FI92356B - Method for determining coefficients in a transverse equalizer and a transverse equalizer - Google Patents

Description

5 92356

Method for determining the coefficients of a transverse equalizer and a transverse equalizer

The invention relates to a method for determining the coefficients of a transverse equalizer and to a transverse equalizer. The method (i) transmits a predetermined training signal through a transmission channel, (ii) generates a sum signal sample in equalizer 10 by summing the received signal samples from the equalizer intermediate outputs, each weighted by its own pin factor, and (iii) generates an error signal representative of said sum signal and error signal. The transverse equalizers according to the invention are in turn according to the preamble of the appended claims 10 and 11.

The method according to the invention is intended to be used in particular for HDTV reception, but it can also be applied at least for the reception of a standard D-MAC or D2-MAC signal. Other possible application areas are fixed or mobile radio equalizers, and data transmission in general.

In synchronous communication systems, the data to be transmitted is in the form of a bit sequence. In the transmitter, the bits are converted into signaling symbols, which are sent to the communication channel at a certain signaling rate 1 / T, where T is the symbol interval. The symbols received at the receiver are detected and converted back to bit sequence. The signal transmitted in the communication channel deteriorates due to various interferences. These include e.g. amplitude and delay distortions that cause interaction between symbols, and noise. To reduce these problems, the receiver of the communication system is often equipped with an equalizer designed to minimize interference, mean square error, and the like prior to detecting the data to be received. One commonly used equalizer is the so-called a transver catch equalizer whose pin coefficients are adjusted so that the desired optimization criterion is achieved. The regulation usually uses the so-called a training signal which is a predetermined cyclic symbol sequence preceding the data transmitted to the transmission channel, by means of which the initial values of the pin coefficients are sought to be adjusted as close as possible to the optimum values.

10 There are many types of optimization criteria for the pin coefficients of equalizers referred to above. Two commonly used criteria are the least squares error method (LMS, Least Mean Square or LMSE, Least Mean Square Error) and the zero forcing method (ZF, Zero Forcing). The former method tends to minimize the mean square error, and the latter in turn tends to adjust the interaction at zero given points. Both control methods are described in more detail, for example, in reference [1] (the reference list is at the end of the explanatory section).

20 Known solutions specifically for video signal correction are also based on the use of (baseband linear) equalizers. Reference [2] describes an HDTV transmission chain using a transverse equalizer in which the least squares method is used as the adjustment criterion. 25 The disadvantage of the least squares error method is that • sample values outside the desired signal sample all gain the same weight when calculating the error. However, a sample value that is far from the desired sample time (e.g., due to reflection) is subjectively more disturbing than a sample value of the same size that is close in time (Ref. [3]).

With the zero forcing method, the interaction terms can be set to zero in a limited number of sample points. The amount depends on the number of adjustable pins of the transverse equalizer used. Typical of the method li 92356 3 is that when adjusting the sample values to zero, the out-of-range sample values become significantly large. This cannot be detected by the zero-forcing method.

Thus, the problem with receiving a video signal in particular is how the correction could also take into account the actual visibility of the error on the screen. The object of the invention is therefore to provide a method by means of which this subjective disturbance of the error can be taken into account and the disadvantages described above can be avoided. This is achieved by the method according to the invention, which is characterized by what is described in the characterizing parts of the appended claims 1 and 8.

The basic idea of the invention is to adjust the pin coefficients of the transverse equalizer so that the systematic error (interaction) and random error (e.g. noise) caused by the signal content 15 are minimized simultaneously, but still so that their mutual weight can be selected. Thanks to such a solution, the adjusting force can in each case be used subjectively against the most disturbing situations.

The solution according to the invention can also avoid large "external" sample values of the zero forcing method, since the solution also includes a control based on the square error.

The method according to the invention is particularly suitable for use in cable ducts for eliminating reflections, or in general for situations in which the approximate location of the expected reflection components relative to the main component is known. However, in the absence of significant reflections, the method still provides good control against random interference, thanks to the square error control it contains.

The operation of the method can be characterized by saying that it converts the interfering single visible component (e.g., reflection) into a slightly elevated random perturbation that is distributed over several sample points. The visibility of such interference in the image is lower than that of a sharp and high individual interference sample. With the method of the invention, the interfering single sample is in a way converted into low noise, which is spread over a wider area.

In the following, the invention and its preferred embodiments will be described in more detail with reference to the examples according to the accompanying drawings, in which Fig. 1 shows the general principle of a transverse equalizer, Fig. 2 shows a first embodiment of an equalizer according to the invention, and Fig. 3 shows a second embodiment of an equalizer.

The general principle of the transverse equalizer is shown in Fig. 1. In the equalizer 10 of Fig. 1, signal samples rk taken from the signal received from the transmission channel come from the left to a shift register formed by successive delay elements 11 20 (delay magnitude T) with intermediate outputs 12 at each delay element. The signal samples are weighted by coefficients C3, where the sum signal sample yk at a given time is 25 = (1) where k is the index of the current sample (index k corresponds to the desired signal sample at each time), and -Mlijs + M2 when M1 + M2 + 1 is the coefficient (pins) ) total number 30. The signal yk after the equalizer is subtracted from the training signal sk transmitted in the summing unit 13, which is thus known to the receiver and which is generated at the receiver. The obtained difference signal ek represents an error. This error can be represented as a (truncated) series development 35 with a signal-dependent part and a signal-dependent 5 92356 non-pum part, ie ek = ^ al8k-1 + Pk (2) 5 where the coefficients a * are later (formula (5)) the coefficients of the series to be determined, and -Nsi ^ + N, N is the integer to which the series development is truncated (in a complete series development, i would, of course, be between -oo <i <+ oc).

10 Assume that the symbols of the training signal sk are uncorrelated and have an average of zero (in practice, the training signal is a rather pseudo-random sequence with a correlation of about zero), i.e. ~ s ^ l = 0 if k * 1 and ~ iTk = 0. (3) 15

In Equation (3) (as in the following formulas), the underscore indicates the statistical expectation value. The selection of the coefficients a ± in terms of least squares is performed, whereby the expression 20 _ ^ aisjc-i) 2 (4) is derived with respect to the coefficients a ±. The solution is: 25 - = f = - (5) • 3k-l This choice of multiples ax thus gives the best fit when the error is to be represented as a function of the teaching signal values sk. This choice is known to result in 30 independent parts pk being orthogonal to the teaching signal samples, i.e. ρλ sJt_i = 0 for all k and i. (6)

The development made (Equation 2) can be interpreted to represent error 6 92356 as the sum of the interference dependent part between the symbol samples and the random part (e.g. due to noise).

Next, an adjustment instruction for the coefficients Cj is sought, taking into account that in a video signal, the visibility and interference of interference between symbol samples (e.g., transmission path echo) is generally greater the farther it occurs from the desired signal sample. For this purpose, the error formula (2) 10 is developed in a new form. The squared expected value of the error is «7 = al3k-i + Pk) 2 (?) 15 - sTi + P * / (8) where the uncorrelation of the symbols sk and the orthogonality of p have been exploited in the transition to the latter form (formulas (3) and ( 6)).

20 In Equation (8), the sum expression represents the signal-dependent average interference and the latter term represents the random average error.

Assuming that the subjective interference of the systematic (signal-dependent) error is worse than that of the random part, it is preferable to weight these factors differently from the above. In this case, a new objective function is defined for minimization. The minimization of the objective function is done by adjusting the coefficients Cj of the equalizer. Let the new objective function be 30 e7 = ^ * \ sTi + ”p Pk / (9) where the weighting factors Wi and wp can be chosen to correspond to subjective perturbation. Typically, wA is small 35 when i is close to zero and greater at large absolute values of index i li 92356 7 (the farther from the desired signal sample, the larger i).

The coefficients C ^ can now be derived by deriving the objective function (9) for each coefficient Ci and noting the result as zero: =? Wi ^ + 2 Ρ * Τ £, = 0 (10) In practice, however, the coefficients C3 are not worth the equation (10). Instead, the obtained partial derivative results can be used to implement, for example, a gradient algorithm known per se. (The gradient algorithm is a calculation algorithm commonly used in pin coefficient control, which is described in more detail in reference [1].) 15 Equation (10) is obtained by performing the marked calculations and taking into account the uncorrelation of the reference signal symbols and using the above dependencies. “Vp) zk-j sk-i * k sk-i + 2 tk zk.j wp. (H) 1 Ski

If a gradient algorithm is used to generate the coefficients, the new coefficient can be calculated from the old one by correcting it in the opposite direction of the gradient (cf. e.g.

