KR20050073325A - Channel equalizer - Google Patents

Abstract

The present invention relates to a channel equalizer for estimating a channel in a data interval with a conjugate gradient algorithm and applying the channel estimate to a modified decision feedback equalizer used in an existing 8VSB receiver to perform equalization. Therefore, unlike the channel estimator using the least square method, Since the calculation is performed by an iterative operation rather than directly, the hardware complexity of the channel estimator is reduced. In addition, by performing channel equalization not only with the field synchronization interval but also with the channel estimates frequently made in the data interval, the performance of the equalizer in the dynamic channel can be improved.

Description

Channel equalizer

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a channel equalizer for compensating for distortion due to a channel. In particular, the present invention relates to performing an equalization by applying a channel estimate estimated by a conjugate gradient method to a modified decision feedback equalizer. Relates to a channel equalizer.

In general, symbols transmitted over a limited bandwidth multipath channel are distorted through intersymbol interference. Therefore, a bit error occurs in the receiver that receives the error. As a result, intersymbol interference has been recognized as a major obstacle in transmitting high-speed data in multipath channels.

In addition, a channel equalization process using an equalizer is performed in a receiver to minimize the intersymbol interference.

Typically, communication channels are not fixed but time-varying, so the equalizer must be able to track the time-varying characteristics of these channels, which is called an adaptive equalizer.

In addition, the most commonly used channel equalizer is a Decision Feedback Equalizer (DFE) based on the LMS algorithm.

The DFE considers the main energy path as the main path when the received signal enters through a multipath channel, and considers all other paths as adjacent signal interference (ISI) or ghost signals through the reflection path. Only the signal coming in through the main path is extracted by correcting phase and magnitude, and the signals coming in through the remaining path are removed.

FIG. 1 shows a block diagram of a general decision feedback equalizer operating in the time domain.

Referring to the operation of FIG. 1, the signals of the path arriving in time earlier than the main path through the feed forward filter 101, that is, the influence of the pre ghost are removed, and the rear filter or feedback filter is removed. The filter 102 eliminates the influence of the signals of the path arriving later than the main path, that is, the post ghost. At this time, the adder 105 adds the output of the front end filter 101 and the output of the feedback filter 102 and outputs the result to the decision device 103, wherein the decision unit 103 is the adder 105. The output signal of is compared with a preset eight-level reference signal, respectively, and the reference signal level closest to the output signal of the adder 105 is determined as a determination value. At this time, the output of the determination unit 103 is fed back to the feedback filter 102 and the control unit 104. In other words, the input of the feedback filter 102 is not an output of the adder 105 but a determination value that has passed through the determining unit 103.

On the other hand, the control unit 104 receives the output of the equalizer, that is, the output value of the adder 105 and the determination value, and updates the coefficients of the front end filter 101 and the feedback filter 102.

In the decision feedback equalizer, when the decision is made accurately by the decision unit 103, the noise amplification phenomenon does not occur because the noise contained in the equalizer output component is re-input to the input of the feedback filter 102. In general, the performance is superior to that of a linear equalizer. In addition, the above-described decision feedback equalizer has an advantage in that performance can be comparable to the maximum likelihood sequence estimator (MLSE), which is an optimum filter when the decision error is small.

On the other hand, the general operation mode of the adaptive equalizer is divided into training mode and tracking mode. First, the transmitter transmits a training string of a predetermined length known to the receiver by mutual appointment. The equalizer of the receiver then takes this training sequence and sets the tap coefficient to the appropriate value. In the current ATSC 8VSB system, a training sequence of 832 symbols consisting of bit strings such as PN511 and PN63 used for frame synchronization is used.

However, if the channel is poor, the equalizer coefficients do not converge within this training sequence. Therefore, another conventional adaptive equalizer proposed for fast initialization so that the tap coefficients of the equalizer can converge even in a poor channel environment is the modified decision feedback equalizer of FIG. 2.

That is, in the crystal feedback equalizer of FIG. 1, the output of the feedback filter 102 is fed back to the output of the shear filter 101, whereas in the modified crystal feedback equalizer of FIG. 2, the output values of the feedback filter 220 are the shear filter. Direct feedback to Tapped delay line (TDL) inputs within 210 is shown.

