JP3471245B2 - Information reproducing device using feedback filter - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information reproducing apparatus for reproducing digital information recorded on a magnetic disk, a magnetic tape, an optical disk, etc., and more particularly to an information reproducing apparatus characterized by data detection.

[0002]

2. Description of the Related Art In a digital recording / reproducing apparatus such as a magnetic disk device or an optical disk, a recording data reproducing method is
P that combines the partial response method and the maximum likelihood detection method
The RML (Partial Response Maximum Likelihood) method is used. Further, in order to cope with the recent increase in recording density, a method has been proposed in magnetic recording in which a modulation method capable of widening the minimum inter-code distance is combined with the EEPR4 method which is a higher-order PRML method (Jaekun Moo).
n and Barrett Brickner, "Maximum TransitionRun Cod
es for Data Storage Systems ", IEEE Transactions on
Magnetics, vol.32, no.5, Sept. 1996).

In addition to the PRML system, a RAM-DFE (Dec
ision Feedback Equalizer) method, FDTS / DF (Fi
The xedDelay Tree Search / Decision Feedback) method is also under consideration.

The FDTS / DF method limits the path length to the maximum likelihood detection method in which the time until data identification is indefinite, and selects the path with the smallest path metric with the limited path length. This is a method for achieving the above-mentioned high recording density.

FIG. 11 is a block diagram showing the structure of an apparatus based on the FDTS / DF method.

The reproduced waveform is the feedforward filter 1
After passing through 101, it is set to ai (i is the time) and then to yi via the adder and input to the FDTS detector 1111. The output d _i- γ ₊₁ of the FDTS detector 1111 is used as detection data, and is simultaneously input to the delay element 1107 ₁ . A plurality of delay elements are provided, and 1107
₁ , 1107 ₂ , ..., 1107 _n are serially connected,
The outputs d _i- γ, d _i- γ _-1 , ... Of the n delay elements respectively
d _i- γ _{-n + 1} is i-γ to i-γ- with respect to the current time i.
It is output to the feedback filter (FB) 6 as a data string D corresponding to each state at time n + 1 .

The output bi of the feedback filter 1106 is input to the adder and added to the output ai of the feedforward filter 1101 to be yi.

In this conventional example, the FDTS detector 1111 obtains and outputs detection data delayed by the time point γ-1 (γ is the tree depth).

Each of FIGS. 12 and 13 shows an FDT.
It is a figure which shows the determination method and impulse response of S.

The impulse response is a response to the recording data "1". In magnetic recording, a reproduced waveform corresponds to two consecutive isolated waves with inverted polarities, that is, a dipulse. 12 and 13 are examples of tree depth γ = 2. As shown in FIG. 12, at time i, branch metrics and path metrics for four paths (trees) are obtained and compared, and the path that gives the smallest path metric is determined. Then, i-γ in that path
The data at +1 time point is output as the determined determination data.

The detected data string before the time point i-γ + 1 obtained as described above is fed back to the feedback filter 110.
As an input value for 6, waveform compensation at the next i + 1 time point is performed.

FIG. 13 is an example of an impulse response waveform. The feedforward filter 1101 sets the output value at the discrimination point to “1” and equalizes the output values before that to “0”. Further, from the point of being delayed by γ bits from the discrimination point, the feedback filter 1106 equalizes to “0”. As a result, the equalized waveform is output values (1 and a) of the remaining γ bits (here, 2 bits).
The metric calculation and the data detection are performed according to the determination method shown in FIG.

[0013]

In the FDTS / DF method, the FDTS method is used for data detection. Since this method limits the path length and obtains the path metric for a limited path, the error rate characteristic may be worse than that of the maximum likelihood detection method.

Further, when the maximum likelihood detection method is used, the time until the data is identified is indefinite, and the detection data cannot be obtained within the limited delay, so that the limited time is required. The feedback loop cannot be operated within (clock).

The present invention has been made in view of the problems of the above-mentioned conventional technique, and it is possible to operate the feedback loop with a predetermined delay amount while using the maximum likelihood detection method, and An object is to realize an information reproducing apparatus using a highly accurate feedback filter with improved rate.

[0016]

An information reproducing apparatus using a feedback filter according to the present invention equalizes a leading edge portion of an impulse response of a reproduced waveform and sets an output value at a discrimination point to "1". A feedforward filter that sets the previous output value to “0”, a feedback filter that compensates for the trailing edge after the time when the tree depth is delayed by γ bits from the discrimination point, and the feedforward filter output and the feedback filter output. The equalization error E for each equalization reference level value, the branch metric for each branch, and the path metric based on the state transition diagram derived from the tree depth γ and the waveform output of the impulse response from the equalized waveforms in which Then, the path metrics for each state are compared, and the path with the smallest path metric is calculated for that state. The metric calculator selected as the remaining path and the survivor path selected by the metric calculator are input, and i-γ to i
A path memory in which n pieces of data up to the point [ gamma] -n + 1 output the value of the final stage as a data string D and detection data corresponding to each state, and a survivor path to each state obtained by the metric calculator. A comparator for finding the path having the minimum path metric from among the above, and the data string D output from the path memory are input, and the data string corresponding to the path giving the minimum path metric found by the comparator is input. A selector for selecting and outputting to the feedback filter, wherein the feedback filter calculates a compensation amount for equalization based on the data string selected by the selector.

In this case, the path memory length may be shorter than the number of input data of the feedback filter, and a delay element for holding the detection data output by the path memory and outputting it to the feedback filter may be provided.

In any of the above cases, the selector selects the equalization error corresponding to the path having the minimum path metric from the equalization errors corresponding to the respective surviving paths, and selects the selected equalization error. It may be used to perform adaptive equalization of the feedforward filter and the feedback filter.

Further, the feedback filter and the metric calculator set the tree depth γ to 2, and the metric calculator has a two-state state transition diagram for the reproduced waveform in which the feedforward filter output and the feedback filter output are superposed. The maximum likelihood detection may be performed by associating with.

Further, the feedback filter and the metric calculator have a tree depth γ of 3, and the metric calculator has a state transition of four states with respect to the reproduced waveform in which the feedforward filter output and the feedback filter output are superposed. Maximum likelihood detection may be performed in correspondence with the figures.

Further, the feedback filter and the metric calculator have a tree depth γ of 4, and the metric calculator has a state transition of eight states with respect to the reproduced waveform in which the feedforward filter output and the feedback filter output are superposed. Maximum likelihood detection may be performed in correspondence with the figures.

In the information reproducing method using the feedback filter of the present invention, the leading edge of the impulse response of the reproduced waveform is equalized, the output value at the discrimination point is set to "1", and the output value before that is set to "1". 0 "and the first step of compensating the trailing edge after the time point of delaying the tree depth γ bits from the discrimination point to obtain an equalized waveform, and the tree depth γ and the impulse response from the equalized waveform. Based on the state transition diagram derived from the waveform output of E, the equalization error E for each equalization reference level value, the branch metric for each branch, and
The second step of obtaining the path metric, comparing the path metrics to each state, and selecting the path having the smallest path metric as the surviving path to that state, and the surviving path selected in the second step. On the basis of,
The third step of generating n data strings D corresponding to respective states from the time point i-γ to the time point i-γ-n + 1 with respect to the current time point i, the detection data being the data at the final stage; A fourth step of obtaining a path having a minimum path metric from among the surviving paths to each state obtained in the second step, and n generated in the third step.
Based on the data string D, a fifth step of selecting a data string corresponding to the path giving the minimum path metric obtained in the fourth step, and based on the data string selected by the selector. And a sixth step of obtaining a compensation amount for the equalization performed in the first step.

In the "action" FDTS detector, the path metric calculation is aborted up to the tree depth γ to obtain the minimum path metric, but in the present invention, maximum likelihood detection is performed as in the PRML method.

In the present invention, first, the branch metric and the path metric are calculated in the metric calculator based on the state transition diagram obtained from the γ value and the impulse response, and the surviving path for each state is determined. However, the judgment standard level is general PRM
Unlike the L method, it is not an integer value but a real value and can be arbitrarily given. Path memory is a general PR
It has the same configuration as the ML system and can perform maximum likelihood detection.

A feedforward filter and a feedback filter are used to perform waveform equalization. FD
Similar to the TS / DF method, the leading edge of the impulse response is equalized by the feedforward filter, and the output value at the discrimination point is set to "1" and the output value before that is set to "0". On the other hand, the feedback filter compensates for the trailing edge after the time point that is delayed by γ bits from the discrimination point.

