JPH10198913A - Decision feedback equalization method and decision feedback equalizer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for signal processing and a device therefor that can remove propagation of error of Decision Feedback Equalization(DFE) and secure the reliability under a condition of a low resolution. SOLUTION: An error detector 700 having a threshold value of ±(b0 +1)/2 determined by an input side tap factor b0 of FBF besides an ordinary decision unit 106 having a threshold value is used for checking a differential value zk between an output value of Feed Forward Filter(FFF) of DFE 702 and that of Feedback Filter(FBF) on a basis of a decision made plural time ago, and if the differential value zk is not less than (b0 +1)/2 or not more than -(b0 +1)/2, then the output signal is regarded as erroneous. And, digital values in a register 701 of DBF are changed by a register alteration means 704, for example, they are all reset to zero.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a signal processing method and apparatus for recording / reproducing digital data at high density on a magnetic disk drive, and more particularly to preventing error propagation in a decision feedback equalizer.

[0002]

2. Description of the Related Art Decision feedback equalizer (Decision Feedback Eq.) Is one of signal processing methods in a reproducing unit of a magnetic recording / reproducing system.
ualization (hereinafter referred to as DFE). This method has been conventionally used mainly in communication systems, but has recently been applied to a signal processing unit of a magnetic disk.

FIG. 1 is a diagram for explaining the outline of the DFE. The data sequence 100 is written to the magnetic recording medium 101 via the head 102. The DFE starts with the head 10
2, a data series 100 including noise on the recording medium 101 is read. The read signal is passed through a feedforward filter (hereinafter, referred to as FFF) 103 to provide a pre-stage equalization characteristic. This is to provide the read signal with appropriate frequency and phase characteristics suitable for the magnetic recording / reproducing system according to the frequency and phase of the read signal. On the other hand, the value obtained as the determination result a plurality of times ago is used as the input value and
The post-stage equalization is performed by an edback filter (hereinafter, referred to as FBF) 113, and the subtractor 104 calculates the difference between the two. Then, this value is determined by a decision unit 1 having a threshold value ± T.
06, it is determined to be 0, ± 1, and the ternary value (0, ± 1) is 2
A decoded sequence 111 is obtained via a conversion circuit 110 for converting the value into a value (0, 1).

FIG. 2 shows the FFF 103 and the FBF 1 shown in FIG.
13 shows a form of a transversal filter used in FIG. The transversal filter includes a shift register 200 and a tap coefficient unit 201, and the input sequence i _k
The output sequence j _k is determined by multiplying (the number of input sequences i _k = the number of registers in the shift register +1) by the tap coefficient C _k of the tap coefficient unit 201 and adding them. FF
In F, the tap coefficients C _k (k = 0, 1,..., M−1) are determined so that only the first part of the input waveform is equalized to 0 by the transversal filter, and the interference of the second part remains as it is. In the FBF, an interference component is calculated from a plurality of previous identification results corresponding to the interference component of the FFF, and is subtracted from the equalized value of the FFF, thereby performing equalization to eliminate the interference component at a later stage of the waveform. Also at this time, the tap coefficients need to be determined in advance similarly to the FFF.

FIGS. 3 and 4 show transversal filters in which the number of taps of FFF and FBF is 7 and 10, respectively. Transversal filter FFF shown in Figure 3, includes a tap coefficient 301 of the shift register 300 and the tap number 7 number register is 6, the tap coefficients of the tap coefficient 301 to the input sequence l _k f _{k (k} = 0,1, ...,
The output sequence x _k is determined by multiplying by 6) and adding. Further, the transversal filter FBF shown in FIG. 4, the number of taps and the shift register 400 in the number of registers 9 comprises a tap coefficients 401 of 10, the input sequence d _k
To the tap coefficient b _k (k = 0, 1,
, 9) are added to determine the output sequence y _k .

Here, as an example when these filters are used, an input signal I =. . . 0001000. . . FIG. 5 shows an equalized waveform corresponding to. However, assuming a noise-free situation, tap coefficients of FFF and FBF use values determined so as to minimize the equalization error. At this time, when the isolated reproduction waveform A corresponding to the input signal is input, the FF is output.
The equalized waveform obtained from F becomes like B after a certain time. On the other hand, the equalized output waveform of the FBF becomes C, and a difference equalized waveform D is obtained by taking the difference between the two. When the difference equalized waveform D passes through the decision unit 106 having a threshold, the decoding result I =. . . 0001000. . . Is output.

