coordinated Tomlinson-Harashima pre-coding or trellis pre-coding.
A) linked to static coordination
B) direct form

Claims (8)

1. Method for digital message transmission over two or more signal paths with mutually statically dependent interferences for the use of these static dependencies for the purpose of increasing the reliability or the bridgeable field length, characterized in that the received signals are mutually coupled via dispersive systems in such a way that the interference for the individual signal paths are reduced as far as possible and the effect of this coupling on the useful signals is canceled by feeding back the signals at the output of the decision-makers at the receiving end and / or inverse coupling at the transmitting end. 2. The method according to claim 1, characterized in that from estimated interference estimates, those in previous steps by subtracting the estimated Useful signals were obtained from the received signals via mutually coupled, possibly dispersive, systems predictive values for the disturbances to the current Step are formed by the associated receive signals for the purpose the reduction of the interference effects are subtracted. 3. The method according to claim 1, characterized in that by means of the current Step and in previous steps based on the difference between received signals and estimated useful signals determined interference estimates via mutually coupled, if necessary, dispersive systems, predictive values for the other disturbances to be formed at the current step, for the purpose of reducing interference be subtracted from the associated received signals. 4. The method according to claim 1, characterized in that the received signals filtered in the way and mutually with each other via optionally dispersive Systems are coupled so that the interference is reduced as much as possible and static Dependencies of the interference values mutually and within the interference signals as possible as well as the effects of these filters and couplings to the useful signals through feedback and mutual coupling of the output signals of the decision-makers at the receiving end about possibly dispersive systems be saved as much as possible. 5. The method according to claim 1, characterized in that under several combinations of symbol sequences presumed to be sent, the one selected for which the performance of the difference between the associated useful signals and the observed reception signals after the mutual coupling of the signals to Interference reduction and for mutual decorrelation is smallest.   6. The method according to claim 1, characterized in that the received signals filtered in this way and mutually via possibly dispersive systems with one another be coupled that the disturbances are as small as possible and the statistical Dependencies of the interference values mutually and within the interference signals are canceled as possible and the effects of the filters and couplings on the useful signals are canceled by inverse filters and coupling on the transmission side, with a primarily multiple representation of the digital symbols by Signal curves or signal values through a suitable selection of the currently used Representatives desired properties of the transmission signals based on given criteria can be awarded. 7. The method according to claim 1 and 2, claim 1 and 4 and according to claim 1 and 6, characterized in that the transmitting and receiving sides are inverse to one another Couplings between the signals made using non-dispersive systems in order to make use of the remaining statistical dependencies simultaneous interference values to further reduce the interference and in particular differences in the performance of the remaining residual disorders as large as possible cause. 8. The method according to claim 1, claim 1, 2 and 7, claim 1 and 4, claim 1, 4 and 7, claims 1 and 5, claims 1 and 6, claims 1, 6 and 7, thereby characterized in that the digital transmission by means of different data rates, Symbol alphabets and waveforms or signal values the different Interference power of the sub-transmission systems is adjusted, which may already be originally existed and / or through the mutual coupling of the signals for the purpose of reducing the interference and generating a possible large spread of interference was produced.