JPH036688B2 - - Google Patents

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention relates to a decision feedback equalizer, and more particularly to a decision feedback equalizer that determines a tap gain coefficient by training with an isolated pulse response. .

[Prior art]

In data communications and the like, decision feedback equalizers are used to automatically remove distortion from signal waveforms distorted due to transmission failures, transmission path conditions, etc., and faithfully reproduce transmitted signal waveforms.

Here, when the transmitted signal sequence is a ₁ , a ₂ , ..., a _k ..., the received signal sequence is b ₁ , b ₂ , ..., b _k , ..., then b _k is transmitted The signal a _k has distortion superimposed thereon, and is expressed as b _k =a _k +N _k (k=1, 2, . . . ). N _k is the sum of subsequent distortions (postcursers) at time k of each of the signal sequences a ₁ , a ₂ , . . . , a _k-1 transmitted before the signal a _k . Now, for ease of explanation, it is assumed that this subsequent distortion affects only one subsequent signal. That is, the value of subsequent strain N _k at time k of a _k-1 is determined by a _k-1 , and the relationship is defined as N _k =A·a _k-1 . Therefore, since b _k = a _k + A・a _k-1 , the received signal
To extract the true transmitted signal a _k from b _k , N _k (=A・
a _k-1 ) can be removed.

In a decision feedback equalizer, the signal necessary for equalization, ie, a _k-1 , is derived by tapping, and this a _k-1
It is configured to obtain a gain coefficient A in order to calculate a distortion component A·a _k-1 . That is, by generating A・a _k-1 based on a _k-1 , a _k =
The original transmission signal is obtained by the subtraction formula b _k -A·a _k-1 .

In such a decision feedback equalizer, it is required to set the gain coefficient A quickly and accurately, but the conventional method for setting this is to successively modify the gain coefficient using the Zero-Forcing method. ing. Specifically, the input isolated pulse response is x
(t), x(0) is an isolated pulse, and its subsequent distortions are x(T ₁ ), x(T ₂ ), x(T ₃ ), etc., and these subsequent distortions are equalized. Decisinon for
According to the directed algorithm method, the error at the judgment time Ti is e(Ti) = X(Ti) - D(Ti) {however, X(Ti)
is the equalized output and D(Ti) is the judgment output}, the direction in which the error e(Ti) decreases at this time is determined, and the tap gain coefficient is successively corrected. this
In the Zero-Forcing method, the tap gain coefficient is successively corrected and converged using the algorithm shown below.

Ci ^(j+1) = Ci ^(j) −△・D(t−Ti)・S _go {X(t
)}...(1) However, equation (1) is approximated and simplified to simplify the circuit configuration and shorten the convergence time. Here, C _i
^(j) is the i-th tap gain coefficient corrected by the j-th training, △ is the correction step, 1, X(t)>0 Sgn{X(t)}={0, X(t)=0 −1 , X(t)<0. As can be seen from equation (1), the gain coefficient is successively corrected in increments of △ from the initial value C ⁽⁰⁾ for each training, and converges to the appropriate value A. That is, A≒lim j→∞C ^(j) . Correct equalization is performed only when C ^(j) converges to A, but if △ is a large value, it converges quickly, but
The difference between the convergence value C ⁽ ∝ ⁾ and the true A becomes large, and conversely, if △ is made small, the error between C ^(∞) and A becomes small, but there is a drawback that it takes time for convergence.

FIG. 1 shows a block diagram of a conventional decision feedback equalizer, and is an example of the case where the algorithm of equation (1) is used. The procedure for determining tap gain coefficients in initial training will be explained.

delay means 31, 32, in which the isolated pulse response x(t) of the first training is connected to the determination means 20;
. . . are input to 3N, and are propagated sequentially with a delay. The output of each delay means is multiplied by means 51 and 5.
2, ..., 5N, the initial setting value of the gain coefficient means 41, 42, ..., 4N of each tap
They are sequentially multiplied by C ₁ ⁽⁰⁾ , C ₂ ⁽⁰⁾ , . . . , C _N ⁽⁰⁾ , and added by the adding means 60. In the signal synthesizing means 10, the addition output of the adding means 60 and the input signal x
(t) are combined to obtain an equalized output X(t).

The tap gain coefficient correction means 70 uses the equalization output X(t) and the output of each delay means to correct the tap gain coefficient means 41 and
2,...,4N initial setting values C ₁ ⁽⁰⁾ , C ₂ ⁽⁰⁾ ,...,
C _N ⁽⁰⁾ is revised and updated to C ₁ ⁽¹⁾ , C ₂ ⁽¹⁾ , ..., C _N ⁽¹⁾ .
Similarly, the correction and updating of the tap gain coefficient is repeated multiple times until the tap gain coefficient converges. In the above, each delay means 31, 32,...
..., 3N _{, the outputs of each delay means are D(t-d1 ) ,} D( _t
-(d ₁ + d ₂ )),..., D(t-(d ₁ + d ₂ +...+
d _N )). Here, if Ti = _i 〓 ^k=1 d _k , the outputs are D(t-T ₁ ), D(t-T ₂ ), ..., D(t-
T _N ) represented.

As described above, the conventional method has the disadvantage that the gain coefficients of all taps are successively modified in a single training session, and that convergence of the coefficients requires long-term training even if the algorithm in equation (1) is used. Ta.

[Purpose of the invention]

The present invention has been made in order to eliminate the above-mentioned drawbacks of the conventional tap gain coefficient, and provides a decision feedback system that shortens the convergence time of the tap gain coefficient and makes it possible to set the gain coefficient of a single tap by training a single isolated pulse response. The purpose is to provide a shape equalizer.

[Summary of the invention]

The present invention has a plurality of taps each generating and deriving a plurality of signals necessary for equalizing the distortion of the received input signal from the received input signal, and multiplies the signals of each of these taps by a predetermined tap gain coefficient to obtain the result of these multiplications. A decision feedback equalizer that equalizes a received input signal by using a training to determine the gain coefficient of a single tap from a single isolated pulse using It is characterized by the following.

〔Example〕

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 2 is a block diagram of an embodiment of the present invention,
Parts equivalent to those in FIG. 1 are designated by the same reference numerals. In this example, we input the isolated pulse response x(t) for training and input the isolated pulse response x
(t) is supplied to the sampling means 80. In this sampling means 80, the time Ti from the isolated pulse x(0) of the isolated pulse response x(t)
Only the subsequent distortion x(Ti) is sampled, and each set value C _i of the tap gain coefficient means 41 to 4N is determined based on this sampled output.
The other configurations are the same as those shown in FIG. 1, and their explanation will be omitted.

In the decision feedback equalizer, the equalized output X(t) is expressed by the following equation.

X(t)=x(t)− _N 〓 ⁱ⁼¹ C _i・D(t−Ti) ……(2) Then, the subsequent distortion x(T ₁ ) of the isolated pulse x(0),
An isolated pulse response x containing x(T ₂ ), ..., x(T _N )
(t), tap gain coefficient means 4
1, 42, ..., 4N setting values C ₁ , C ₂ , ...,
This is possible by setting C _N to the same value as the subsequent distortions x(T ₁ ), x(T ₂ ), . . . , x(T _N ).

To explain this principle in a simplified manner using FIG. 3, if the subsequent distortion of the isolated pulse x(0) has a waveform as shown by diagonal lines, then the subsequent distortions A ₁ , A ₂ , A ₃ . . .
..., the subsequent distortion at the time of interest of the tap can be sampled from the isolated pulse response as the subsequent distortion value normalized by the isolated pulse amplitude, and this can be set as the gain coefficient as it is. Therefore, in the present invention, in order to set the tap gain coefficients C ₁ , C ₂ , _. conduct.

Isolated pulse response x of the j-th training input at time intervals such that the subsequent distortion is sufficiently small
(t) is expressed as xj(t), only the subsequent distortion xj(Ti) at time Ti from the isolated pulse xj(0) is sampled, and the set value C _i of the tap gain coefficient means is determined.

Setting the gain coefficients for all N taps is completed by performing training similar to the above multiple times. If the number of trainings required to set the gain coefficients for all taps is J, then a function J≧N holds true. However, {i} and {j} are independent numerical sequences, and {i} includes the element number of the tap,
{j} contains the training number required to set the gain coefficient of the tap. Here, by taking the input of the sampling means 80 from the output of the signal synthesizing means 10 and keeping the output of the adding means 60 open during the training period, X(t)=x(t)
Therefore, it is clear that tap gain coefficient setting similar to that described above is possible.

Through the above series of operations, each tap gain coefficient ci
After the setting is completed, it operates as a decision feedback equalizer according to equation (2), and the equalized output becomes the original transmission signal. The equalization operation itself at this time is the same as the conventional one. That is, the judgment value D(t) of the equalized output X(t) of the input pulse x(t) is input to the delay means 31, 32, ..., 3N connected to the judgment means 20, and is delayed. It propagates sequentially. The outputs of each delay means D(t-T ₁ ), D(t-T ₂ ), ..., D(t-
T _N ) is calculated by multiplying means 51, 52, . . . , 5N by gain coefficient means 41, 42, .
. . , _4N are sequentially multiplied by set values C ₁ , C ₂ , . In the signal synthesizing means 10, the addition output of the adding means 60 and the input pulse are synthesized to obtain an equalized output. A series of these operations of the decision feedback equalizer can compensate for continuous distortion and obtain a normal decision value.

〔Effect of the invention〕

According to the present invention, the gain coefficient of a single tap is determined by sampling a single trailing distortion from a single isolated pulse response in training, and a simple algorithm and a short training time are used to determine the gain coefficient of all taps. This has the advantage that the gain coefficient can be determined.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a conventional decision feedback equalizer, FIG. 2 is a block diagram of an embodiment of the present invention, and FIG.
The figure is a diagram illustrating the operating principle of the apparatus shown in FIG. 2. Explanation of symbols of main parts, 10... Signal synthesis means, 20... Judgment means, 31-3N... Delay means, 41-4N... Gain coefficient means, 51-5N
. . . Multiplication means, 60 . . . Addition means, 80 . . . Sampling means.

Claims (1)

[Claims] 1. It has a plurality of taps each generating and deriving a plurality of signals necessary for equalizing the distortion of the received input signal from the received input signal, and multiplies the signals of each of these taps by a predetermined tap gain coefficient and uses these multiplication results. A decision feedback equalizer that equalizes the received input signal is configured to sample the subsequent distortion at the time of interest due to the isolated pulse when it is input, and uses this sampled output to Training is performed to determine the gain coefficient of a single tap from a single isolated pulse, and this training is repeated at least a number of times equal to the number of taps possessed by the decision feedback equalizer, thereby setting the total tap gain coefficient. A decision feedback equalizer characterized by: