JPH03284012A - Decision feedback type equalizer - Google Patents

Abstract

PURPOSE:To perform powerful adaptive equalization by monitoring the impulse response of a line by shifting the center tap of a forward equalizer in a decision feedback equalizer, and performing the switching control of tap correction algorithm corresponding to the above monitoring. CONSTITUTION:A subtractor 33 takes difference between the output of a c+1 tap multiplier 18 in the forward equalizer 10 and that of a d1 tap multiplier 23 in a backward equalizer 20, and a subtractor 34 outputs the difference between the above difference and a decision signal as an error signal epsiloni+1. Tap correction from the center tap c0 to the c+1 tap of the equalizer 10 by LMS algorithm is performed by switching the error signal epsilon1 to the signal epsiloni+1 corresponding to the increment of the front edge h1 of the impulse response to main response ho. Also, a tap coefficient control circuit 71 is provided which performs the tap correction by switching the signal to the signal epsilon1 corresponding to the decrement of the front edge h-1 for the main response, and furthermore, changes a tap coefficient by multiplying by a coefficient less than 1 sequentially, and also, returns the tap correction to the one by the LMS algorithm performed by using the signal epsilon1 when the front edge h-l goes to a constant value or zero.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a decision feedback equalizer, and more particularly to a decision feedback equalizer that removes waveform distortion caused by multipath fading propagation.

(Prior Art) Conventionally, adaptive vases have been used for the purpose of removing waveform distortion caused by multipath fading lines and the like. This adaptive equalizer is classified into a linear equalizer (1) and a nonlinear equalizer. FIG. 4 is a block diagram of the linear equalizer (LE), and FIG. 5 is a block diagram of the nonlinear equalizer. These two isoflowers are both composed of transversal filters, and the tap coefficients of the transversal filters are set so that the root mean square or m-th root mean of absolute values of the error signal ε between the input and output of the judger is minimized. controlled. Linear equalizers are mainly used in terrestrial digital microwave communications, and linear equalizers multiply the received signal, which includes code amount interference and noise distributed on each tap of a transversal filter, by a tap coefficient. Since the code amount interference is removed by linear combination, the residual code amount interference of the equalizer becomes large for strong multipath distortion. In particular, linear equalizers can sufficiently reduce the received signal in response to severe selective fading, which increases the multipath delay dispersion (Delay 5 pread) relative to the transmission symbol length as the transmission speed increases or the propagation length increases. cannot be equalized.

As shown in Fig. 5, the nonlinear equalizer has a linear forward equalizer (FE) 10 and a nonlinear backward equalizer (BE) 10 for input.
It is a decision feedback equalizer (D F E) consisting of 20,
The center tap of the front equalizer 10 is normally the front equalizer 10.
is set to the final tap. The code amount interference due to the leading edge (Percursor) of the impulse response is removed by the forward equalizer 10, and the code amount interference due to the trailing edge (Pos tcursor) is removed by the backward equalizer 20. Judgment device 4
Since the decision signal output from 0 does not include code amount interference or noise, the equalization ability of the backward equalizer 20 based on decision feedback is greater than that of a linear equalizer, and the impulse response Pos tcu
rsor completely removes code amount interference, that is, delayed multipath distortion with respect to the main character within the tap range of the backward equalizer 20, so if fading is due to delayed multipath (minimum phase shift fading) Therefore, the decision feedback equalizer has better equalization ability than the linear equalizer. On the other hand, code amount interference due to the Precursor, that is, progressive multipath distortion, is equalized not by the backward equalizer 20 but by the forward equalizer 10, which is equivalent to a linear equalizer.
With respect to fading due to progressive multipaths (non-minimum phase shift fading), a decision feedback equalizer exhibits only the same equalization ability as a linear equalizer. Therefore, in terrestrial digital microwave communications, where strong progressive multipath conditions may occur, linear equalizers, which are easy to implement, are mainly used, and decision feedback equalizers are not often used. .

As a method for improving the equalization ability of a decision feedback equalizer against progressive multipaths, an adaptive matched filter (
The MF/DFF reception system with MF) 610 was published by the Institute of Electronics and Communication Engineers, Communication Systems Study Group, “Adaptive reception system in multipath transmission paths” by Takatsugu Watanabe, February 1979 (CS78-
203). This MP/DFE reception method performs decision-feedback equalization against progressive multipaths by maximizing the S/N ratio with an adaptive matched filter and performing optimal reception by equalizing distortion with a decision-feedback equalizer 620. 620 can be improved.

In FIG. 7, 701 is the waveform of the impulse response of the line before the adaptive matched filter 610 is output, and 702 is the waveform of the impulse response after the output of the adaptive matched filter 610.

Since the adaptive matched filter 610 symmetrizes the impulse response, a strong Precursor in the section t<Q of 701
The components are focused on the main response, as shown at 702, and some power is also distributed to the 1>0 Pos tcursor component. In other words, part of the leading distortion is converted into lagging distortion, which reduces the load on the forward equalizer, and conversely, the increased Pos tcursor is equalized by the decision feedback backward equalizer. Decision feedback equalizer 62
The equalization ability for zero progressive multipath is improved.

For such progressive multipaths, a decision feedback equalizer provided with an adaptive matched filter at the front stage exhibits better equalization ability than a decision feedback equalizer alone. However, a decision feedback equalizer with an adaptive matched filter installed at the front stage has an MP
/The first priority is to maximize the SN ratio by using diversity reception using an adaptive matched filter for multipath fading lines where the SN ratio is limited, such as non-line-of-sight communications where the DFE reception method has already been put into practical use. It is not necessarily optimal for a wired line with a relatively high S/N ratio and where only waveform distortion is a problem. Its equalization ability is, for progressive multipath distortion,
A decision feedback equalizer with an adaptive matched filter installed at the front stage is much better than a decision feedback equalizer alone, but a decision feedback equalizer alone is much better at dealing with lagging multipath distortion. is better. This is caused by new waveform distortion caused by the adaptive matched filter, and in the case of multilevel QAM transmission, this distortion becomes impossible to ignore as the multilevel level increases. The waveform distortion caused by the adaptive matched filter is due to the convolution of the adaptive matched filter and the delayed-dispersed impulse response, and the impulse response after passing through the adaptive matched filter is focused on the main response at t=0 as shown in Figure 7. However, this is due to the response spreading even though the level is low, and to equalize this it is necessary to considerably increase the number of taps of the decision feedback equalizer, which is not realistic.

(Problem B to be solved by the invention) The conventional decision feedback equalizer described above lacks equalization ability to realize high-speed transmission using multilevel QAM under strong multipath fading conditions. There was a drawback.

SUMMARY OF THE INVENTION An object of the present invention is to provide a decision feedback equalizer with excellent equalization ability.

(Means for Solving the Problems) The decision feedback equalizer of the present invention is a transversal filter with a symbol length of T intervals, and the center tap C0 of the transversal filter is shifted by N taps from the final stage side to the previous stage side. a forward equalizer that performs linear equalization on the input signal and a backward equalizer that is a transversal filter with a symbol length T and performs nonlinear equalization on the input signal; a first subtracter that takes the difference between the outputs of the forward equalizer and the backward equalizer; and a determination that inputs the output signal of the first subtracter and outputs a determination signal to the backward equalizer. a second subtractor that obtains an error signal ε1 by taking the difference between the input and output of the determiner; a first tap coefficient corrector that calculates and corrects each tap coefficient of the forward equalizer and the backward equalizer from the signal on each tap using an LMS algorithm; A linear filter for filtering, a delay device for giving a certain delay to the received signal and outputting it,
A third subtracter outputs the difference between the outputs of the delay device and the linear filter as a difference signal, and a main response h0 and a leading edge (P
recursor) h -1+ ...+h-4+ ・
...+h−N, and each C*I+ of the forward equalizer after the center tap C0.
C*2. ..., C, i, ... c, H taps and each d lll+ after the first stage tap of the backward equalizer
l? +..., d5,..., d, a fourth step that takes the difference between the C4, tap multiplier output of the forward equalizer and the d, tap multiplier output of the backward equalizer for d, 4 taps. a subtracter, a fifth subtracter that outputs the difference between the difference obtained by the fourth subtracter and the determination signal as an error signal εi+1, and a leading edge h- of the impulse response monitored by the impulse response estimator. , in accordance with the increase in the main response h0 of the forward equalizer, tap correction is performed by switching from the error signal ε1 to the error signal εi+1 using the LMS algorithm for the C41 tap after i taps from the center tap C0 of the forward equalizer, and the impulse The tap correction controlled by the error signal 81+1 is performed by switching to the first error signal 81 in response to a decrease in the leading edge h- of the response with respect to the main response, and further a coefficient smaller than 1 is successively added to the tap coefficient. a second tap coefficient control unit that changes the values collectively and returns to the tap correction based on the LMS algorithm that performs tap correction using the error signal ε1 when the leading edge h-, of the impulse response becomes steady or zero; It is characterized by comprising:

(Example) Next, the present invention will be explained with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of the present invention. In FIG. 1, 10 is a forward equalizer, 11, 12°, 13.14 is a delay element having a delay time of transmission symbol length T, 15, 1
6.17.1.8 is a multiplier, 19 is a combiner, 30.31
, 32.33.34 is a subtracter, 40 is a determiner, 20 is a backward equalizer, 2122 is a delay element having a delay time of transmission symbol length T, 23.24 is a multiplier, 25 is a combiner, 7
1 is a tap coefficient control circuit ■, 72 is a tap coefficient control circuit ■, 50 is a linear filter, 51.52.53°5, 55
is a delay element having a delay time of transmission symbol length T, 56
．． 57, 58, 59.60 are multipliers, 61 is a combiner, 7
7 is a delay device, and 75 is an impulse response estimator.

Here, the transmission symbol sequence is ai (n=-■...-1-
oo), and the discrete value of the impulse response of the transmission system until it is input to the forward equalizer 10 is -, then the discrete value rn of the received signal is expressed as . In FIG. 1, rn is input to a forward equalizer 10, and delay elements 11. ．． 12.13.14 and on each tap u-1, u-1, u6, r-g as u+l,
r--, r (1+ r Distributed in the order of 41, multipliers 1, 5., .16. 1.7. 18 multiply the tap coefficients 01, C-1, CO+ c- at each tap. , are combined by the combiner 19 and then output to the subtracter 30.

The determiner 40, which is a nonlinear element, generates a signal y obtained by subtracting the difference between the output of the forward equalizer 10 and the output of the backward equalizer 20 using the subtracter 30. is input, and a judgment signal (identification signal) '1° for yo is output. On the other hand, the judgment signal j0 is input to the backward equalizer 20, passes through the delay elements 21 and 22, and the judgment signal ↑1° on each tap is the tap coefficient d−d,
is multiplied by multipliers 23 and 24 to obtain 2. After being combined by the combiner 25, it is fed back to the subtracter 3° as the output of the backward equalizer 20. Therefore, the human power of the judge is y.

is expressed by the following equation.

Further, the subtracter 31 takes the difference between the input and output of the determiner 40, and obtains a first error signal εl = y o - Miya. <3)
5. In order for the decision feedback water equalizer to work as an equalizer, the root mean square value ξ1=E[C1'' of the error signal ε1 of the decider 40, similar to the linear equalizer (Wiener filter), must be set.
) (E indicates an expected value) It is necessary to set the tap coefficients of the forward equalizer 1o and the backward equalizer 20 so that (4) is minimized. Therefore, the tap coefficient can be more easily determined, and the ideal value of the tap coefficient can be obtained by solving the following 1-7j equation.

Here, * indicates a complex conjugate, and the tap coefficients are expressed as the following tap coefficient vector 7.

In addition, (5) the 6×6 matrix including the small matrix A on the left side is a correlation matrix for the entire decision feedback equalizer, and is also as follows. The above A is composed of correlation coefficients ajj, and its elements are expressed by the following equation based on an impulse response.

n=-D.

On the other hand, in FIG. 1, the tap coefficient control circuit 71 (■) is a tap-on signal u-2 of the front vase 1o excluding the c-tap.
, u-1+ uO and the tap signal on the rear vase 2o”
j −1+↑−2 and the error signal ε1, the normal LMS (least raean 5qua
re ) algorithm, i.e. dt = at + v at an-1” (1=11
2) Tap coefficients C-2+ c- of the forward equalizer lO by (8).

co and the tap coefficients dd of the backward equalizer 20.

is calculated sequentially for each symbol. Here, tap correction coefficient μ
By setting and ν within the convergence range, the tap coefficients can be obtained without solving the normal equation (51).By the way, the operation up to the above is the same as a normal decision feedback equalizer, but the In the embodiment shown in FIG. 1, the center tap c0 of the forward equalizer 10 is shifted by one tap from the final stage of the forward equalizer 10. This is an example of application to wave communication. Multipath propagation in normal terrestrial digital microwave communication is mainly approximated by a two-wave model in which there is one multipath that leads or lags the main wave. It is known that the delay time difference with the main multipath wave rarely exceeds the symbol length T.For such multipath fading lines, the forward equalizer 10
By simply shifting the center tap C, from the final tap position of the conventional forward equalizer 10 to the input side of the forward equalizer 10 by one tap, conventional decision feedback type equalization can be applied to progressive multipath distortion. It is derived from the normal equation that the equalization ability is greater than that of the This will be explained using FIG. 2.

In FIG. 2, 200 is a transmission symbol sequence, 201 is 2
Impulse response of multipath line shown in wave model,
210 is an embodiment of the decision feedback equalizer of the present invention; 221
are each tap of the forward equalizer 10 shown as a shift register, 212 is a multiplier, 213 is a combiner, 214 is each tap of the backward equalizer shown as a shift register, 215 is a multiplier, and 216 is a combiner. vessel, 217, 218, 219, 220
is a subtracter, and 221 is a determiner. In Figure 2, 2
01, the impulse response is due to progressive multipath,
The leading edge of the impulse response at time t=-T (Precu
rsor) h - r exists at almost the same level as the main response h0 and in opposite phase, indicating an extremely severe multipath line. Expressing this in the frequency domain, an extremely deep wedge-shaped drop (day, 7b) occurs in the carrier frequency level of the modulated wave signal band. 200 transmission symbol sequences (...a, -, a.

a, 1, . . .) are incorporated with the impulse response of 201 and input manually to 210. When the output of the determiner 221 is 10, the received signal on the center tap c0 of the front vase 10 is r
6 = hoa o + h-, a, HtonaF),
Kono Nakari. a, is the desired signal component due to the main response h0, and % h -1a 41 is the Precursor h-,
This is code amount interference. 2nd for each tap other than c0
The received signals are distributed as shown in the figure. In the case of a normal decision feedback type equalizer, at 210 in FIG.
(final stage), the signals distributed on each tap of the forward equalizer 10 are multiplied by the tap coefficients, and then linearly combined in the combiner 213 so that code amount interference is removed. , the rear vase 20 does not operate. For example, 00 in Figure 2
Tap output is Co (ha a6 +h-tact)
becomes. The code amount interference component here is C6h-1a+1, and this a. In order to remove the interference caused by , the main response component glue distributed on the C-i tap one tap before Co. a+I is multiplied by C-+, and the combiner 213 combines the c0 tap and C-+ tap outputs to obtain a
, 1 is removed. However, the interference component h-18+1 due to the Precursor on the C-x tap also
-, is multiplied and synthesized by the synthesizer 213, so ao2
interference occurs. To remove this, an additional CIl tap is required. The interference component due to 8.3 is included on the C4 tap, and to remove this, an additional C-1 tap is required, and if this operation is repeated, an infinite number of taps will be required. In this way, the powerful P
When removing code amount interference due to a recursor with a finite number of taps, a conventional decision feedback equalizer uses a linear isoflower (L
B) Similarly, equalization becomes impossible.

By the way, in the decision feedback equalizer according to the present invention, by shifting the center tap by 1 tap, C and taps can be prepared, and on this tap, r-+=h□ a-1
+h-186 received signals are distributed. In addition, in FIG. 2, the main response and Precursor of 201 are
h,=1. When Oh-1=-0,99, the normal equation (5) for the decision feedback equalizer 210 is solved (and co l C-1t c-z and d2 become almost zero, and (-= 1.01.d+ = -1,01.Here, it is impossible for the center tap 00 to become zero in a normal flower vase, and the desired signal component existing on the center tap. In order for ao to be led to the determiner 221 as the main signal and determined as ↑, co must be not zero but have a necessary and sufficient magnitude. However, in the solution of the normal equation above, co Instead of becoming zero, CHl becomes -1,01, so
The output of the forward equalizer 10 is C+Ir-1""C+1 (1111a-, +h-+a
o) = 1.01 ・(1, Oa-+ 0199a
o) = 101a-, +a.

becomes. On the other hand, the output of the backward equalizer 20 is d, ↑−1=
1.016-+, and if the decision signal output from the decider 221 is not incorrect, ↑-1 can be approximated to a-+, so the output of the forward equalizer 10 and the output of the backward equalizer 20 By taking the difference between , the code amount interference due to a-+ is completely removed, and only ao is input to the determiner 221 and correctly determined. That is, for a strong Precursor, the desired signal component due to the main response on the center tap,
Precursor on C◆1 tap set by shifting the center tap instead of ao
By using the ao of the interference components h-,a, and ao as the main signal and making a decision in the decider 221, leading multipath distortion is equivalently converted to lagging distortion, and strong equalization is performed by decision feedback. This makes it possible to achieve extremely high equalization capability even for progressive multipath distortions of intensity that were conventionally impossible to equalize. MF strong against progressive multipath distortion
Comparing the /DFE reception method and this embodiment, it is found that since this embodiment does not use an adaptive matched filter (MF), there is no waveform distortion caused by the adaptive matched filter, and adaptive matching can be achieved with a relatively small number of taps. Equalization ability exceeding that of a decision feedback type equalizer provided with a filter at the front stage can be obtained. In addition, for delayed multipath distortion, in Fig. 2, the center tap c0 becomes the main signal root, so that its tap coefficient becomes the dominant level, and the operation is similar to that of the conventional decision feedback equalizer. and high equalization ability can be obtained.

By the way, so far we have mentioned that a decision feedback equalizer with a shifted center tap has higher equalization ability than the normal equation solution, but when adaptive equalization is used, only the normal LMS algorithm Next, we will explain why the tap coefficients do not converge. In FIG. 2, the forward isomorphism 10 is basically a linear isomorphism, and when the backward isomorphism 20 is not operating, the C41 tap attempts to eliminate the code amount interference due to a-l, which precedes ao by one symbol. When the backward equalizer 20 operates, d and taps are a-
attempts to remove code amount interference caused by +. That is, C+
I and dl match the objects to be equalized, and when viewed from center tap C0, it can be said that C4 and dl correspond in time. In a decision feedback equalizer with a shifted center tap, the eigenvalues of the correlation matrix for the temporally corresponding (overlapping) taps of the forward equalizer 10 may approach zero as follows. show. From the normal equation (5), d = H"c - (9) Φc = H,... OI Here, Φ is 4 for only the forward equalizer 10 with N taps.
It is a ×4 correlation matrix, and Φ=A−HH”... αυ.

Diagonal component ali of matrix A (i = -2゜-1,0,+
1) is an autocorrelation coefficient, which is a real number n nee ■ and takes the same value regardless of i, and its magnitude is determined by the main response h0 of the impulse response and the responses in its vicinity.
...depends on high-level components such as h-z, h-+, h+, hz ``''''.

For example, Equation 00 is as follows for the configuration of the decision feedback equalizer shown in FIG.

-e, J:, e, E Q Flash Flash + 10++ Pu, J: +1:, e J: Buyo, α Stone, α 4, α d, e +1: , α 10++ + 11% s % 11 % % S % +++ 10J:J:+1: Deep -1 yo, J:J yo, α +g , 繻Here, the forward equalizer 10 temporally corresponds to the d1 tap of the backward equalizer 20. The autocorrelation coefficient for c+ and tap is φ8. From the above formula, φ■=...h-, h
-;+h-, h-i+h2h2, h;+... If the main response h0 component is not included and the Precursor h- of 201 in Fig. 2 is large, φ
has a value, but when h-, is zero, that is, there is no leading multipath distortion, or in a lagging multipath state,
For cases where hl has a value, φ approaches zero. By the way, for the correlation matrix Φ, there exists an eigenvector Q and an eigenvalue matrix A, and ΦQ=ΔQ...
・ On the first floor, satisfy the following.

The eigenvalue matrix A is a Jacobi matrix that performs orthogonal transformation on the correlation matrix Φ and sets the non-diagonal components to O.
If the autocorrelation coefficient is zero,
The eigenvalue for it also becomes zero. As shown above,
φ1. When becomes close to zero, for that, that is, C1
, the eigenvalue λ1 for the taps also approaches zero.7 This phenomenon is not observed in normal decision feedback equalizers, and when the center tap of the front isometric vase 10 is shifted, the forward This occurs only for the equalizer 10 taps.

By the way, if the eigenvalue becomes small, the convergence of the corresponding tap coefficient becomes extremely poor, and if it becomes zero, it no longer converges.

Let cOpL be the solution of the normal equation for the αω-type forward equalizer 10, that is, the ideal tap coefficient vector, and the difference from the tap coefficient vector C due to adaptive control as the coefficient error vector V = c c OD t .
・ When introduced as (2), the evaluation function ξ1 of equation (4)
is transformed as follows.

C1-ξ, t+ + (c all
L ) T Φ ((c 019 t-1ξ-, +V
' ΦV = ξ-9+78 ... αR here ξ7,
7 is the minimum value of C1, and
It is a quadratic form of the order differential coefficient, and the eigenvalue determines the shape of the quadratic surface of ξ. By the solution of the normal equation, as long as there is an ideal value C,Opt for the i-th tap C1,
C1 indicates a change in a quadratic curve that minimizes 1=0 in the direction of one axis of coefficient error. However, as the eigenvalue λ in λ becomes smaller, the curve gradually flattens, and when λl becomes zero, C1 becomes a straight line in the axial direction.

If the tack coefficient is corrected using the LMS algorithm in this state, it will no longer converge and the tap will maintain its initial value.

A description will be given of changes in the tap coefficient when the impulse response of the two-wave model changes as shown in FIG. 3 for the decision feedback equalizer shown in FIG. 2.

In FIG. 3, 301 is a distortion-free fc in which only the main response h0 exists, and the tap coefficient for this is 304, which is found from a normal equation. Since there is no need to equalize 301, in 304, the value of only the center tap c0 of the front equal vase O is waited for.

302 is a Precursor of -t=-T for 301
h-.

is slightly increased, and the tap coefficient for it is 30.
It is 5. When the Precursor is smaller than the main response, the solution to the normal equation shows that the distortion is equalized by the forward equalizer lO, similar to a normal decision feedback equalizer, and they overlap each other. C++ and dl are zero. 303, with the Precursor further increased, the solution to the normal equation in this case is 306, co
Instead, use c- as the main signal route, and use Precursor
d of the backward equalizer 20, which is decision feedback, is in the increasing direction, that is, from 301-302 to 303.
For the impulse response change, the initial value of the tap coefficient is set to 304, and all the taps of the decision feedback equalizer are L.
The tap coefficient changes obtained by correcting the taps using the M and S algorithms are 307→308→309.

301-,'(Up to 02, the eigenvalue λ1 for CH1 tap is extremely small, and c+1 of 308 maintains the initial value O of 307. Therefore, the distortion due to 1, t, is linearly equalized by the C knob. 5. However, if t increases further and a strong progressive multipath state occurs as in 303, λ will have some value. However, C1, C
Eigenvalue λ for I. .

Since λ 1 is much larger, c(1, c-1 converges faster. As shown in 309, tap 00 is the signal component due to the main response. Continue to hold a as the main signal. As ,h-,increases, the,c-,tap grows further.

Further, during the equalization by C-, interference from the signal a4g which is 2 nm later than ao occurs, and C~2 attempts to equalize this. In this way, when the progressive multipath increases from a non-distorted state, a decision feedback equalizer with the center tap of the forward equalizer shifted is normally operated using the M S algorithm. It operates in the same way as the decision feedback equalizer, and does not converge to the ideal solution of the normal equation 306. Next, when the change in 303-302-301, that is, the progressive multipath is in the decreasing direction, the initial value of the tap coefficient is set to the ideal solution of 306, and the decision feedback equalizer is operated using the LMS algorithm. And the tap coefficient change is 3
12→311-310. Precursor h
When −+ decreases like 302, λ1 becomes smaller and c
The tap correction of +, cannot follow the fluctuation of the impulse response, and c+ of 311 maintains almost the value at 312. On the other hand, the C-+ tap with an eigenvalue larger than C,t is 30
According to the ideal solution of 5, we try to equalize the distortion caused by h-. Here, CB is relatively small compared to 305, but this is because the C4 tap remains large, and due to the C01 tap, the response component h-Ia due to h-
This is because signal a0 in o is added to the main signal route from center tap c0, and C0 does not require as much level as shown at 305. On the other hand, since C11 has a value, the overlapping dl eliminates the distortion that occurs. Furthermore, when the state becomes undistorted from 302 to 301,
Since there is no distortion due to h-, C-+ becomes zero, and the signal component supplied from C4 and tap -□IaO becomes zero, so C0 reaches the same normal level as 37. But C0
I and d still hold the state of 41.

The above is the case for progressive multipath fluctuations, but
Regarding lagging multipath fluctuations, the normal LMS algorithm converges satisfactorily. In this case, the rear etc. 20 operates while the tap coefficient on the terminal side of the front etc. 10 is kept zero from the center tamp of the forward etc. 10, so it can be said that it is completely equivalent to a normal decision feedback type equalizer .

Next, we will discuss the equalizer configuration and control algorithm to solve this convergence problem.

The reason why the change from 302 to 303 in FIG. This is because the decision continues to be made by the ao decider of
The main signal route changes to C4I tap, Precu
The transition to the ideal solution of 306 does not occur unless ao of h-Iao is determined by rsor. Incidentally, c+ and dI, which overlap with each other, are not independent, and d is dependent on C11. In particular, since the eigenvalue of the c+ tap is smaller than that of other taps, even if the C+I tap coefficient deviates from the ideal value, it does not affect the evaluation function ξ1 much. This means that the distortion caused by the deviation of C4, d
This is because one tap is equalized dependently. This is confirmed by the fact that although some residual distortion remains in the state 310 in FIG. 3, the distortion caused by C3 is equalized by d. Therefore, in order for the main signal to gradually shift from ao on the center tap C0 to C+ and ao on the tap as the progressive multipath component increases, the c+1 tap is normally set. It is necessary to introduce an algorithm that grows the tap, rather than the LMS algorithm. Furthermore, it must not significantly disturb the control system of the decision feedback equalizer. Here, we will focus on the phenomenon that the eigenvalue of c- is small and the tap coefficient of the ideal solution C increases as ri increases. C3
One tap correction is performed using the error signal ε1 of the determiner 221.
Instead of the usual LMS algorithm using C2, we introduce a second error signal ε2 such that the C31 tap decreases in the direction of growth as h- increases, and LMS algorithm using C2
Corrections are made using an algorithm.

As shown in FIG. 1, the subtracter 33 subtracts d and the tap output from c+ and the tap output, and the subtracter 34 subtracts the output of the determiner 40, resulting in a second error signal ε2. The tap coefficient control circuit 72 uses the received signal u- on the C2, C, and taps to calculate C,
Taps are sequentially corrected for each symbol.

Here μ is a correction coefficient.

As shown in FIG. 2, from the output C61 (h 6 a −1+ h −186) of the C9I tap multiplier 212
The tab output d (6-1 is subtracted by the subtracter 217, and then the output terminal of the determiner 221 is subtracted by the subtracter 218.
The difference from the output of 7 is an error signal ε2, and when the judgment signal is correct, the following equation is obtained.If C+1 is zero, d is also zero, so C2 in the above equation has the value -a. On the other hand, u-1=h, a-1
+h-1a6, and by averaging the second intersection on the right side of the formula (to), it becomes the correlation between the error signal ε2 and the input signal, and has the value E Cat u-+") = h-+. Therefore, (to) ) while modifying each symbol one by one,
Even if the initial value of c,i is zero, as h−, increases, C
01 begins to grow. When c+1 starts to have a value, c1
Interference occurs with respect to a-+ from the tap, and d of the rear equalization layer 20 equalizes this, and the first term in the second row of the αη equation approaches zero. If the value of c+1 is not sufficient, the second term in the second line of the αη formula has a value as an error signal for the C11 tap to become the main signal route, and while controlling the C0 tap with the 00 formula, It gradually grows so that tap C4 becomes the main signal route.

Formula 00 does not use the error signal of the determiner, so it cannot be called a normal LMS algorithm, but the root mean square value of C2 ξ 2 = E (82"
- Correct the 0.1 tap coefficient to minimize am.

(to) can be expressed as follows.

n=−■ (d, he +h−+)+ (ci+”+t)
...α9 Therefore, c+1 that minimizes C2 is C2
/c-=0n;-■, indicating that c+1 grows as h- increases. By the way, with C2 introduced here,
When the c+1 tap coefficient is corrected, the evaluation function ξ1 of equation (4) defined by C1 is dependent on the value of the C4° tap, and under the condition that the c0 tap is corrected by the ae2 algorithm, Error signal ε of the determiner 221
In order to minimize the number of evaluation intervals ξ1 for 1, C4, tap is divided (all taps are controlled by the LMS algorithm. Therefore, in this case, C◆rei, and in the normal equation (5), C, l tap When the term related to is deleted and C and the taps are corrected by 82, the normal equation for the taps excluding C and I has a 5 × 5 correlation matrix, and from ■ and equation (5), we get the following become.

here, Also, in Figure 2, 20 The impulse response of 1 is For example, h0=1. o h−+= Equation (21) becomes as follows.

0.9 If it becomes 9, ... (22) Furthermore, finding the eigenvalue of the correlation matrix on the left side of equation (22), we get
λ-,=2.39 λ-1=1.98 λo=1.
98λ, = 0.50 λg = 1.00, neither of which is zero. Therefore,
Each tap in equation (22) converges to a matrix solution in equation (22) by the LMS algorithm using C1. On the other hand, C41 is given by equation (20), so by substituting ct, =-1,0 in (23), C
or.

=-1, O. This result is expressed as ξ1/c in the decision feedback equalizer shown in FIG. This is almost equivalent to the solution of the normal equation (5) including 1=0. That is, even if the second error signal ε2 is introduced and 82 is used for 0.1 tap correction, there is a solution in which the error signal ε1 of the determiner 221 is minimized without being disturbed by C2, and that is only sl. It can be approximated to the solution of the normal equation (5) focusing on . By the way, what has been explained above is the ideal solution of equation (5), (20) with 82 introduced, and (
Although this is an example in which the solution from equation 21) almost matches, it is an approximate solution when Precursorh- is increasing, and does not necessarily match completely. In such a case, the solution from equations (20) and (21) almost coincides with the α-barrel algorithm using C2, but it is just a Precursor algorithm.
This is an approximate solution when h-, is increasing, and does not necessarily match completely. In such a case, once the algorithm (2) using C2 converges to the approximate solution given by equations (20) and (21), C4 and tap correction can be done as in equation (7). By switching to the LMS algorithm using 81, (5)
It is possible to converge to the ideal solution of the equation. This is because h-I is relatively large compared to ho, so when ct and tap are corrected by C1, the eigenvalue for 0.1 tap is smaller than other taps, but it is not zero and has a value. Therefore, once C2 is brought to a state close to the ideal solution, C4,
This is because the coefficient has grown sufficiently and the main signal route has shifted to the C+1 tap, so even if it is switched to C1, the ideal solution will converge.

However, switching the adaptive algorithm in this way will give discontinuity to the adaptive operation, but the correction coefficient for tap correction in -m is sufficiently small, and the speed of multipath fading fluctuation is relatively small compared to the transmission speed. By being slow enough, the adaptive behavior remains nearly continuous. By the way h
Even if -, is in the increasing direction, if h-, is smaller than a certain value compared to ho, the solution to the normal equation (5) is the normal decision feedback type where the main signal route is kept at 00. It is close to the tap coefficient of the converter. In such a case, tap correction by C2 is not required, so, for the switching point of the adaptive algorithm for this C,l, if,h-,increases and exceeds a certain value, 82 is used; After that, when it becomes steady, it switches to C1.

Next, when h- is decreasing, if tap modification by 62 of the C0 equation is used, the main signal route continues to be held at the C11 tap, and the ideal solution to equation (5) is not converged. In this case, C1 is used to correct the C41 tap, and in order to further reduce C1 and the tap coefficient, and gradually return the main signal route to the center tap c0, the following tap correction algorithm is introduced.

In equation (24), the normal LMS algorithm that uses 81 for tap correction of ct is multiplied by (1=μ), so 312-311 for the impulse response change from 303-302 to 301 in Fig. 3 At a tap coefficient change of -310, C4° does not hold a constant value, but gradually decreases. In this case, the value of CHl is (5
Although it does not necessarily match the ideal solution of the equation C4, the distortion due to any deviation from the ideal solution of C4 is removed by the backward isotherm 20, d, so that it does not affect the equalization operation of multipath distortion due to propagation. I won't give it. In addition, if h- is in the decreasing direction and reaches a steady state in the middle, then in equation (24), (1=
Similarly to the case where h-1 becomes stationary in the increasing direction without being multiplied by μ), tap correction is performed using the normal LMS algorithm according to 81 of equation (7). In the above switching of the correction algorithm for the c+1 tap, 82 moves the main signal route to the C+1 tap as a means to obtain an approximate solution, and as in equation (24), multiplying by (1-μ) is It is mainly used as a means to return the main signal route to the center hub Co. Regardless of whether h increases or decreases, the final result is tap correction by C1 when hl is steady. , converges to the ideal solution of the normal equation in (5). Furthermore, since the fading speed is sufficiently slow compared to the transmission speed, even if h- fluctuates over a long period, it remains in a state that can be considered steady over a short period, and by switching the tap correction algorithm described above, It is possible to converge a decision feedback equalizer with a shifted center tap in response to fluctuations in multipath. On the other hand, in the case where h- becomes zero and delayed multipath occurs, as mentioned above, the C4I tap correction can be sufficiently converged with the LSM algorithm using C1, so switching the tap correction algorithm is not necessary. do not need.

As explained above, the decision feedback equalizer of the present invention needs to monitor the fluctuation of Precursorh- with respect to the main response h0 of the impulse response in order to control the switching of the tap correction algorithm. As a means for this purpose, as shown in FIG. The determination signal outputted from 40 and the output tap coefficient W of impulse response estimator 75 are convolved. The convolution value is a reproduced waveform (r6plica) obtained by reproducing the received signal.

The received input signal delayed by the delay device 77 and the synthesizer 61
The difference between the two outputs is taken by the discharger 32 and becomes an error signal e of the reproduced waveform with respect to the received signal. Impulse response estimator 7
5 inputs the error signal e and the judgment signal output from the judgment unit 40, and sequentially calculates the estimated impulse response discrete values WI using the LMS algorithm shown below.

Here, β is a correction coefficient. In the above formula, the parameter n
indicates the time of each symbol, and the output of the determiner 40 is . At this time, the signal ↑A on the center taps of the delay elements 51, 52, 53°5, and 55 becomes ↑-3. That is, since a three symbol time delay occurs in the impulse response estimation process, the delay amount of the delay device 77 is set to 3T. From the above operations,
Control is performed so that the root mean square value of the error signal e is minimized, and an estimated value Wl of the impulse response is obtained. W6+W-t of the impulse response estimator output, that is, the main response h0 and Precu
rsor h-, it is possible to monitor changes in h- relative to ho. The tap coefficient control circuit a of 72 is Wo.

W-, the error signal ε1 of the determiner 40, the error signal ε2, and the received signal U on the C41 tap of the forward isotherm 10. 1 is input, and by Wo and W-, the tap correction algorithm is switched as shown below, and C4 and tap coefficients are sequentially calculated for each symbol.

(t) h-, when increasing (Ih-11>γ1hol) (2) h-, when increasing (l h-, <Tl ho 1
) or when h-, steady or zero (3) When h-1 decreases In (1) and (2) above, γ is Precurs
In a region where or h-+ is smaller than the main application h0 and the distortion due to h- is equalized by the forward equalizer as in a normal decision feedback equalizer, the main signal route is connected to the c and l taps. This is a control threshold to prevent transition.

Through the above operation, the tap coefficients can be converged for both leading and lagging multipath distortions, and equalization is performed using decision feedback for leading multipath distortions, as in the case of lagging ones. This makes it possible to powerfully equalize fading due to progressive multipaths (non-minimum phase shift fading), which is a weakness of conventional decision feedback equalizers.

(Effects of the Invention) As explained above, the present invention provides a decision feedback equalizer (
By shifting the center tap of the forward equalizer of D F E), monitoring the impulse response of the line, and switching and controlling the tap correction algorithm accordingly, fading due to progressive multipath (non-minimum phase shift fading) is eliminated. It is possible to perform powerful adaptive equalization using decision feedback, and excellent equalization performance can be obtained with a relatively small number of taps, making it more suitable for multi-level QAM transmission on severe multipath fading lines. This has the effect of making it possible to further increase the transmission speed and extend the length of the line section.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram explaining the operation of this embodiment, FIG. 3 is a diagram explaining adaptive equalization in this embodiment, and FIG. 4 is a diagram explaining the operation of this embodiment. Fig. 5 is a block diagram of a conventional linear equalizer, and Fig. 6 is a block diagram of a conventional decision feedback equalizer.
This figure is a block diagram of a conventional MF/DFE receiver, and FIG. 7 is a diagram showing the waveform of the impulse response of the line before inputting and after outputting the adaptive matched filter. 10... Forward equalizer, 11.12, 13.14°21
．． 51.52.5, 55...delay element, 15°16.
17.1B, 23.2, 56.57°5B, 59.60
... Multiplier, 19.25.61 ... Combiner, 20-
・Backward equalizer, 30.31.32°33.34...
Subtractor, 40... Determiner, 5o... Linear filter,
71...Tap coefficient control circuit I, 72...Tap coefficient control circuit ■, 75 River impulse response estimator, 77...
・Delay device.

Claims (1)

[Claims] Forward equalization is a transversal filter with a symbol length T interval, and the center tap c_0 of the transversal filter is shifted by N taps from the final stage side to the previous stage side, and performs linear equalization on the input signal. The vessel and
a backward equalizer that is a transversal filter with a symbol length T interval and performs nonlinear equalization on an input signal; and a first subtractor that takes the difference between the outputs of the forward equalizer and the backward equalizer. a determiner that inputs the output signal of the first subtracter and outputs a determination signal to the backward equalizer; and a second subtraction that takes the difference between the input and output of the determiner to obtain an error signal ε1. The vessel and
LMS is calculated from the error signal ε1 between the input and output of the determiner and the signals on each tap of the forward equalizer and the backward equalizer.
a first tap coefficient corrector that calculates and corrects each tap coefficient of the forward equalizer and the backward equalizer by an algorithm; a linear filter that linearly filters the judgment signal output from the judge; and a received signal. a delay device that applies a certain delay to and outputs it; a third subtractor that outputs the difference between the outputs of the delay device and the linear filter as a difference signal;
The main response h_0 and the leading edge (Precursor) h_- of the impulse response of the line are determined from the difference signal and the judgment signal.
an impulse response estimator that estimates _1, ..., h_-_i, ..., h_-_N; and each c_+_1, c_+_2, at a stage subsequent to the center tap c_0 of the forward equalizer.
..., c_+_i, ..., c_+_N taps and each d_1_y, d_2 after the first-stage tap of the backward equalizer
a fourth subtractor that takes the difference between the c_+_1 tap multiplier output of the forward equalizer and the d_1 tap multiplier output of the backward equalizer for _y, ..., d_i, ..., d_N taps; a fifth subtracter that outputs the difference between the difference obtained by the fourth subtracter and the determination signal as an error signal εi+1;
Tap correction is performed using an LMS algorithm for c_+_1 taps after i taps from the center tap c_0 of the forward equalizer in accordance with an increase in the leading edge h_−_1 of the impulse response monitored by the impulse response estimator relative to the main response h_0. from the error signal ε1 to the error signal ε
i+1, and the leading edge h of the impulse response
The error signal ε is increased in response to a decrease in the main response of _−_1.
The tap correction controlled by i+1 is applied to the error signal ε1.
The tap coefficient is changed by successively multiplying the tap coefficient by a coefficient smaller than 1, and the leading edge h of the impulse response is changed.
a second tap coefficient control unit that returns the tap correction to the LMS algorithm-based tap correction using the error signal ε1 when _-_1 becomes steady or zero; Maker.