The above waveform equalizing operation is the same as that of the FDTS / DF method. The input value to the feedback filter in the present invention is not the FDTS detection result, but the detection value corresponding to the most probable path at the present time from the plurality of detection data strings recorded in the path memory is used as the temporary detection data string. This tentative detection data string is a detection data string corresponding to the path when the path metric of the surviving path corresponding to each state is compared and the path giving the minimum path metric is obtained.

Summarizing the above operations, the state transition diagram which is the basis of maximum likelihood detection is determined by the metric calculator from the γ value and the impulse response waveform and recorded in the path memory. The comparator determines the minimum value of the path metric corresponding to the plurality of surviving paths recorded in the path memory. The selector selects the most probable data string at the present time (the data string corresponding to the path having the smallest path metric) from the data string D of the path memory as the temporary detection data according to the minimum value of the path metric obtained by the comparator. The selected temporary detection data is input to the feedback filter. The feedback and feedforward filters are used to compensate for the leading and trailing edges of the impulse response.

As described above, in the present invention, the FDT
Since the maximum likelihood detection method (Viterbi detection method) is used in place of the FDTS detection method used in the S / DF method, the error rate is improved. In addition, by using the most probable data string at this time from the data string recorded in the path memory as the temporary detection data, it is possible to perform the feedback operation with a predetermined delay amount, and it is a highly accurate detector. .

[0029]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing a main structure of an information reproducing apparatus using a feedback filter according to the present invention, showing a structure of a portion for generating detection data from a reproduced waveform obtained from a reproducing head which is a detector. Shows.

This embodiment uses a metric calculator (ACS) 2 and a path memory 4 used in the PRML system or the like in place of the FDTS detector in the FDTS / DF system data detector. It is composed of a feedforward filter (FF) 1, a comparator 3, a selector and a feedback filter 6 which are provided in addition.

A reproduction waveform obtained from a reproduction head (not shown) is equalized by a feedforward filter (FF) 1 from the discrimination point of the impulse response waveform to "0" at the leading edge portion thereof and "1" at the discrimination point. It is output as a waveform ai (i is time) rising to "". The output ai of the feedforward filter 1 is a feedback filter (FB)
It is added to the output bi of 6 to be yi, which is output to the metric calculator (ACS circuit) 2. The output ai of the feedforward filter 1 is added to the output bi of the feedback filter 6 to generate a waveform in which the trailing edge from the number of bits (tree depth γ) determined from the rising position is equalized to "0". Becomes The definition of the tree depth γ is the same as the definition in the FDTS / DF method, and indicates the number of bits from the identification point to the position where the feedback filter 6 equalizes.

The metric calculator 2 calculates the equalization error E for each equalization reference level value from the input yi (equalization waveform) based on the state transition diagram derived from the tree depth γ and the waveform output of the impulse response. A branch metric and a path metric for each branch are obtained. Then, the path metrics to each state are compared, and the path having the smallest path metric is selected as the surviving path to that state.

The metric calculator 2 performs the above operation,
The equalization error E and the path metric L of the surviving path corresponding to each state are output to the comparator 3, and the surviving path P (selection result) is output to the path memory 4.

The comparator 3 obtains the minimum value of the path metric L of the surviving path to each state, outputs the state number S corresponding to the path giving the minimum value to the selector 5, and at the same time equalizes the state number S. From the error E, the equalization error εi corresponding to the most probable path is selected and output to the feedback filter 6.

The selector 5 selects one of the plurality of data strings D corresponding to the surviving paths recorded in the path memory by using the selected state number S, and temporarily detects the data string d.
Output ^* . The tentative determination data string d ^* is a data string before the time point that is delayed by γ bits from the current time point i. In FB, based on this d ^{* value} , the trailing edge portion of the impulse response of the FF output (which is delayed by γ bits from the rising edge). A compensation value for equalizing (before the time point) to “0” is calculated and output. On the other hand, as the final detection data d, the final stage output of the path memory is used without using the temporary detection data.

Next, the operation of this embodiment will be described.

