KR100677915B1 - Noise prediction decision feedback equalizer - Google Patents

Abstract

A noise predictive decision feedback equalizer is provided to remove an inter symbol interference and reduce wrong decision probability of a symbol caused by AWGN(Additive White Gaussian Noise) by adding noise removal algorithm to the equalizer. A noise predictive decision feedback equalizer includes a first adder(260), a second adder(240), a noise predictive filter unit(250), a symbol decision unit(230), and a third adder(220). The first adder(260) subtracts an output value of an inverse filter unit from an output value of a forward filter unit. The second adder(240) subtracts an output value of the symbol decision unit(230) from the first adder(260). The noise predictive filter unit(250) receives an output value of the second adder(240) and outputs a predictive value about the noise. The third adder(220) subtracts an output value of the noise predictive filter unit(250) from the output value of the first adder(260), and provides the subtracted output value to the symbol decision unit(230).

Description

Noise prediction decision feedback equalizer {ROBUST NOISE PREDICTIVE DECISION FEEDBACK EQUALIZER}

1 is a diagram showing the structure of a crystal feedback equalizer according to the prior art.

2 shows the structure of a crystal feedback equalizer according to the present invention;

3 illustrates a linear cross filter of a crystal feedback equalizer in accordance with an embodiment of the present invention.

4 illustrates a detailed structure of a crystal feedback equalizer according to an embodiment of the present invention.

<Explanation of symbols for the main parts of the drawings>

100, 200: forward filter unit 110, 410: forward tap coefficient update unit

120, 210: reverse filter unit 129, 419: reverse tap coefficient update unit

130, 260, 420: first adder 170, 450: fourth adder

140, 230: symbol determination unit 150: preamble generation unit

160: switch 220, 440: third adder

240 and 430: second adder 250: noise prediction filter unit

301, 302, 303, 304: delay

311, 312, 313, 314, 315: Multiplier 320: Summer

330: tap coefficient update unit 401, 402, 403, 404: delay

405, 406, 407, 408, 409: Multipliers 411, 412, 413, 414: Delay

415, 416, 417, 418: multipliers

The present invention relates to an equalizer for reducing interference between symbols generated due to multipath fading in a radio channel, and more particularly, to a decision feedback equalizer for determining a received symbol using a preamble.

In general, in wireless communication, inter-symbol interference (hereinafter, referred to as 'ISI') occurs due to distorted multi-path fading in addition to the main channel. Since this greatly affects communication performance, an equalizer is used in the receiver of the wireless terminal to remove the ISI.

The basic equalizer is a linear equalizer (hereinafter, referred to as 'LE') using a Least Mean Square (LMS). Recently, however, a Decision Feedback Equalizer (hereinafter referred to as 'DFE') using a preamble is often used to compensate for errors caused by continuously feeding back noise.

The DFE improves performance with a high probability decision value due to the preamble without reinforcing noise by using a preamble signal as a basic equalizer at the receiving end. In this case, the preamble signal is a signal commonly known by the transceiver in order to synchronize the transmitter and the receiver and induce timing.

On the other hand, the performance of the equalizer is generally influenced by the number of actual coefficients of the equalizer, that is, not only the reference signal but also the characteristics of the channel and the phase of the sampler. However, the DFE can compensate for amplitude distortion without as much noise enhancement as in LE. Thus, the ability of the feedback portion to eliminate ISI due to many previous symbols in the DFE is much more free to choose the coefficient of the forward portion. At this time, the combination of the impulse response and the forward portion of the channel receives a nonzero sample along the main pulse. In other words, the forward portion of the DFE does not need to approach the inverse of the channel characteristics, and does not need to be sensitive to excessive noise enhancement and sampler phase. Therefore, most of the DFE is used to remove the ISI.

Hereinafter, a structure of a conventional general DFE will be described in detail with reference to FIG. 1.