FIG. 6 shows equalized waveforms A, B, C, and D when the input signal I is a random input sequence. In the figure, a waveform A is a waveform obtained by superimposing an isolated waveform on an input value I, a waveform B is a waveform obtained by passing through an FFF,
A waveform C represents a waveform equalized by the FBF, and a waveform D represents a waveform obtained by calculating a difference between the waveform B and the waveform C. If this waveform is determined to be 0 or ± 1 by the determiner 106 and −1 → 1 is converted by the circuit 110 for converting a ternary value into a binary value, the input value I completely matches. In this case, the response waveforms described in FIG. 5 are linearly added.

[0008] In order to calculate the linear addition of interference by the FBF, a ROM that outputs the interference value calculated in advance by using the determination result of a plurality of bits as an input address is used. In other words, the DFE is a method in which the interference component expected by the FBF part is calculated from the determination result, and the signal component is faithfully extracted by subtracting from the transmission signal passing through the FFF part.

[0009]

Since the DFE is a system in which an interference component is subtracted by the FBF, a signal component can be faithfully extracted. However, in the DFE, once an error occurs in the determination result due to noise or the like, the error propagates because an incorrect interference is subtracted. This is called error propagation, and this long lasting propagation is
This is a big problem in using.

The present invention has been made in view of such a problem of DEF, and it is an object of the present invention to use DFE as a signal processing technique for realizing high-density recording and to realize signal processing without error propagation. Aim.

[0011]

According to the present invention, the DFE F
FBF based on the output value of the FF and the judgment result a plurality of times ago
Using an error detector having a threshold value dependent on the tap coefficient b ₀ closest to the input side of the tap coefficient device of the FBF, in addition to a normal decision unit having a threshold value, The above object is attained by preventing the error propagation by changing the digital value in the register of the FBF assuming that a determination error has occurred when the value exceeds the threshold value.

That is, the present invention provides a method for determining a difference value between a pre-stage equalized signal obtained through a feedforward filter and a post-stage equalized signal obtained by passing a value determined by a determiner through a feedback filter. In a decision feedback equalization method for decoding a digital recording signal by inputting the difference to a signal, a decision error is detected when the difference value exceeds a predetermined threshold value.

When the tap coefficient closest to the input side of the tap coefficient unit of the feedback filter is b ₀ , the threshold value is set between ± b ₀ and ± 1 (in the same sign). For example, if the threshold is set to ± (b ₀ +1) / 2,
The difference value is equal to or more than (b ₀ +1) / 2 or − (b ₀ +
If it is 1) / 2 or less, it is regarded as a judgment error. Tap coefficient b
₀ generally changes according to the normalized linear density of the recording medium. Therefore, the fact that the threshold value for detecting a determination error is set according to the tap coefficient b ₀ means that the threshold value is set according to the recording density of the recording medium. When the recording medium is disk-shaped and the recording density differs for each recording zone set in the radial direction of the disk, the threshold is set according to the recording zone.

When a determination error is detected, the digital value in the register of the feedback filter is changed to a predetermined value,
For example, by resetting to zero, error propagation can be prevented. Alternatively, there is an effect that error propagation can be prevented by setting the digital value in the register at the input terminal of the feedback filter to zero. This is because there is a time lag between the input values of FFF and FBF,
Since the input value to the FBF can be forcibly changed after a determination error has occurred and the input pattern is random, there is a pattern in which the error always converges at a certain time by resetting. That's why.

When a decision error is detected, the digital information in the register of the feedback filter is replaced with a decision result based on the equalization characteristic not including the feedback system, for example, a partial response and a decoded sequence obtained by using the maximum likelihood decoding method. This also eliminates a determination error and prevents error propagation. Further, the present invention provides a feed-forward filter for providing a pre-stage equalization characteristic to a signal read from a recording medium, a determiner, a feedback filter for providing a post-stage equalization characteristic to a determination result by the determiner, and a feed-forward filter. A decision feedback equalizer including a subtractor that supplies a difference value between an output and an output of a feedback filter to a decision unit, including a decision error detector that detects a decision error when the difference value exceeds a predetermined threshold. It is characterized by the following.

The decision feedback equalizer of the present invention includes register changing means for changing a digital value in a register of the feedback filter when a decision error detector detects a decision error. This register changing means may change all values in the register of the feedback filter to zero, change only the digital value in the register of the input terminal to zero, or convert the digital value in the register of the feedback filter into a partial response and The value may be changed to a value obtained by using the maximum likelihood decoding method.

The decision feedback equalizer of the present invention can be provided in a magnetic storage device. The digital magnetic recording apparatus according to the present invention is suitable for high-density recording and can achieve a large capacity as compared with the case where a conventionally used PRML circuit is used.

[0018]

Embodiments of the present invention will be described below with reference to the drawings. First, the principle of the present invention will be described in detail. First, the feedforward filter FF
The input sample value of F is 1 _k , the tap coefficient of the FFF is f _i (i = 0, 1,..., M−1), and the tap coefficient of the feedback filter FBF is b _j (j = 0, 1,. −
1). At this time, the output sample value x _k of the FFF is represented by the following [Equation 1].

[0019]

[Number 1] _{x k = Σ i = 0,} m-1 f i · l ki Further, the output sample value y _k of the FBF is represented by the following expression (2).

[0020]

Y _k = Σ _{j = 0, n-1} b _j · d _kj At this time, the difference value of [ _Equation 1] and [Equation 2] is _represented by z _k = x _k −y
If it is defined by _k , it is z _k {-1, 0, 1}, and the sample value after determination is d _k = {-1, 0, 1}. However,
The relationship between z _k and d _k is as shown in the following [Equation 3].

[0021]

D _k = −1 (z _k −T), d _k = 0 (−T ≦ z _k ≦ T), d _k = 1 (z _k > T) In the above [Equation 3], T is This is a threshold value of the determiner 106. Here, if incorrectly determined d _k Assuming now d _k ', it can be considered divided case of the following [A] [B] [C] [D]. However, when d _k = a is incorrect as d _k '= b,
(D _k , d _k ′) = (a, b). [A] When (d _k , d _k ′) = (0, 1) or (d _k , d _k ′) = (− 1, 0) d _k ′ = d _k +1 and y _k ′ = b ₀ d _k '+ b ₁ d _k-1 + ...
From + b _n-1 d _{k-n + 1} , y _k ′ is expressed as the following [Equation 4].

[0022]

Y _k ′ = b ₀ (d _k +1) + b ₁ d _k-1 +... + B _n-1 d _{k-n + 1} = b ₀ + b ₀ d _k + b ₁ d _k-1 + · + B _n-1 d _{k-n + 1} = b ₀ + Σ _{j = 0, n-1} b _j d _kj = b ₀ + y _k Therefore, from [Equation 4], the following [Equation 5] holds.

[0023]

(5) z _k ′ = x _k −y _k ′ = x _k −b ₀ −y _k = −b ₀ + z _k [B] (d _k , d _k ′) = (1, 0) or (d _{k , d k ') = (0} , if the _{-1) d k' = d k} -1 _{and,, y k '= b 0} d k' + b 1 d k-1 + ··
From + b _n−1 d _{k−n + 1} , y _k ′ is expressed as the following [Equation 6].

[0024]

Y _k ′ = b ₀ (d _k −1) + b ₁ d _k−1 +... + B _n−1 d _{k−n + 1} = −b ₀ + b ₀ d _k + b ₁ d _k−1 + ... + b _n-1 d _{k-n + 1} = −b ₀ + Σ _{j = 0, n−1} b _j · d _kj = −b ₀ + y _k Therefore, from [Equation 6], the following [Equation 7] Holds.

[0025]

(7) z _k ′ = x _k −y _k ′ = x _k + b ₀ −y _k = b ₀ + z _k [C] In the case of (dk, dk ′) = (− 1, 1) d _k ′ = d _k + 2 and y _k ′ = b ₀ d _k ′ + b ₁ d _k−1 +.
From + b _n−1 d _{k−n + 1} , y _k ′ is expressed as the following [Equation 8].

[0026]

Y _k ′ = b ₀ (d _k +2) + b ₁ d _k-1 +... + B _n-1 d _{k-n + 1} = 2b ₀ + b ₀ d _k + b ₁ d _k-1 + · + B _n-1 d _{k-n + 1} = 2b ₀ + Σ _{j = 0, n-1} b _j d _kj = 2b ₀ + y _k Therefore, from [Equation 8], the following [Equation 9] holds.

[0027]

In the case of z _k ′ = x _k −y _k ′ = x _k −2b ₀ −y _k = −2b ₀ + z _k [D] (d _k , d _k ′) = (1, −1) d _k '= d _k -2, _{and, y k' = b 0 d} k '+ b 1 d k-1 + ··
From + b _n-1 d _{k-n + 1} , y _k ′ is expressed as the following [Equation 10].

[0028]

Y _k ′ = b ₀ (d _k −2) + b ₁ d _k−1 +... + B _n−1 d _{k−n + 1} = −2b ₀ + b ₀ d _k + b ₁ d _k−1 + ... + b _n-1 d _{k-n + 1} = -2b ₀ + Σ _{j = 0, n-1} b _j · d _kj = -2b ₀ + y _k Therefore, from [Equation 10], the following [Equation 11] Holds.

[0029]

Z _k ′ = x _k −y _k ′ = x _k + 2b ₀ −y _k = 2b ₀ + z _{k From the} above-mentioned [Equation 5], [Equation 7], [Equation 9] and [Equation 11],
In each of the cases [A], [B], [C], and [D], it can be seen that the difference value at the next time when the determination is incorrect depends on the term of the tap coefficient b ₀ closest to the input side of the FBF.

According to the above principle, when there is no error, z _k ≒
Since {1, 0, -1}, fluctuations due to noise are also taken into consideration, and in the present invention, a threshold τ, for example, ± (b ₀ +1) / 2, is set between ± b ₀ and ± 1 (in the same order as the sign). = ± τ, a determination error is detected. FIG. 7 is an explanatory diagram of an example of the DFE according to the present invention. This DEF 702 is an FFF 10
0, an error detector 700 and a register changing unit 704 in addition to the subtractor 104, the determiner 106, and the FBF 113. A reproduction signal from a magnetic disk (not shown) is shown in FIG.
Is input to the FFF 100 in the same manner as described above. FFF100 and F
The difference value z _{k of the} BF 113 is input to an error detector 700 having a threshold τ. Then, the error detector 700 compares the difference value z _k ′ after the determination error has occurred with z _k ′> (b ₀
If it is detected that +1) / 2 or z _k ′ <− (b ₀ +1) / 2, an error flag 703 is generated on the assumption that the decoding result is incorrect. Register changing means 70
4 receives the error flag 703 from the error detector 700, changes the digital information in the register unit 701 of the FBF, and resets it to, for example, all 0s. This makes it possible to restart decoding correctly after a certain period of time. The register changing unit 704 may change only the digital value in the register at the input end of the register unit 701 to zero.

Here, as an example, the normalized linear density K =
Consider the case of 3.0. At this time, the number of taps of the FBF is set to 1
When the value of the tap coefficient b ₀ closest to the input side of the FBF was determined so as to minimize the equalization error by setting it to ₀ , b ₀ =
1.9007. Therefore, the threshold τ is 1 and 1.9007
For example, τ ≒ 1.5 is set. Here, z _k ′ in the combination of (z _k−1 , z _k ) due to erroneous d _k
8 are shown in FIG. However, since z _k−1 and z _k are true values, the values were determined assuming that z _k ≒ d _k . For example, [A]
In the above case, d _k = 0 and d _k ′ = 1. At this time, z _k
= 0, then from equation (5), z _k ′ = −b ₀ + z _k
= −1.9007 + 0 = −1.9007 (* mark in FIG. 8). From the table in FIG. 8, all of the patterns except for the four patterns in which the numerical values are circled satisfy | z _k ′ |> 1.5. Furthermore, since it has been confirmed that these four patterns always exceed the threshold value τ after a few times, there is no problem in error determination.

The tap coefficients of the FFF and the FBF including the tap coefficient b ₀ should be set to appropriate values depending on the recording density, more specifically, the normalized linear density. Table 1 below shows the normalized linear density K, and the value of the tap coefficient b ₀ closest to the input side of the FBF obtained so as to minimize the equalization error with respect to the normalized linear density K. I have. As is clear from Table 1, the value of the tap coefficient b ₀ increases as the recording density increases. Therefore, it is necessary to change the setting of the threshold value related to the tap coefficient b ₀ depending on the recording density and the magnetic disk zone.

[0033]

[Table 1] FIG. 9 is an explanatory diagram showing another example of the DFE according to the present invention. In this example, in addition to the DFE including the FFF 100, the subtractor 104, the determiner 106, and the FBF 113, equalizing means not including a feedback system, such as a partial response class 4 (PR4) 901, a clock 902, and a maximum likelihood decoding circuit 903 Equalization means to be used are provided.

In parallel with the output of the DFE, the output result of the PRML composed of the partial response class 4 (PR4) 901, the clock 902 and the maximum likelihood decoding circuit 903 is stored in a register 9 having the same register length as the FBF 113.
04 temporarily. Then, when the error detector 700 detects that the difference value z _k has exceeded the threshold τ, and the error detector 700 generates an error flag 905,
The register change unit 907 supplies the determination result stored in the register 904 to a circuit 906 for converting a binary value to a ternary value, converts the result into ternary data, and replaces the ternary data with the digital information in the register 701 of the FBF 113. ,
Prevent DFE error propagation. The delay circuit 900 is provided to match the timing at which the contents of the register of the FBF 113 are replaced with the contents of the register 904 from the determiner 106.

[0035]

According to the present invention, when the DFE is used as a signal processing means for realizing high-density recording, if it is assumed that the determination is wrong at a certain time k, the output of the FFF at the next time k + 1 and the output of the FFF at the time k are obtained. By setting a threshold value for the difference value of the output of the FBF and detecting an error, and changing the digital value in the register of the FBF at that time, error propagation can be prevented.

[Brief description of the drawings]

FIG. 1 is a signal system diagram of a recording / reproducing system including a decision feedback equalization system.

FIG. 2 is a diagram of a transversal filter in which the number of taps is m.

FIG. 3 is a diagram of an FFF transversal filter having seven taps.

FIG. 4 is a diagram of an FBF transversal filter having ten taps.

FIG. 5 is a diagram of a signal waveform of an isolated pattern after passing through an FFF and an FBF, and a signal waveform obtained by taking a difference between the signals.

FIG. 6 is a diagram of a signal waveform after passing through an FFF and an FBF with respect to a random pattern and a signal waveform obtained by taking a difference between them.

FIG. 7 is an explanatory diagram of an example of a DFE according to the present invention.

FIG. 8 is a diagram showing a difference value at a time next to a time when a determination error has occurred.

FIG. 9 is an explanatory diagram of another example of the DFE according to the present invention.

[Explanation of symbols]

101: recording medium, 102: magnetic recording head, 103 ...
Feed forward filter (FFF), 104: subtractor, 106: determiner, 110: circuit for converting a ternary value into a binary value, 113: feedback filter (FBF), 20
0, 300, 400 ... shift register, 201, 30
1,401: tap coefficient unit, 700: error detector, 70
REFERENCE SIGNS LIST 1 register, 702 DFE, 703 error flag, 704 register changing means, 900 delay circuit, 9
01: PR4 equalizer, 902: clock, 903: maximum likelihood decoding circuit, 904: register, 906: circuit for converting binary to ternary, 907: register changing means

Claims (12)

[Claims] 1. A difference value between a first-stage equalized signal obtained through a feedforward filter and a second-stage equalized signal obtained by passing a value determined by a determiner through a feedback filter is input to the determiner. A decision feedback equalization method for decoding a digital recording signal by detecting a decision error when the difference value exceeds a predetermined threshold value. 2. The threshold value represents a tap coefficient closest to an input side of a tap coefficient unit of the feedback filter as b _0.
The decision feedback equalization method according to claim 1, wherein the value is set between ± b ₀ and ± 1 (same order of sign). 3. The method according to claim 1, wherein the threshold is set according to a recording density of a recording medium. 4. The method according to claim 1, wherein the threshold value is set according to a recording zone provided in a radial direction of the recording medium. 5. The decision feedback system according to claim 1, wherein a digital value in a register of said feedback filter is reset to a predetermined value when said decision error is detected. Method. 6. The decision feedback or the like according to claim 1, wherein when the decision error is detected, the digital value in the register at the input terminal of the feedback filter is set to zero. Method. 7. The method according to claim 1, wherein when the decision error is detected, digital information in a register of the feedback filter is replaced with a partial response and a decoded sequence obtained by using a maximum likelihood decoding method. 5. The decision feedback equalization method according to any one of 4. 8. A feed-forward filter for providing a pre-equalization characteristic to a signal read from a recording medium, a determiner, a feedback filter for providing a post-equalization characteristic to a result determined by the determiner, and the feed-forward filter And a subtractor that supplies a difference value between the output of the feedback filter and the output of the feedback filter to the determiner, wherein a determination error is detected when the difference value exceeds a predetermined threshold value. A decision feedback equalizer characterized by comprising a device. 9. The decision feedback equalizer according to claim 8, further comprising register changing means for changing a digital value in a register of said feedback filter when said decision error detector detects a decision error. 10. The decision feedback equalizer according to claim 9, wherein said register changing means sets a digital value in a register at an input terminal of said feedback filter to zero. 11. The decision feedback system according to claim 9, wherein said register changing means changes a digital value in a register of said feedback filter to a value obtained by using a partial response and a maximum likelihood decoding method. Equalizer. 12. A magnetic storage device comprising the decision feedback equalizer according to claim 8. Description: