KR100480881B1 - A Blind Adaptive Decision Feedback Equalizer using Error Feedback - Google Patents

Abstract

The present invention relates to an adaptive decision feedback equalizer using error feedback, and in particular, by using an MMA algorithm and an LMS algorithm in combination, the present invention can adapt to a channel without training permutation, and error in a conventional adaptive decision feedback equalizer. A blind adaptive decision feedback equalizer using error feedback that improves channel equalization performance by adding an error feedback filter, which is a part capable of feedback, reduces the correlation of error signals.

Description

A blind adaptive decision feedback equalizer using error feedback

 The present invention relates to a blind adaptive decision feedback equalizer using error feedback, and in particular, by using an MMA algorithm and an LMS algorithm in combination, the present invention can adapt to a channel without training perturbations, and the conventional adaptive decision feedback. The present invention provides an adaptive decision feedback equalizer using error feedback that improves channel equalization performance by adding an error feedback filter capable of feeding back an error to the equalizer to reduce the correlation of error signals.

In a digital communication system, various distortions occur as a signal transmitted from a transmitter passes through a channel. Factors that cause the distortion include Gaussian thermal noise, impulse noise, and addition or multiplication noise due to fading, which is a phenomenon in which the strength of a signal changes in time.

In addition, inter-symbol interference due to multipath channels, non-ideal frequency response, group delay, etc. is a major cause of degrading the performance of high-speed digital data communication systems.

The technique of compensating for the distortion of the channel and reducing the bit error at the receiving side is called channel equalization, and the apparatus for performing the channel equalization is called equalizer. The equalizer compensates for the magnitude and delay characteristics of the signal received at the receiver, thereby increasing the quality of the communication path without increasing the power of the transmitted signal or changing the channel bandwidth.

The equalizer operates in tracking mode and decision directed mode. In the tracking mode, the transmitter sends a promised training sequence and the equalizer receives the training sequence to initialize the tap coefficients. When the tap coefficients of the equalizer converge, the mode is directly switched from the tracking mode to the judgment mode, and the residual error of the channel is compensated for by the determined data instead of the training permutation.

However, standards such as HDTV, Local Multipoint Distribution Service (LMDS), and downlink in DOCSIS require the adaptation of channels without training permutations or pilot channels. A blind adaptive algorithm is proposed to support this standard.

The blind adaptation algorithm starts channel equalization without prior information of the transmitted signal using the statistical properties of the transmission signal, and the algorithms used are RCA (Reduced Constellation Algorithm; RCA) and CMA (Constant Modulus Algorithm). ) And MMA (Multi Modulus Algorithm).

The RCA is a blind algorithm that starts channel adaptation by reducing the constellation of the transmission signal, and the CMA draws a circle around the origin of the constellation and calculates the distance from the circle to reduce the distance. It is a blind algorithm that adapts.

In addition, MMA is a blind algorithm that independently divides the real axis and the imaginary axis of the constellation diagram, so that the equalizer using the MMA allows the real axis and the imaginary axis to operate independently of each other using two transversal filters. Designed.

1 to 2 show a general configuration of an equalizer using a conventional MMA algorithm.

First, FIG. 1 shows an equalizer using a conventional MMA algorithm. The equalizer is composed of two filters 20a and 20b having a real axis and an imaginary axis to update a tap coefficient C _n and D _n to adapt a channel. The output of the filter unit 20, the determiners 30a and 30b for discriminating and determining the waveform inputted through the equalization filter unit 20 into the real axis and the imaginary axis, and the output of the equalization filter unit 20 ( Error function () ) And real / imaginary error function generators 10a and 10b (coefficient generators) supplied to the two filters 20a and 20b.

Next, Fig. 2 shows an equalizer using the MMA algorithm for the decision feedback structure, which includes a front filter 40a for processing using a signal input through a channel and a rear filter 40b for processing with an already determined signal. The output of the equalization filter unit 40, the determiner 60 which discriminates and determines the waveform inputted through the equalization filter unit 40 by the real axis and the imaginary axis, and the output of the equalization filter unit 40 ( Error function () ) And an error function generator 50 (coefficient generator) for supplying to the front filter 40a and the rear filter 40b in the equalization filter section.

3 shows a filter structure of an equalizer using the MMA algorithm of the prior art, which is one filter used in the equalization filter section of FIGS. That is, in the equalizer shown in FIG. 1, the equalization filter unit comprises two filters shown in FIG. 3 in parallel, and in the equalizer of the crystal feedback structure shown in FIG. 2, the filters of FIG. Configure.

The filter is the input of the equalizer Delay unit 70 for delaying the Pair of complexes The conversion unit 80 and the output from the conversion unit 80 And error functions generated by the error generator Multiply each complex by a complex multiplier 90 and an accumulator 100 for generating tap coefficients, a register 110 for storing the tap coefficients, and a tap coefficient stored in the register 110 for each delay unit. And a multiplier 120 and an adder 130 for generating an output.

Referring to the operation of the equalizer using the MMA algorithm of Figures 1 and 2 configured as described above are as follows.

First, the equalizer using the MMA algorithm shown in FIG. 1 divides the real axis and the imaginary axis by using the filter structure shown in FIG. 3 so that two FIR filters are independently adapted. Error function and Is obtained by dividing the real and imaginary axes as shown in Equation 1 by using the output of the equalization filter unit.

Equation 2 shows the coefficient update equation of the MMA algorithm.

Shown in Equation 2 Denotes the complex conjugate of the input signal of the equalizer. and Are the tap coefficients of the real axis and the imaginary axis, respectively.

In addition, referring to the operation of the decision feedback equalizer shown in FIG. 2, unlike the equalizer shown in FIG. 1, the inter-symbol interference is eliminated using a predetermined value. That is, unlike the equalizer of FIG. 1, channel adaptation is possible by only one tap coefficient update. In addition, an error function for generating an error in the error function generation unit is obtained as in Equation (3). The error function of Equation 3 is not obtained by dividing the real axis and the imaginary axis like Equation 1, but simultaneously obtains the values of the real axis and the imaginary axis.

Equation 4 shows the coefficient update equation of the MMA algorithm. Unlike Equation 2, Equation 4 simultaneously updates the values of the real axis and the imaginary axis.

In the front filter of FIG. 2, the signal received through the channel is processed to compensate for the distortion, and the rear filter compensates for the remaining residual error by using the determined value.

In the above equations Is a step size that determines the channel adaptation rate. The larger the value, the faster the adaptation, but the greater the residual error, the greater the likelihood that the equalizer will diverge or vibrate. Also, the Smaller values have the advantage of slower adaptation speed but smaller residual error.

Also, adaptive constant Is obtained by equation (5).

The adaptation constant Is the squared value of the equalizer output and the adaptive constant as shown in equation (1). The error function is generated by the difference of the squares of.

The adaptation constant The value has the statistical characteristics of the transmitted signal The value has a fixed value when the modulation and demodulation method is determined. 64 QAM , 128 QAM At 256 QAM to be.

The generalized form of the MMA is GMMA (Generalized MMA; hereinafter referred to as GMMA). The GMMA has a very large error function when the MMA is applied to a constellation modulation method such as higher order QAM. If the error function is very large, the equalizer is more likely to diverge or vibrate. To prevent this case, the proposed algorithm is GMMA.

The GMMA algorithm is a constant applied based on the size of the output of the equalizer It is an algorithm that suggests the value of error function created by changing the value of within the proper range. The error function generation method of GMMA is shown in FIG. 4, which shows a case where GMMA is applied to 256 QAM.

Applied above Is changed according to Equation 6 according to the absolute value of the equalizer output.

As shown in FIG. 4, the output of the equalizer is 64 QAM modulation scheme in the range of (1). Same as Is used and the equalizer output is in the range of (2) Is used, in the range of (3) This is applied to reduce the size of the error function to stabilize the operation of the equalizer.

As such, the MMA algorithm is a channel adaptation algorithm proposed to be suitable for the QAM and CAP methods, which are orthogonal transmission modulation and demodulation methods.

As described above, since the equalizer using the conventional MMA algorithm of FIG. 1 updates the tap coefficients of the real axis and the imaginary axis separately, the number of registers, multipliers, and adders is doubled.

In addition, in the case of an equalizer using the MMA algorithm of FIG. 2 in the decision feedback structure, the precursor part of the inter-signal interference can be removed by using the determined value, but the after-part of the inter-signal interference cannot be removed and remains completely removed. Poor inter-symbol interference deteriorates the performance of the equalizer and causes a performance degradation and divergence at low signal-to-noise ratios. In addition, when the decision feedback equalizer is applied, a signal mixed with an error is initially fed back, which acts as a factor for degrading the performance of the equalizer, and the initial error propagation problem is a factor that degrades the initial performance of the decision feedback equalizer. there was

Therefore, in order to solve the above problems, the present invention can apply the MMA algorithm to the channel adaptation by updating only one tap coefficient, reduce the hardware size by interworking with the LMS algorithm, and maintain the inter-code remaining in the decision feedback equalizer. The aim is to reduce the initial error propagation problem and improve the channel equalization performance by reducing the correlation by using the error feedback filter.

In order to achieve the above object, the present invention includes an equalization filter unit for adapting the channel by updating the tap coefficient by the front filter, the rear filter and the error feedback filter to filter the signal received from the transmission channel in parallel; An adder for generating an output of the equalization filter unit; A determiner for discriminating and determining the waveform output through the adder into a real axis and an imaginary axis, and supplying the determined signal to the rear filter; After receiving a complex signal that is an output of the equalization filter unit to generate an error function, the adaptive decision feedback equalizer using an error feedback including an error function generator for supplying the front filter and the error feedback filter.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

FIG. 5 shows a structure of a decision feedback equalizer in which an error feedback filter is added according to the present invention, and is present by using a forward filter (140a, feedforward filter) for filtering and outputting a signal received from a transmission channel. The feedback filter 140b, which removes the interference between the signals, and the error feedback filters 140c, which reduce the correlation of the remaining errors by feedback, are configured in parallel to form a tap coefficient. , , The equalizer filter 140 for adapting the channel to the channel, the adder 150 for generating the output of the equalization filter 140, and the real and imaginary axes of the waveforms output through the adder 150. Generate an error function by receiving a determiner 160 for supplying the determined signal to the rear filter 140b and a complex signal which is an output of the equalization filter unit, and calculating an error function Part 210 (coefficient generator).

While the filter structure used in the conventional equalizer is calculated by dividing the imaginary axis and the real axis, the structure of the filter constituting the equalizer using error feedback performs a complex operation as shown in FIG. Delay unit 220 for delaying the Pair of complexes The conversion unit 230 and the output from the conversion unit 230 And error functions generated by the error generator Multiply each complex by a complex multiplier 240 and an accumulator 250 for generating tap coefficients, a register 260 for storing tap coefficients, and a tap coefficient stored in the register 260 for each delay unit. And a multiplier 270 and an adder 280 for generating an output of the equalization filter unit. That is, the channel is equalized through one tap coefficient update.

The equalizer according to the present invention of FIG. 5 has the same number of taps as the conventional decision feedback equalizer shown in FIG. 2, and an error feedback filter is added to the equalization filter unit to improve channel adaptation performance than the conventional equalizer. You can see that. In addition, when comparing the number of gates, it was confirmed that the number of gates is smaller than that of the equalizer of FIG. 1 and is similar to the gate number of the equalizer of FIG. In addition, since the MMA algorithm has the same tap coefficient update formula as the LMS algorithm, it can be linked without additional control logic.

7 shows an error generator for generating an error function in conjunction with an LMS algorithm, wherein the error function generator includes an MMA and GMMA error function generator for generating an error function of MMA and GMMA algorithms as an output of the equalization filter; An LMS error function generator that generates an error function of the LMS algorithm using the output of the equalization filter unit and the output of the determiner, and one of the error functions generated by the MMA and GMMA error function generators and the LMS error function generator is added to the MSE value. And a multiplexer for selecting a constant for determining an adaptation rate based on the MSE value.

When the constellation degree of the modulation / demodulation method is 128 QAM or more, three equalizers (390a, 390b, 400) are used to select an adaptation constant and a constant (stepsize) for determining the adaptation speed according to the output value of the equalizer so that the equalizer operates stably. Limit the value of the error function to a certain range.

In the error function generator of FIG. 7, the error function of the MMA algorithm and the error function of the GMMA algorithm are output values of the equalization filter unit. Can only be generated. The error function generator 210 outputs the equalization filter Is divided by the imaginary part and the real part, and the squares and adaptive constants of the second square operator 360 and the third multiplication operator 380, and the multiplication operator 360 and 380, respectively. A second subtractor 350 and a third subtractor 370 for obtaining a difference from a value, and a difference generated by the subtractors 350 and 370; It consists of a first multiplier 330 and a second multiplier 340 to multiply with and generates an error function of the MMA and GMMA algorithm.

When the constellation is above 128 By value , , The GMMA algorithm that selects one of the values is used. The constant stepsize, which determines the adaptation speed, is obtained from many simulations. , It is used to select from the values of. Fixed when constellation is less than 128 Only the value is used as the adaptive constant, and the value of the constant that determines the adaptation speed It is fixed and used.

Next, the error function of the LMS algorithm in the error function generator of FIG. Output of equalizer Is generated through the first subtractor 290. The error function of the LMS algorithm and the error function of the MMA and GMMA algorithms are selected through the multiplexer 320 as a control signal using an MSE value obtained by squaring the error function of the LMS algorithm using the first square calculating unit 300. In this case, a squarer is used instead of a multiplier in the first square operation unit 300 to reduce the size of hardware and enable high-speed operation. It also consists of a multiplier 310 that multiplies the selected error function and a constant to determine the adaptation speed.

When the channel adaptation starts for the first time or the channel condition deteriorates, the channel adaptation succeeds by selecting the error function of the MMA algorithm.If the MSE value is less than the reference value, the error function of the LMS algorithm is selected and the residual channel remains. Compensate for errors In addition, the error values obtained as described above are used as inputs to the error feedback filter to reduce the correlation of residual errors.

The error function and tap coefficient update equation of the MMA algorithm are as shown in Equations 7 and 8.

The error function of the MMA algorithm calculated in Equation 7 simultaneously generates the LMS algorithm error function shown in Equation 8, and selects one of two error functions based on the MSE value.

The tap coefficient update equation calculated in Equation 9 is similar to the tap coefficient update equation of the LMS algorithm. Until the error function shown in Equation 7 is generated, the real axis and the imaginary axis operate independently, but in the filter, it is regarded as one complex signal to generate tap coefficients. The tap coefficient update equation shown in Equation 9 is applied to the decision feedback equalizer to which the error feedback filter is added.

Equation 10 , , Denotes the tap coefficients of the front filter, the rear filter, and the error feedback filter, respectively. Compared with the MMA equalizer of the conventional crystal feedback structure, the number of tap coefficients updated is the same. But The tap coefficient is a tap coefficient added by applying an error feedback filter, and the MMA algorithm according to the present invention changes the tap coefficient updating formula so that one tap coefficient is However, the channel can be adapted.

Therefore, the output Y_n of the equalization filter unit is defined as in Equation 11 below.

Here, M, N, and L represent the number of taps of a forward filter, a feedback filter, and an error feedback filter.

The operation of the decision feedback equalizer using the MMA and LMS algorithms and adding an error feedback filter will now be described with reference to FIG. 5.

Generated by the sum of the front filter 140a, the rear filter 140b, and the error feedback filter 140c. Create MMA and GMMA error functions with values, The LMS error function is generated from the value and output of the determiner 160. At this time, the square of the LMS error function becomes the MSE value, and one of the MMA and GMMA error functions and the LMS error function is selected as the multiplexers 200a and 200b based on the MSE value.

The selected error function is supplied to the front filter 140a and the rear filter 140b, and the tap coefficient is generated by the filter function and the delayed filter input value by the filter structure. The selected error function is used as an input to the error feedback filter 140c, and the tap coefficient of the error feedback filter 140c is generated by the delayed error function similarly to the front filter 140a and the rear filter 140b by the filter structure. .

In the equalizer structure, the front filter 140a is used to remove the intersymbol interference of the current symbol by the signal before being determined, and the rear filter 140b is the intersymbol of the current symbol caused by the previously determined signal. It is used to remove the interference. The error feedback filter 140c is used to feedback the selected error function to reduce the correlation of residual components of intersymbol interference that could not be removed in the front and back filters.

The output value of the filter is generated by multiplying the tap coefficient thus generated and the filter input value. In addition, the equalizer of the present invention is the same as the LMS algorithm and the tap coefficient update formula can be configured without the need for additional control logic.

As described above, the decision feedback equalizer which adds the error feedback filter of the present invention can be used by interworking with the MMA and the LMS algorithm, so that it can adapt to the channel without training permutation, and is generated by the same equation as the LMS algorithm and the tap coefficient update equation. It is easy to interwork with LMS algorithm, and only one tap coefficient can be updated.

Residual error that was not eliminated by the front and rear filters by feedback of the current intersymbol to remove the current intersymbol interference by returning the signal determined by the decision feedback structure with the error feedback filter. By reducing correlation, residual error is reduced and channel adaptation performance is increased. In addition, the decision feedback equalizer has a disadvantage in that an error is increased during initial channel adaptation because a symbol that is not completely removed from the inter-symbol interference is determined at the initial channel adaptation. The error feedback filter reduces an error correlation by using an error feedback filter. By reducing the errors that occur, you can reduce the initial increase in errors.

Therefore, the present invention can be applied to a system requiring channel adaptation without training permutations such as downlink of HDTV, LMDS, and DOCSIS.

1 shows an equalizer using a conventional MMA algorithm.

2 shows an equalizer using a conventional MMA algorithm for the decision feedback structure.

3 shows a filter of an equalizer using a conventional MMA algorithm.

4 is a diagram illustrating a GMMA used for a general 256 QAM.

5 illustrates an adaptive decision feedback equalizer in which an MMA algorithm and an LMS algorithm are linked to a decision feedback structure using error feedback according to the present invention.

FIG. 6 illustrates a filter used in an adaptive decision feedback equalizer using the error feedback of FIG. 5; FIG.

FIG. 7 is a diagram illustrating an error function generator for generating an error function of an MMA algorithm and an error function of an LMS algorithm in an adaptive decision feedback equalizer using the error feedback of FIG. 5; FIG.

<Description of Symbols for Main Parts of Drawings>

10a: real error generating unit 10b: imaginary error generating unit

20,40,140: Equalization filter part 20a: Real FIR

20b: Imaginary FIR 30a: Real Judgment

30b: Imaginary judgment 40a, 140a: Front filter

40b, 140b: Rear filter 50,210: Error function generator

60,160: Determinator 70,220: Delay unit

80: conversion unit 90: tap coefficient generation complex multiplier

100: tap coefficient generation accumulator 110: tap coefficient storage register

120: output generator multiplier 130: output generator adder

140c: error feedback filter 150: adder

160: determiner 170: LMS error generation unit

180: MMA & GMMA error generator 190: MSE generator

200a, 200b: multiplexer 230: conjugate complex conversion unit

240, 270: multiplier 260: register

250, 280: Accumulator 290, 350, 370: Subtractor

300, 360, 380: square calculation unit 310, 330, 340: multiplier

320, 390a, 390b, 400: Multiplexer

Claims (6)

The front filter that filters and outputs the signal received from the transmission channel, the rear filter that removes the interference between the existing signals using the already determined signal, and the error feedback filters that reduce the correlation of residual errors by feedback An equalization filter unit configured to update tap coefficients to adapt to a channel; An adder for generating an output of the equalization filter unit; A determiner for discriminating and determining the waveform output through the adder into a real axis and an imaginary axis, and supplying the determined signal to a rear filter; And an error function generator for receiving an complex signal that is an output of the equalization filter unit, obtaining an error function, and supplying the error function to the equalization filter unit. The blind adaptive decision feedback equalizer using the error feedback unit. The method of claim 1, The filter of the equalization filter unit performs a complex arithmetic operation and delays an input, a conversion unit representing the input as a complex conjugate, a tap coefficient by multiplying a result output from the conversion unit and an error function generated by an error generation unit. A complex multiplier and accumulator for generating a multiplier, a register for storing the tap coefficients generated by the multiplier and the accumulator, and a multiplier for multiplying and storing the tap coefficients stored in the register for each delay unit to generate an output of an equalization filter unit. A blind adaptive decision feedback equalizer using error feedback, comprising: and an adder. The method of claim 1, The error function generating unit generates an error function of the MMA and GMMA algorithms using the output of the equalization filter unit and the MMA and GMMA error function generating unit; An error function generator, a multiplexer selected by an MSE value from one of the error functions generated by the MMA and GMMA error function generators and the LMS error function generator, and a constant for determining an adaptation rate based on the MSE value. A blind adaptive decision feedback equalizer using error feedback, comprising: selecting a multiplexer. The method of claim 3, wherein The MMA and GMMA error function generators divide an output value of an equalization filter into an imaginary part and a real part, and square each of them to obtain a difference between an adaptive constant value and a value generated by the square / subtraction operation part and the equalization filter. A blind adaptive decision equalizer using error feedback, comprising: a multiplier for generating error functions of MMA and GMMA algorithms by multiplying negative output values The method of claim 3, wherein The LMS error function generation unit generates an error function of the LMS algorithm by using a difference between the output value of the determiner and the output value of the equalizer, and a signal generated by squaring the error function of the LMS algorithm generated by the subtractor. A blind adaptive decision equalizer using error feedback, comprising: a multiplexer for selecting one of an error function of an LMS algorithm and an error function of an MMA and GMMA algorithm. The method of claim 1, The tap coefficient updated in the front filter of the equalization filter part is Is renewed by The tap coefficient updated in the rear filter is Updated by The tap coefficient updated in the error feedback filter is (Where Is the tap coefficient, Is an error function, Is a constant for determining a channel adaptation rate). The blind adaptive decision feedback equalizer using error feedback.