1 is a view showing the structure of a crystal feedback equalizer according to the prior art. Referring to FIG. 1, a conventional DFE structure includes a forward filter 100 and a backward filter 120 formed of a linear transversal filter.

The floating value obtained by passing through the filter is determined by the symbol determiner 140 to be 0 or 1, thereby affecting a subsequent incoming symbol. In addition, the preamble generator 150 generates a preamble used to give reliability to the initially determined decision value.

In more detail, the signal x (k) received by the DFE is first input to the forward filter 100. The signal input to the forward filter 100 is delayed through a plurality of delayers 101, 102, 103, 104, and the received signal and the signal passing through the delayers are forward tap coefficients updating units. The tap coefficients provided at 110 are multiplied by multipliers 105, 106, 107, 108, and 109, respectively.

The output signals of the multipliers 105, 106, 107, 108, 109 are then summed in the first adder 130 and from each of the multipliers 125, 126, 127, 128 of the reverse filter 120. The output signal is subtracted. The value z (k) summed and subtracted by the first adder 130 is input to the symbol determiner 140. The symbol determiner 140 determines a value of 0 or 1 from the input value.

)을 입력받아, 복수의 지연기들(121, 122, 123, 124)로 입력하여 지연시키게 된다. 상기 역방향 필터(120)로 입력된 신호는 상기 복수의 지연기들(121, 122, 123, 124)을 통해 지연되고, 상기 수신 신호 및 각 지연기들을 거친 신호는 역방향 탭 계수(Tap coefficients) 갱신부(129)에서 제공되는 탭 계수들과 곱셈기들(125, 126, 127, 128)에서 각각 곱하여진다. 이때, 상기 각 곱셈기들(125, 126, 127, 128)에서 곱하여진 신호는 상술한 바와 같이 제1 가산기(130)로 입력되어 감산 연 산시키게 된다.In this case, the reverse filter 120 determines the value determined by the symbol determiner 140 (

) Is inputted to the plurality of delayers 121, 122, 123, and 124 to be delayed. The signal input to the reverse filter 120 is delayed through the plurality of delayers 121, 122, 123, and 124, and the received signal and the signal passing through the delayers are updated with reverse tap coefficients. The tap coefficients provided by section 129 and the multipliers 125, 126, 127, and 128 are respectively multiplied. In this case, the signals multiplied by the multipliers 125, 126, 127, and 128 are input to the first adder 130 and subtracted as described above.

On the other hand, the DFE is provided with a preamble generating unit 150 as described above to be used for error correction to increase the probability of determining the received signal. The preamble generator 150 generates and outputs a preamble or training sequence with a predetermined length.

)이 선택적으로 제4 가산기(170)로 입력된다. 상기 제4 가산기(170)에서는 상기 제1 가산기(130)의 출력 신호 z(k)에 상기 심볼 결정부(140)의 출력 값(

)을 감산 연산함으로써 에러 값(e)을 피드백시킨다. 즉, 상기 제4 가산기(170)의 출력 값(e)은 상기 순방향 필터(100)의 순방향 탭 계수 갱신부(110) 및 상기 역방향 필터(120)의 역방향 탭 계수 갱신부(129)로 제공되어, 다음 필터링시 연산되는 탭 계수들을 갱신하게 된다.That is, the signal of the preamble generator 150 or the output value of the symbol determiner 140 may be selected by the selection of the switch 160.

) Is optionally input to the fourth adder 170. In the fourth adder 170, the output value of the symbol determiner 140 is equal to the output signal z (k) of the first adder 130.

) Is fed back to the error value e. That is, the output value e of the fourth adder 170 is provided to the forward tap coefficient updater 110 of the forward filter 100 and the reverse tap coefficient updater 129 of the reverse filter 120. Then, the tap coefficients calculated at the next filtering are updated.