25 reference [1] p. 140). Then a new value C3 (k + 1) is obtained as follows: 1. C] (k + 1) (ir) ^ wi ~ ^ rk-j sk-i «1 ^ 1 +» (12)

Sk7 30 where μ is the step size parameter. This equation already allows for practical hardware implementations, although calculating expectation values requires both memory and computing power. Due to the complexity of the device design, formula (12) thus does not provide the best possible practical solution.

35 However, the result of formula (12) can be simplified by realizing the set minimization of the set objective function. In this case, only the terms that have the greatest impact on the final result are included in the calculation. At its simplest, the terms where i = j are taken, because in a conventional reasonably good transmission channel, these terms predominate. Formula (12) is then given the form ^ (* + 1) = <^ {*) - μ [^ {wrwp) sk_j tk + tk rk_j wp] (13) sk 10 At this stage it can be stated that formula (13) is a combination of square error minimizing control (MSE) and zero coercion control (ZF). If all the weighing coefficients w are equal, a normal square error minimizing adjustment is obtained, as it should be.

15 A new value is thus obtained for each adjustable pin coefficient by correcting the old value for the first correlation of the received signal samples corresponding to the intermediate output of that pin coefficient and the error signal samples corresponding to the desired sample. through. (Referring to the term s ^ in this context (a-nalogically with the term rk-j) reference is made to a training signal sample corresponding to an adjustable coefficient interval, although - training signal samples are not obtained from the same intermediate outputs as received signal samples, and not necessarily from a similar signal sample. the weighting factor wp to choose as one, while the coefficients w3 may deviate in one direction or another from one. Furthermore, it can be said that the term r ^ s * .., has the same value regardless of js. In this case, the algorithm for calculating the equalizer coefficients takes the form: »92356 9

Cjik + l) = Cjik) - μ [g {wj-l) tk + tk rH], (14) where the first term inside the square bracket expression represents zero-forced type control and the latter ne-5 slip error control. Combining the coefficients gives a new form that is more advantageous for practical implementation:

Cjik + 1) = Cj (k) - Gj tk - G0 ek rk.j], (15) where the value of Gj can vary depending on which multiplier Cj is adjusted. The second coefficient G0 is the same for all coefficients C.,. In practice, the weighing factors Gj and G0 (or their ratio) are determined experimentally according to the visibility of the disturbance on the screen.

The principle of the equalizer implementing Equation (15) is shown in Fig. 2. For the sake of clarity, only the formation of one coefficient Cj is plotted in the figure, but it should be noted that the other coefficients are formed by the same principle. In this case, the equalizer first comprises the basic part 21 described above, from which a sample rk-j representative of the received signal 20 and an error signal sample ek are obtained. These signal samples are multiplied by a coefficient unit 22. The teaching signal sample sk_., Is obtained from the shift register 23 from its own intermediate output, and is multiplied by an error signal sample in a coefficient unit 24. Each coefficient unit is; 25 connected to its own correlator unit 25a and 25b, respectively. expected values of the above revenues. These correlations are weighed with their own weighing coefficients G0 and Gj, respectively, in coefficient units 26a and resp. 26b, after which the weighed correlations are summed in the summing unit 27 and the obtained sum is inverted in the inverting unit 28. The obtained result is summed to the old value of the pin coefficient C.j (k) in the summing unit 29 and the obtained new value is given to said tap coefficient Cj.

92356 10 For a practical solution, the solution according to Figure 2 can be further simplified. Starting from formula (14) and making a simplification known per se, replacing the statistical mean with an instantaneous value of 5 (in which case the formation of Cj by low-pass filtering or averaging also takes care of the averages in question)

Cj (k + 1) = Cj (k) - μ (aj tk s ^ + tk rk.j), (16) where a3 is the coefficient g (Wj-l), which may vary depending on the weighting used for different values of j . This solution is shown in Fig. 3, in which, as in Fig. 2 above, the formation of only one pin coefficient Cj is shown. The other pin coefficients are again formed on the same principle. In this solution, a sample rk_j representing the signal received in the summing unit 31 is summed and the training signal sample sk_j multiplied by the weighing factor 15 in the multiplier unit 32, which is again obtained from its own intermediate output from the shift register 33. The sum thus obtained is coupled to the coefficient 34. The result obtained is coupled to a coefficient unit 35, where it is loaded with a negative step size parameter -μ. The obtained result is summed with the old value Cj (k) of the pin coefficient Cj in the summing unit 36, and the obtained new value is given to the adjustable pin coefficient Cj.

The above inventive idea can be applied in several alternative ways. A particularly advantageous implementation is obtained by selecting Wj (formula (14)) as one with small values of | j |. Then the corresponding Gj (formula (15)) and the corresponding aj (formula (16)) are zero, so that the implementation for these pin coefficients is simple, corresponding to the usual square error adjustment. For other tap pic-factors, Wj> l is selected. These other Wj? T may be equal to or different from each other, depending on the disturbance of the error on the screen, and the resulting weighing effect as a result.

11 9 '/ ό 56

If the interference consists mainly of delayed reflections, as is the case, for example, in cable distribution, it may be advantageous to choose WjSt to differ from one only by positive values of j, or e.g. so that a }s are 5 zeros when jsJ, where J is a small positive index, otherwise a3s are equal positive numbers. A more general weighing rule in this case could be such that a3st increase as j increases. This means that the weighing factor increases further from the respective desired sample by the corresponding signal sample.

One conventional implementation option for this type of equalizer is also to quantize some or all of the signals to simplify the computational work. In the case of the solution according to the invention, the first simplification could be to quantize the error signal ε * to two levels, thus taking into account only the sign of the error. This procedure is known to slow down the speed of adjustment somewhat, but instead saves on circuit complexity.

Although the invention has been described above with reference to the example according to the accompanying drawing, it is clear that the invention is not limited thereto, but can be modified in many ways within the scope of the inventive idea set forth above and in the appended claims.

25 For example, the manner in which a training signal is generated at a receiver can vary in many ways.

Reference list: 30 [1] · John G. Proakis: Advances in Equalization for

Intersymbol Interference, Advances in Communication Systems, ss. 124-144, Academic Press Inc., New York, San Francisco, London 1975.

[2]. G.Duvic, J.Palicot, J.Veillard, M.Veillards 35 Operational Implementation of a D2-HDMAC / Packet Chain, 12 92356

International Broadcasting Convention, Brighton September 1990, Conference Publication Number 327.

[3.] R.J.J. Bartelink: Noise, reflection and group delay analysis of an HDMAC signal distributed by the Dutch 5 cable television network. Publication PHNL-AMV-91-8, Philips Nederland B.V., July 1991.

li

Claims (11)

A method for determining the tapping coefficients (Cj) of a transverse equalizer in a data transmission system, comprising a transmission channel through which a data signal is transmitted, in which a predetermined instruction signal (sk) is transmitted through the transmission channel, the equalizer's image equalizer is transmitted (yk) by summing received signal samples (rk + M1... rk_M2) from the equalizer's intermediate socket, of which samples are each weighted with their own tapping coefficient (C_M1... C + M2), and - as the difference between said sum signal sample (yk) and the instruction signal sample (sk), an error-representing error signal (Fk) is generated, characterized in that at least for a portion of the tapping coefficients (Cj) a new value (Cj (k + 1) is determined) )) by correcting the old value (Cj (k)) using a weighted combination of a first correlation (rk_j £ k) of received signal sample (rk_j) corresponding to the tapping coefficient in question (Cj) between lanterns and error signal samples (£ k) corresponding to the desired sample and a second correlation (so called k) of instruction signals (sk_j) corresponding to the coefficient of the relevant coefficient and error signal samples (£ k) corresponding to the desired sample. 2. A method according to claim 1, characterized in that, for each tapping coefficient (Cj) in the equalizer, a new value is formed in the manner described above. 30 3. A method according to claim 1, characterized in that the weighting coefficient (Gj) in said second correlation is changed starting from which tap coefficient (Cj) is determined. Method according to claim 1, characterized in that for the first part tap coefficient (Cj) a new value (Cj (k + 1)) is determined by correcting the old value (Cj (k)) with a weighted combination of a first correlation (rk_j £ k) of received signal samples (rk.j) corresponding to the intermediate tap of the relevant coefficient (Cj) and error signal sample (£ k) corresponding to the desired sample, and a second correlation (sk_j £ k) of instruction signals (sk_j) corresponding to the intermediate socket of the coefficient in question and error signal sample (kk) corresponding to the desired sample, and that the second part of the tapping coefficients (Cj) is controlled by a square error control known per se. 5. A method according to claim 4, characterized in that said first part consists of the tapping coefficients (Cj) whose respective signal samples are located furthest from the desired signal sample. The method according to claim 4, characterized in that said first part consists of the tapping coefficients (Cj) whose respective signal samples are delayed relative to the currently desired signal sample. 7. A method according to claim 1, characterized in that the weighting coefficient (Gj, Wj, aj) in the second correlation is increased where a corresponding sample is preferably further away from the currently desired signal sample. ? A method for determining the tapping coefficients (Cj) of a transverse equalizer in a data transmission system comprising a transmission channel through which a data signal is transmitted, in which method 30 - a predetermined instruction signal (sk) is transmitted through the transmission channel, - in the equalizer sum signal sample (yk) by summing received signal samples (rk + M1 · · · rk_M2) from the equalizer terminal, of which samples are each weighted with their own tapping coefficient (C_M1. C + M2), and t 92356 19 - an error representing error signal (κ k) is generated as the difference between said sum signal sample (γk) and the instruction signal sample (κ), characterized in that at least for a part of the tapping coefficients (C j) a new value (C j (k + 1) is determined) ) by correcting the old value (Cj (k)) using a weighted combination of a first product (rk.j £ k) of received signal samples (rk_j) corresponding to the tapping coefficient in question the intercept of the client (Cj) and received error signal sample (£ k) corresponding to the desired sample, and a second product (sk_j £ k) of instruction signals (sk_j) corresponding to the coefficient of the patient's intermediate socket and error signal sample (£ k) corresponding to the desired sample. . 9. A method according to claim 8, characterized in that the weighting coefficient (aj) of said second product is changed starting from which tapping coefficient (Cj) is determined. 10. Transverse equalizer for use in a data transmission system comprising a transmission channel through which a data signal is transmitted and in which system transmits through the transmission channel a predetermined instruction signal (so-called) which comprises a transverse equalizer - a plurality of consecutive delay elements (11) for providing intermediate terminals (12) connecting to the delay elements relatively delayed signal samples (rr + M1. .rk_M2), - means for weighing each signal sample with its own tap coefficient ( C_M1 ... C + M2), means for summing the weighted signal samples to form a sum signal sample (yk), and - means (13) for generating an error representing error signal (£ k) as the difference between said sum signal sample (γk) ) and the instruction signal sample (so-called), characterized in that at least part of the weighing means comprises means (22, 24 - 27) for calculating a weighted combination. n of a first correlation (rk.j £ k) of received 9 c 5 5 6 20 signal sample (rk_j) corresponding to the intermediate tap coefficient of the relevant coefficient (Cj) and error signal sample (£ k) corresponding to the desired sample, and a second correlation (sk_j £ k) of instruction signals (sk_j) corresponding to the intermediate output of the relevant coefficient and error signal sample (£ k) corresponding to the desired sample, and means (28, 29) to correct the value of the relevant tap coefficient (Cj) by means of said weighted combination. A transverse equalizer for use in a data transmission system comprising a transmission channel through which a data signal is transmitted and in which system transmits through the transmission channel a predetermined instruction signal (so-called), which includes a transverse equalizer - a plurality of sequential delay sequences Means (11) to provide, in intermediate terminals connecting to the delay elements, relatively delayed signal samples (rr + M1.. .Rk.M2), - means for weighing each signal sample with its own tapping coefficient (C_M1.). C + M2), means for summing the weighted signal samples to form a sum signal sample (yk), and - means (13) for generating an error representing error signal such as the difference between said sum signal sample (yk) and the instruction signal sample (so-called), characteristics 25. that at least some of the weighing means comprise means (31 - 34) for calculating a weighted combination of a first product (rk.j £ k) of received signal samples (rk_j) corresponding to the intermediate tap coefficient (Cj) intermediate socket and error signal sample (6k) corresponding to the desired sample and a second product (sk_j £ k) of instruction signals (sk_j) corresponding to the present the intermediate socket of the coefficient and error signal sample (k) corresponding to the desired sample, and means (36) for correcting the value of the respective tapping coefficient by means of said weighted combination. Il