In FIG. 2, reference numeral 210 includes a TDL 211, a post-ISI remover 212, a coefficient updater 213, and an adder 214 as a front end filter. The TDL 211 of the front end filter 210 is composed of a plurality of serially connected delayers that sequentially delay input symbols. The post-ISI removal unit 212 is composed of a plurality of subtractors, each subtractor subtracting each weighted sum returned from the feedback filter 220 from the input / output signals of the respective delay units of the TDL 211. Remove ISI The coefficient updater 213 is composed of a plurality of multipliers, and each multiplier performs coefficient update by multiplying a previous coefficient value corresponding to the output of each subtractor of the post-ISI remover 212. The outputs of the multipliers of the coefficient updater 213 are output to the adder 214, are added together, and then output to the determiner 230. Since the determination unit 230 performs the same configuration and operation as the above-described determination unit of FIG. 1, detailed description thereof will be omitted.

In addition, reference numeral 220 denotes a TDL 221 and a weighted summation 222 as a feedback filter. The channel estimator 240 calculates a cross-correlation value between the training sequence constituting the field sink and the training sequence known in advance in the receiver, and then calculates the channel correlation value. As a result, it is output to the weighting adder 222 of the feedback filter 220. The weight adder 222 obtains a weighted sum as shown in Equation 1 below by using the channel estimate of the channel estimator 240 and the input / output signals of the respective delayers of the TDL 221. Output to each subtractor of the post-ISI removal unit 212 of the filter 210.

Here, j represents a burst time, q is from 0 to F-1, and x _nm is an input / output signal of each delay unit of the TDL 220.

Postcursors are obtained by subtracting each weighted sum calculated and output from the input / output signals of the respective delayers of the TDL 211 in the post ISI removal unit 212 of the front end filter 210 as shown in Equation 1 above. In other words, the post-ISI is eliminated.

The modified feedback feedback equalizer removes Post ISI due to past symbols, and then removes the remaining Pre-ISI components by adapting the shear filter 210.

Therefore, if the channel estimation is accurate at the beginning, only the pre-ISI portion of the front end filter 210 needs to be adaptive equalized, and thus the initialization convergence speed is faster than that of the decision feedback equalizer structure of FIG. 1.

However, if the channel is dynamic, the existing modified decision feedback equalizer is difficult to guarantee the performance. This is because the feedback filter coefficients of the modified decision feedback equalizer described above are not updated in the data section following the training sequence section, so it is difficult to equalize to remove post ghost.

Therefore, there is a need for a method that allows the modified feedback feedback equalizer in a dynamic channel to converge and estimate.

An object of the present invention is to solve the above-mentioned problems of the prior art.

Another object of the present invention is to perform a channel estimation in the data section by applying a conjugate gradient algorithm in the channel estimator and apply the channel estimate to the modified decision feedback equalizer to perform channel equalization in the data section. The present invention provides a channel equalizer for improving equalization performance in a dynamic channel.

A channel equalizer according to an embodiment of the present invention for achieving the above object, by applying a conjugate-gradient algorithm to the input decision value channel estimation for estimating the channel in the field sync interval and data interval government; And a shear filter, a feedback filter, and a determiner, wherein the weighted sums are obtained from the channel estimates estimated by the channel estimator and the determined values delayed by the feedback filter, respectively, and the input values are respectively delayed by the shear filter. And a channel equalizer which removes the interference due to past symbols by subtracting the respective weighted sums from each other, removes the interference due to future symbols by updating coefficients, and outputs a determination value determined by the determiner to the channel estimator. It is characterized by.

The front end filter of the channel equalizer is composed of a plurality of serially connected delayers, and includes a tap delay line (TDL) for sequentially receiving a received signal passing through the channel and a plurality of subtractors, each subtractor of the TDL. A post ISI removal unit for eliminating interference due to past symbols by subtracting the corresponding weighted sum fed back from the feedback filter from the output signal of each stage, and a plurality of multipliers, each multiplier being composed of each post-ISI removal unit A coefficient updater that removes interference due to future symbols by multiplying a previous coefficient value corresponding to the output of the subtractor by performing a coefficient update, and an adder that adds and outputs outputs of each multiplier of the coefficient updater. It is characterized by.

The feedback filter of the channel equalizer is composed of a plurality of series-connected delayers, and sequentially outputs a TDL for delaying the decision value of the determiner and a channel estimation value of the channel estimator in which the channel estimation is performed even in the data section and the output of each stage of the TDL. It is characterized in that it comprises a weighted summation (weighted summation) for receiving a signal to obtain a weighted sum and output to the shear filter.

A channel equalizer according to another embodiment of the present invention includes a channel estimator for estimating a channel in a field sync interval and a data interval by applying a conjugate gradient algorithm to an input MLSE decision feedback signal; A feedback filter that receives the 8-level decision value or the MLSE decision feedback signal and sequentially delays it, and obtains a weighted sum from each delayed signal and the channel estimate estimated by the channel estimator; Receives the received signal passing through the channel and delays it sequentially, removes the interference caused by the past symbols by subtracting the corresponding weighted sum returned from the feedback filter from each delayed signal, and updates the coefficients to remove the interference caused by future symbols. Shear filter to remove; The output of the front end filter is compared with the eight-level reference signal defined by the 8-VSB ATSC standard, respectively, and the reference signal closest to the output of the front end filter is determined as an eight-level decision value and outputs to the feedback filter. A level determining unit; The MLSE unit estimates a transmission signal by applying a maximum likelyhood sequence estimation (MLSE) method that uses correlations of past symbols with respect to an output signal of a front end filter, and outputs an estimated MLSE decision feedback signal to the feedback filter and the channel estimator. Characterized in that it comprises a.

The feedback filter includes a plurality of series-connected delayers, a TDL configured to selectively receive an 8-level decision value of the 8-level decision part or an MLSE decision feedback signal of the MLSE part and sequentially delay the channel, and the channel for which channel estimation is performed in the data section. And a weighted summation unit configured to obtain a weighted sum after receiving the channel estimate value of the estimator and the output signal of each stage of the TDL, and output the weighted sum to the front end filter.

The TDL of the feedback filter may be sequentially delayed after receiving MLSE decision feedback signals having different decoding depths at intervals of 12 symbols.

Other objects, features and advantages of the present invention will become apparent from the following detailed description of embodiments taken in conjunction with the accompanying drawings.

Hereinafter, with reference to the accompanying drawings illustrating the configuration and operation of the embodiment of the present invention, the configuration and operation of the present invention shown in the drawings and described by it will be described as at least one embodiment, By the technical spirit of the present invention described above and its core configuration and operation is not limited.

4 is a block diagram of the overall configuration of the channel equalizer according to the present invention, and is largely comprised of a channel estimator 400 and a channel equalizer 500. The channel estimator 400 estimates a channel even in a data interval by applying a conjugate-gradient method to the received signal. The channel equalizer 500 obtains the weighted sum using the impulse response of the channel estimated by the channel estimator 400, and then subtracts the weighted sum from each TDL output of the front end filter to remove the post-ISI from the data interval. do.

The channel equalizer 500 may use the modified crystal feedback equalizer of FIG. 2 as it is. The present invention is applicable not only to the modified decision feedback equalizer described above, but also to the channel equalizer using the estimation result of the channel estimator.

According to an embodiment of the present invention, the channel equalizer 500 uses the modified crystal feedback equalizer of FIG. 2. That is, the channel equalizer 500 includes a front end filter 510, a feedback filter 520, and a determination unit 530. The front end filter 510 includes a TDL 511, a post-ISI remover 512, a coefficient updater 513, and an adder 514. The rear filter 520 includes a TDL 521 and a weight adder 522. The determination unit 530 is a determination unit having eight levels of reference signals as shown in FIGS. 1 and 2.

The present invention configured as described above first describes the detailed operation of the channel estimator 400.

When the transmission signal is called x (n), the impulse response of the discrete equivalent channel is called h (n), and the white noise is called w (n), the input signal y (n) input to the receiver is expressed as Can be expressed as:

The channel estimator 400 receives y (n) as an input and estimates the impulse response h (n) of the discrete equivalent channel that the original signal x (n) has passed through, thereby estimating the finite impulse response of the channel. Outputs the function.

Here, the simplest method of channel estimation by the channel estimator 400 assumes that the training signal periodically added to the transmission signal is a white signal, and then the training signal passed through the channel and the training signal previously stored in the receiver. There is a Simple Correlation Method (SCM) that obtains a cross correlation value and outputs it as a frequency response of an estimated channel. This method is simple, so it can be implemented with less hardware. On the other hand, if the training signal is not white, the estimation error will be large. Moreover, the wider the channel estimation region, the more data will be present on both sides of the training signal. It is greatly influenced by, so accurate channel estimation is impossible.

Meanwhile, the LS estimation method (LSM), which is known as a more accurate method, can accurately estimate a channel even when the training signal does not have a white characteristic, compared to the simple correlation method. In other words, the training time is detected, and the cross correlation value p between the training signal transmitted through the channel and the training signal known to the receiver during the training time is obtained, and the autocorrelation matrix R of the training signal is obtained. Then, to remove the autocorrelation portion existing in p that is the cross correlation value between the received signal and the original training signal. By performing the matrix operation of, we can estimate the channel more accurately.

That is, the least square method described above can obtain a more accurate estimation channel at the cost of a complex implementation than the simple correlation method.

However, as we saw above, the least squares method is complex Since the operation must be performed, the calculation amount is large and the hardware complexity is high. Therefore, when estimating a channel using only field synchronization, a channel estimation process is performed at intervals of about 25 ms so that the computational amount is not high compared to the actual time, and thus the channel can be implemented. However, the update rate of the channel estimate is too slow to operate on a real dynamic channel, and the computational amount is too large to increase the update rate, which is difficult to physically implement.

Therefore, in the present invention The channel equalization is performed by applying a conjugate-gradient method of estimating a channel by an iterative operation instead of performing the operation directly.

The conjugate-gradient method described above is derived from the conjugate-direction method. The conjugate-direction method is shown in the August 4, 1994 document entitled "An introduction to the conjugate gradient method without the agonizing pain" by Jonathan Richard Shewchuk.

3 is a view for explaining the concept of the conjugate-direction method. In Figure 3 The point corresponding to is x and the initial value set as the starting point of the algorithm is x (0) . E.g Assuming that the values exist in two-dimensional space, we can reach x (0) to x in two steps if we move exactly as needed in each direction of the Cartesian axis at a time. In each step, the same operation as in Equation 3 below is repeatedly performed.

In Equation 3 described above Is in Indicates the search direction for the furnace. x in the i + 1th step The error vector of As can be seen in Figure 3 Wow Are orthogonal to each other.

therefore, Can be obtained as in Equation 4 below.

Substituting Equation 3 into Equation 4 described above may be expressed as Equation 5 below.

3, Navigation vector from x to x Direction of In the opposite direction. therefore It can be seen that and are perpendicular to each other. In other words, the following relationship is established.

However, in the actual situation, the value of x cannot be used because the value of x is not known.

Therefore, when using the conjugate-direction method, it is assumed that the direction vectors satisfy the following Equation 7 instead of being orthogonal to each other.

then, Can be obtained as in Equation 8 below.

here, Is a value defined as in Equation 9 below.

At this time, the direction search vector Set of linearly independent vectors It can be obtained from Equation 10 below.

And, the above formula (9) Satisfies the relationship as shown in Equations 11 to 13 below.

If r vectors are used instead of the vector set U in Equation 10, Equation 14 is obtained.

Equation 14 described above is called a conjugate-gradient method.

And, with the above equation (12) Taking the inner product of can be expressed as Equation 15 below.

Applying Equation 13 described above May be expressed as in Equation 16 below.

Therefore, by the above equation (10) May be expressed as in Equation 17 below.

And, To If you apply the equation (8) above May be expressed as in Equation 18 below.

Equation 10 described above in Equation 18 If is applied as shown in Equation 19 below Can be represented.

Direction search vector by iterative operation as above And then update the value of x using that value Converge to a value.

Accordingly, the channel estimator 400 applies the above-described conjugate-gradient method to the determination value of the determiner 530 to perform channel estimation not only in the training section but also in the data section. The estimated channel estimate is input to the weighted adder 522 of the feedback filter 520 of the channel equalizer 500.

In this case, the TDL 511 of the front end filter 510 of the channel equalizer 500 includes a plurality of serially connected delayers, delaying the input signals sequentially and outputting the input / output signals of the respective delayers to post-ISI. Output to reject (512). The post-ISI remover 512 is composed of a plurality of subtractors, each subtractor being fed back from the weighted adder 522 of the feedback filter 520 in input / output signals of each delayer of the TDL 511. Each corresponding weighted sum is subtracted to remove the effect of post-ISI, which is interference due to past symbols. The coefficient updater 513 is composed of a plurality of multipliers, and each multiplier performs coefficient update by multiplying a previous coefficient value corresponding to the output of each subtractor of the post-ISI remover 512. The outputs of the multipliers of the coefficient updater 513 are output to the adder 514, are added together, and then output to the determiner 530.

The determination unit 530 determines the reference signal closest to the output of the front end filter 510 by comparing the output of the front end filter 510 with the 8-level reference signal defined by the 8-VSB ATSC standard. Determined by the value. The determination value is input to the TDL 521 and the channel estimator 400 of the feedback filter 520. That is, the decision unit 530 may be an eight-level decision or slicer, or may be a decision predictor using a slice predictor. The determination unit 530 may use other known crystals. Here, decision and slice are used interchangeably.

The TDL 521 of the feedback filter 520 is composed of a plurality of serially connected delayers, which sequentially delay the determination signal of the determination unit 530, and weighted the input / output signals of each delayer. summation) (522). The weight adder 522 is weighted as shown in Equation 1 using the channel estimation value of the channel estimator 400 in which channel estimation is performed even in the data interval and the input / output signals of each delay unit of the TDL 521. After summation, the sum is output to each subtractor of the post-ISI remover 512 of the front end filter 510.

The postcursors, i.e., subtract the corresponding weighted sum of the weighted adder 522 from the input / output signals of the respective delayers of the TDL 511 in the post ISI remover 512 of the front end filter 510. Post-ISI is removed.

The signal from which post ISI has been removed by the post ISI remover 512 is output to the coefficient updater 513. In each multiplier of the coefficient updater 513, a coefficient update is performed by multiplying a previous coefficient value corresponding to the output of each subtractor of the post ISI remover 512, and in the adder 514, the coefficient updater 513 Adding and outputting the output of each multiplier eliminates pre-ISI, the interference due to future symbols, and ends all equalization.

That is, the conventional channel estimation is performed only through the PN sequence of the field sync interval. In the present invention, since the frequent channel estimation is possible even in the data interval in a conjugate-gradient manner, the equalizer coefficient can be quickly updated even in the dynamic channel. Therefore, the performance of the dynamic channel is improved.

In the conventional simple correlation method or the least-square method, the channel estimation is necessary only because the exact matrix and the vector value are required.

However, in the present invention Since the channel is estimated by an iterative operation without performing the matrix operation directly, the output of the channel equalizer may be determined using a maximum likelihood sequence estimator (MLSE) instead of the determination unit 530 of FIG. 4.

In this case, a highly reliable determination value can be obtained even in the data section instead of the field synchronization section. Therefore, the autocorrelation matrix and the cross-correlation matrix are continuously updated in the data section, thereby making it possible to update the equalizer coefficient faster. This improves performance on dynamic channels.

FIG. 5 is a block diagram illustrating an exemplary embodiment of a channel equalizer using the MLSE, wherein the TDL of the feedback filter 520 is determined by the determination unit 530, and the MLSE determination estimated by the MLSE unit 700. It receives the feedback value and delays it sequentially.

In addition, the channel estimator 600 applies the conjugate-gradient method with the MLSE decision feedback signal and the equalizer input signal of the MLSE unit 700 having high reliability even in the data section to perform channel estimation in the data section as well as the training section. Perform.

The MLSE unit 700 estimates a transmission signal from the output of the channel equalizer 500, and outputs the estimated MLSE decision feedback signal to the TDL of the channel estimator 400 and the feedback filter 520. At this time, the current estimated signal is obtained using the correlation of past symbols. The number of past symbols used can be arbitrarily determined, and the more symbols used, the higher the probability of estimating an accurate signal. The MLSE unit 700 may use various algorithms according to the type of coding of the transmitted signal. For example, when convolutional encoding is used at the transmitter side, the MLSE unit 700 estimates a transmission signal using a Viterbi decoding algorithm. In addition, when trellis encoding is used at the transmitting side, the MLSE unit 700 estimates a transmission signal using a trellis decoding algorithm.

That is, the MLSE unit 700 determines and outputs an 8-level VSB signal by applying a Viterbi decoding algorithm or trellis decoding algorithm at every decoding depth. As the decoding depth is increased, the reliability of the MLSE decision is improved.

6 illustrates this and shows the symbol error rate of the MLSE decision according to the decoding depth.

6, it can be seen that the symbol error rate of the decision using the slice predictor is lower than that of using the 8-level determiner. In particular, as the decoding depth of the MLSE increases, the symbol error rate of the decision decreases gradually. Able to know.

Accordingly, the determination value fed back from the MLSE unit 700 has a delay of 12 symbols in time as the decoding depth increases by one. That is, the decision value returned at depth 2 is a decision value 12 symbols earlier in time than the decision value feedback at depth 1. As described above, since the reliability of the decision increases as the decoding depth increases, inputting a decision value output at every decoding depth into the existing TDL can minimize the error of the decision included in the feedback filter, that is, the TDL. It can improve performance.

FIG. 7 illustrates in detail the feedback of the determination values output at all decoding depths of the MLSE to the TDL. When the delay when the MLSE is implemented in hardware is m symbols, the determination value output at depth 1 has a (12 + m) symbol delay compared to the determination value output from the determination unit 530. Therefore, the determination value output from the determination unit 530 is used as the input of the first tap, and the determination value output at depth 1 of the MLSE is used as the input of the (12 + m + 1) th tap of the TDL. On the other hand, the decision value output from depth 2 has a 12 symbol time delay compared to depth 1, so it is used as the input of the (24 + m + 1) th tap of TDL. In this way, the determination value output at each decoding depth is used as an input of the TDL at intervals of 12 taps.

Therefore, the improved performance can be obtained as compared with the channel equalizer using only the conventional decision unit 530.

As described above, the present invention estimates a channel even in the data interval by using a conjugate gradient algorithm, and applies the channel estimate to the modified decision feedback equalizer used in the existing 8VSB receiver, thereby improving the equalizer performance in the dynamic channel.

As described above, the channel equalizer according to the present invention performs channel equalization by applying channel estimation values through channel estimation of the conjugate gradient method to the modified decision feedback equalizer.

First, unlike the channel estimator using the least square method Since the calculation is performed by an iterative operation rather than directly, the hardware complexity of the channel estimator is reduced.

Second, since the MLSE can be used to obtain a highly reliable determination value not only in the field synchronization section but also in the data section, the autocorrelation matrix R and the cross-correlation vector p can be calculated in the data section. Accordingly, by performing channel equalization not only with the field synchronization period but also with the channel estimates frequently made in the data interval, the performance of the equalizer in the dynamic channel can be improved.

Those skilled in the art will appreciate that various changes and modifications can be made without departing from the spirit of the present invention.

Therefore, the technical scope of the present invention should not be limited to the contents described in the embodiments, but should be defined by the claims.

1 is a block diagram of a general decision feedback equalizer

2 is a general block diagram of a modified decision feedback equalizer;

3 is a diagram for explaining a conjugate gradient method for channel estimation.

4 is a block diagram illustrating a channel equalization apparatus according to an embodiment of the present invention.

5 is a block diagram illustrating a channel equalization apparatus according to another embodiment of the present invention.

FIG. 6 is a diagram illustrating an example of a symbol error rate of an MLSE determination according to a decoding depth in the MLSE unit of FIG. 5. FIG.

7 is a diagram illustrating an example of a TDL configuration of a feedback filter through feedback of an MLSE unit of FIG. 5;

Explanation of symbols for main parts of the drawings

400: channel estimator 500: channel equalizer

510: Shear filter 511,521: TDL

512: post ISI removal unit 513: coefficient update unit

514: adder 520: feedback filter

522: weighted addition unit 530: determination unit

700: MLSE part

Claims (10)