As described above, the general DFE is simply an equalizer having a crystal feedback capability using a preamble to LE. The DFE has the advantage of eliminating the forward error of the ISI in this way, but especially if an incorrect decision is fed back, the output of the DFE has an error effect on the next few samples while the incorrect decision is returned. have. Accordingly, there is a problem that the influence of the previous sample is likely to make a more inaccurate decision in the next samples. In particular, in a noisy environment, there is a risk that the initial error probability is increased due to the decision feedback of the equalizer used to remove the ISI.

It is therefore an object of the present invention to provide a noise prediction decision feedback equalizer which reduces the effects of noise by adding a noise cancellation algorithm to the equalizer for reducing intersymbol interference in a wireless environment.

Noise prediction decision feedback equalizer of the present invention for achieving the above object; A decision feedback equalizer comprising a forward filter unit and a reverse filter unit, comprising: a first adder which subtracts an output value of the reverse filter unit to an output value of the forward filter unit; and an output value of the symbol determiner unit to an output value of the first adder. Receiving a second adder for subtracting, an output value of the second adder, a noise prediction filter for outputting a predicted value for noise, and subtracting an output value of the noise prediction filter for an output value of the first adder And a third adder provided to the symbol determiner.

In this case, the noise prediction filter unit may calculate a noise prediction value through an average operation of a current input value and a previously input value from the second adder, and calculate a result by the following equation. .

In this case, λ is an oblivion element of the noise prediction filter unit, and N (k) represents an input value of the noise prediction filter unit.

The equalizer may further include a fourth adder provided to each of the forward filter unit and the reverse filter unit by subtracting an output value of the symbol determination unit to an output value of the first adder.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that in the following description, only parts necessary for understanding the operation according to the present invention will be described, and descriptions of other parts will be omitted without departing from the scope of the present invention.

Since the present invention generally aims at simply eliminating ISI, the widely used DFE is noise predictive (ie, reducing noise effects under additive white Gaussian noise (AWGN)). We propose an algorithm.

That is, the noise predictive equalizer according to the present invention reduces the influence of noise by using a preamble already known at the receiving end in an AWGN (Additive White Gaussian Noise) environment. Therefore, by adding to the noise prediction algorithm DFE, it is possible to enter a faster stabilization phase with attenuated noise during actual preamble operation.

Hereinafter, a structure of a noise prediction decision feedback equalizer according to an embodiment of the present invention will be described with reference to FIGS. 2 to 4.

2 is a diagram showing the structure of a crystal feedback equalizer according to the present invention. 2, the noise prediction decision feedback equalizer according to the present invention includes two linear transversal filters 200 and 210, a symbol decision unit 230, and a noise prediction filter unit. Filter 250). That is, in the configuration of the noise prediction feedback equalizer (Noise Predictive DFE) according to the present invention, a noise prediction filter matrix is added to the noise prediction filter unit 250 to the existing DFE.

First, the linear transversal filter is a structure included in a basic linear equalizer. The DFE includes two filters, a forward filter 200 and a backward filter. 210).

The output value z (k) of the forward filter unit 200 and the output value of the reverse filter unit 210 are added and subtracted by the first adder 260 and input to the third adder 220. In this case, the operation result of the noise prediction filter unit 250 is input to the third adder 220 and subtracted according to the present invention. Then, the operation result of the third subtractor 220 is input to the symbol determiner 230.

In the symbol determiner 230, a non-integer value obtained by passing through the linear traverse filters, that is, the forward filter unit 200, the reverse filter unit 210, and the noise prediction filter unit 250 is 0 or 0. Determining 1 will affect incoming symbols later.

Meanwhile, the output value D (k) of the symbol determiner 230 is input to the second adder 240. In this case, the second adder 240 subtracts the output value of the symbol determiner 230 from the output value z (k) of the first adder 260 and inputs the calculated value to the noise prediction filter unit 250. do.

As described above, in general communication, signals cause intersymbol interference with multipath fading, and the received distorted symbols pass through a forward filter. In this case, the coefficient of the filter is improved by applying a weighting factor to the difference between the previous symbol and the determined value by an adaptive least mean square (LMS). Accordingly, a decision value of 0 or 1 is determined by providing a value passing through the forward and reverse filters as an input to a decision device. In the conventional DFE algorithm, only the inter-symbol interference was removed by setting the decision value as a symbol. However, in the present invention, the noise prediction filter unit 250 as described above is added to prevent the wrong symbol determination due to noise, AWGN. Will be.

The signal processing method in the noise prediction filter unit 250 added according to the present invention is as follows. First, referring to the variables used for the signal processing, C _F (k) and C _B (k) denote the tap coefficient matrix of the forward filter unit 200 and the reverse filter unit 210, respectively, and y _F ( k) and y _B (k) represent the input signal matrices of the forward filter unit 200 and the reverse filter unit 210, respectively. As a result, the output signal z (t) and the error signal e (k) of the DFE may be defined as in Equations 1 to 4 below.

In this case, the result value o (k) calculated by the noise prediction filter unit 250 according to the present invention may be defined as in Equation 5 below.

In Equation 5, λ is a forgetting factor of the noise prediction filter 250, and N (k) is an input value of the noise prediction filter 250. In addition, o (k) represents an output value of the noise prediction filter unit 250.

On the other hand, the N (k) value is calculated as shown in Equation 6 below.

In Equation 6, D (k) means an output value of the symbol determiner 230 as described above. In this case, the value N _D (k) input to the symbol determiner 230 reflecting the result value of the noise prediction filter unit 250 according to the present invention may be defined as in Equation 7 below.

Therefore, when the DFE including the noise prediction filter unit 250 is implemented according to the present invention as defined by the above-described equations, it may be implemented as shown in FIG. 2.

Although not shown in FIG. 2, a preamble generating unit may be added. The preamble generator generates a preamble or training sequence of a length already determined by the preamble generator in order to give reliability to the decision value initially fed back.

3 illustrates a linear cross filter of a crystal feedback equalizer according to an embodiment of the present invention. Referring to FIG. 3, the linear cross filters used in FIG. 2 (ie, the forward filter unit 200 and the reverse filter unit 210) may include a plurality of delayers 301, 302, 303, and 304. The multipliers 311, 312, 313, 314, and 315, the adder 320, and the tap coefficient updater 330 may be configured.

In other words, the linear traverse filter has a basic linear equalizer structure. In more detail, the unequalized input signal is delayed in each of the delayers 301, 302, 303, and 304, and the input signal and each delayed signal are delayed in the tap coefficient updater 330. The provided tap coefficients are multiplied by the plurality of multipliers 311, 312, 313, 314, 315. Then, the output signals of the respective multipliers are summed in the summer 320.

Meanwhile, the tap coefficient updater 330 updates the tap coefficient by reflecting the noise predicted result value according to the present invention.

Hereinafter, a specific implementation example of the decision feedback equalizer including the noise prediction filter 250 according to the present invention described with reference to FIG. 4 will be described.

4 is a diagram illustrating a detailed structure of a crystal feedback equalizer according to an embodiment of the present invention. Referring to FIG. 4, according to an embodiment of the present invention, a structure of a DFE to which a noise prediction function is added includes a forward filter 200 and a backward filter including a linear transversal filter. 210, a symbol determiner 230, a noise prediction filter 250, and a plurality of adders 420, 430, 440, and 450.

In more detail, the signal x (k) received by the DFE is first input to the forward filter 200. The signal input to the forward filter 200 is delayed through a plurality of delayers 401, 402, 403, 404, and the received signal x (k) and the respective delayers 401, 402, 403, 404 The multiplied signal is multiplied by the tap coefficients provided by the forward tap coefficients updater 410 and the multipliers 405, 406, 407, 408, and 409, respectively.

The output signals of the multipliers 405, 406, 407, 408, 409 are then summed in a first adder 420 and from each of the multipliers 415, 416, 417, 418 of the reverse filter 210. The output signal is subtracted. The value z (k) added and subtracted by the first adder 420 is input to the third adder 440 and the fourth adder 450.

The third adder 440 subtracts an output value o (k) of the noise prediction filter unit 250 added according to the present invention to the output signal z (k) from the first adder 420 to determine a symbol. Input to the unit 230. At this time, the value input to the symbol determiner 230 is N _D (k) as calculated by Equation (7).

The symbol determiner 140 determines a value of 0 or 1 from the input value. In this case, the reverse filter 210 receives the value D (k) determined by the symbol determiner 230 and inputs the delays 411, 412, 413, and 414 to delay. The signal input to the reverse filter 210 is delayed through the plurality of delayers 411, 412, 413, and 414, and the received signal and the signal passing through the delayers are updated with reverse tap coefficients. The tap coefficients provided in section 419 and multipliers 415, 416, 417, 418 are respectively multiplied. In this case, the signals multiplied by the multipliers 415, 416, 417, and 418 are input to the first adder 420 and subtracted as described above.

Meanwhile, according to the present invention, the second adder 430 subtracts the output signal D (k) of the symbol determiner 230 from the output signal z (k) of the first adder 420 to the noise. The input is input to the prediction filter unit 250. At this time, the value input to the noise prediction filter unit 250 is N (k) and is calculated as shown in Equation 6 above.

As described above, the noise prediction filter unit 250 calculates a noise prediction value o (k) by the calculation of Equation 5. The calculated output value o (k) sms of the noise prediction filter unit 250 is input to the third adder 440 again.

At this time, the third adder 440 determines the symbol by subtracting the output value o (k) of the noise prediction filter unit 250 from the output value z (k) of the first adder 420 as described above. Provided to the unit 230.

Meanwhile, as in the general DFE, an error value is input to each linear prediction filter by comparing the output value of the first adder 420 with the output value of the symbol determiner 230. That is, in the fourth adder 450, the output signal z (k) of the first adder 420 is subtracted from the output signal D (k) of the symbol determiner 230 so that the forward filter unit 200 and The tap coefficient updating units 410 and 419 of the reverse filter unit 210 are provided. In this case, the output value of the fourth adder 450 is calculated as Equation 4 as e.

As described above, in the noise prediction filter unit 250 added according to the present invention, the noise predicted value is used as the output value of each linear traverse filter and the output value of the symbol determiner 230 for symbol determination to remove the noise more effectively. It is possible to provide a determined feedback feedback equalizer.

In the above detailed description of the present invention, specific embodiments have been described, but various modifications are possible without departing from the scope of the present invention. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be defined not only by the scope of the following claims, but also by the equivalents of the claims.

As described above, the present invention has the advantage of lowering the probability of erroneous determination of a symbol due to AWGN as well as removing intersymbol interference, which is a function of a basic equalizer, by adding a noise prediction algorithm.

Claims (4)

In the crystal feedback equalizer comprising a forward filter part and a reverse filter part, A first adder configured to subtract the output value of the reverse filter unit to the output value of the forward filter unit; A second adder configured to subtract the output value of the symbol determiner from the output value of the first adder; A noise prediction filter unit receiving an output value of the second adder and outputting a prediction value for noise; And a third adder which subtracts an output value of the noise prediction filter unit to an output value of the first adder and provides the calculated value to the symbol determiner. The method of claim 1, wherein the noise prediction filter unit, And calculating a noise prediction value by averaging a current input value and a previously input value from the second adder. The equalizer of claim 2, wherein the noise prediction filter unit calculates a result by the following equation.

In this case, λ is an oblivion element of the noise prediction filter unit, and N (k) represents an input value of the noise prediction filter unit. The method of claim 1, wherein the equalizer, And a fourth adder which subtracts an output value of the symbol determiner to an output value of the first adder and provides the fourth adder to each of the forward filter unit and the reverse filter unit.

Images