A channel equalizer for recovering an original signal from a digital TV received signal passing through a channel, A channel estimator for estimating a channel in a field sync interval and a data interval by applying a conjugate gradient algorithm to an input determination value; And And a weighted sum of the channel estimate estimated by the channel estimator and the determined values delayed by the feedback filter, respectively, and then the input delayed by the shear filter. And a channel equalizer for removing interference due to past symbols by subtracting each weighted sum, removing interference due to future symbols by updating coefficients, and outputting a determination value determined by the determiner to the channel estimator. Channel equalizer, characterized in that. The method of claim 1, wherein the front end filter of the channel equalizer A tap delay line (TDL) configured to receive a plurality of serially connected delayers and sequentially receive a received signal passing through a channel; A post ISI removal unit configured to remove the interference due to past symbols by subtracting a corresponding weighted sum fed back from the feedback filter from the output signal of each stage of the TDL; A multiplier comprising a plurality of multipliers, each multiplier performing a coefficient update by multiplying a previous coefficient value corresponding to an output of each subtractor of the post-ISI elimination unit to remove interference due to future symbols; And an adder for adding and outputting the outputs of the respective multipliers of the coefficient update unit. The feedback filter of claim 1, wherein the feedback filter of the channel equalizer is A TDL composed of a plurality of serially connected delayers, and sequentially delaying the decision value of the determinant; And a weighted summation outputting the channel estimation value of the channel estimation unit and the output signal of each stage of the TDL in the data section to obtain a weighted sum and output the weighted summ to the front end filter. Channel equalizer. The determinator of claim 1, wherein the channel equalizer And an eight-level determiner for comparing the output of the front end filter with an eight-level reference signal defined by the 8-VSB ATSC standard and outputting a reference signal closest to the output of the front end filter as a determined value. Equalizer. The determinator of claim 1, wherein the channel equalizer MLSE unit for estimating a transmission signal by applying a maximum likelyhood sequence estimation (MLSE) method using the correlation of past symbols to the output signal of the front end filter and outputting the estimated MLSE decision feedback signal as a decision value Equalizer. A channel equalizer for recovering an original signal from a digital TV received signal passing through a channel, A channel estimator for estimating a channel in a field sync interval and a data interval by applying a conjugate gradient algorithm to an input MLSE decision feedback signal; A feedback filter that receives the 8-level decision value or the MLSE decision feedback signal and sequentially delays it, and obtains a weighted sum from each delayed signal and the channel estimate estimated by the channel estimator; Receives the received signal passing through the channel and delays it sequentially, removes the interference caused by the past symbols by subtracting the corresponding weighted sum returned from the feedback filter from each delayed signal, and updates the coefficients to remove the interference caused by future symbols. Shear filter to remove; The output of the front end filter is compared with the eight-level reference signal defined by the 8-VSB ATSC standard, respectively, and the reference signal closest to the output of the front end filter is determined as an eight-level decision value and outputs to the feedback filter. A level determining unit; And An MLSE unit for estimating a transmission signal by applying a maximum likelyhood sequence estimation (MLSE) method using correlation of past symbols to an output signal of a previous filter and outputting an estimated MLSE decision feedback signal to the feedback filter and a channel estimator Channel equalizer, characterized in that configured to. The method of claim 6, wherein the shear filter is A tap delay line (TDL) configured to receive a plurality of serially connected delayers and sequentially receive a received signal passing through a channel; A post ISI removal unit configured to remove the interference due to past symbols by subtracting a corresponding weighted sum fed back from the feedback filter from the output signal of each stage of the TDL; A multiplier comprising a plurality of multipliers, each multiplier performing a coefficient update by multiplying a previous coefficient value corresponding to an output of each subtractor of the post-ISI elimination unit to remove interference due to future symbols; And an adder for adding and outputting the outputs of the respective multipliers of the coefficient update unit. The method of claim 1, wherein the feedback filter A TDL comprising a plurality of serially connected delayers, which selectively receives the 8-level decision value of the 8-level decision part and the MLSE decision feedback signal of the MLSE part, and sequentially delays And a weighted summation outputting the channel estimation value of the channel estimation unit and the output signal of each stage of the TDL in the data section to obtain a weighted sum and output the weighted summ to the front end filter. Channel equalizer. The method of claim 8, wherein the TDL is A channel equalizer, characterized in that for receiving the MLSE decision feedback signal at 12 symbol intervals delayed sequentially. The method of claim 8, wherein the TDL is A channel equalization apparatus, characterized in that the decoding depth is sequentially increased from the MLSE and sequentially received in parallel with the decision feedback signals having a time delay of 12 symbols.