First, the operation principle of the feedforward filter 1 and the feedback filter 6 in FIG. 1 will be described. FIG. 2 shows impulse response waveforms corresponding to several γ values. Here, assuming a magnetic recording device,
"1" of the recorded data corresponds to a dipulse (a continuous waveform of two solitary waves having different polarities) in the reproduced waveform. That is, (1-D) conversion is performed on the recording data, and the solitary wave is made to correspond to the point where the conversion data is "± 1". γ is defined similarly to the tree depth in FDTS / DF, and γ bits from the identification point are used for maximum likelihood detection without being equalized by the feedback filter 6.

FIG. 2A shows γ = 2, and FIG. 2B shows γ =
3 and FIG. 2 (c) are equalized waveforms of γ = 4 (impulse response)
Is an example of.

In each impulse response waveform,
The feedforward filter (FF) 1 performs equalization on the leading edge portion from the time when it first rises from 0 to +1 (discrimination point), and all are equalized to "0".

The waveform output values a, b, and c after the identification point shown in FIG. 2 are arbitrary values, and may be determined in consideration of the characteristics of the reproduced waveform and the error rate.
Since the value of c is determined by the feedforward filter 1 if the reproduced waveform does not change, it can be considered that the value of c is equalized by the feedforward filter 1.

On the other hand, the equalization for the trailing edge portion after the time point γ from the discrimination point is performed by the feedback filter (FB) 6 and all equalized to "0". Since the feedback filter 6 performs compensation based on the determination result at the discrimination point, it is not affected by noise unless there is a determination error. By the above, an impulse response waveform in which all the bits except the γ-bit from the identification point are equalized to “0” is obtained.

The equalized waveform y obtained as described above is
It is input to the metric calculator (ACS circuit) 2 of the maximum likelihood (Viterbi) detector shown in FIG.

FIG. 3 shows a basic state transition diagram for performing metric calculation for maximum likelihood detection in this embodiment. 3A shows an example of γ = 2, FIG. 3B shows an example of γ = 3, and FIG. 3C shows an example of γ = 4. These state transition diagrams are constructed based on the impulse response waveform shown in FIG. 2, and the values shown in each branch represent the detection data and the equalization reference value. Based on such a state transition diagram,
The branch metric and the path metric are obtained from the difference from the equalization reference value (equalization error E), and the surviving path for each state is determined. Although the conventional ACS circuit only outputs the surviving path P to the path memory, the metric calculator (ACS circuit) in this embodiment.
In 2, the path metric L and the equalization error E for each surviving path are also output at the same time.

The output path metric L and the equalization error E
Is input to the comparator 3 shown in FIG. The comparator 3 compares the path metric L corresponding to each state,
The state number S corresponding to the path giving the minimum path metric value is calculated. Further, a corresponding value is selected from the equalization error E according to the state number S and is output as the equalization error εi. The equalization error εi is not necessary when the feedforward filter (FF) 1 and the feedback filter (FB) 6 are fixed parameter filters, but when performing adaptive equalization, this equalization error εi is used. Adaptive equalization can be performed on the parameters of the feedforward filter (FF) 1 and the feedback filter (FB) 6 to sequentially change them. In the adaptive equalization of the feedforward filter (FF) 1, the leading edge of the impulse response is set to “0” and the parameters are changed so that the γ-bit waveform output value from the discrimination point matches the value shown in FIG. And the feedback filter (F
B) In the adaptive equalization of 6, the trailing edge of the impulse response is "0".
Is to change the parameters so that

The structure of the path memory 4 is basically the same as the path memory in the conventional PRML system. The changes are that the value at the first stage is changed based on the state transition diagram, and that the contents of the path memory 4 are data strings D corresponding to each state, i-γ to i- for the current time i. γ-n
Up to +1 time point, the number n is the feedback filter (F
B) It is output as an input candidate to 6.
Since the initial stage selector becomes meaningless by changing the initial stage value of the path memory 4, the initial stage delay element and selector can be deleted.

In the selector 5, from these data strings D
The most probable tentative judgment data at the moment depending on the state number S
Row d^*Is selected. Feedback filter (F
B) 6 is this data string d^*Waveform equalization based on
Compensation value for) is output. On the other hand, in this embodiment
The detection data d as a detector is the above-mentioned temporary detection data d.
^*Instead, it is output from the final stage of the path memory.
Therefore, make the path memory long enough to converge the paths.
The maximum likelihood determination result can be obtained.

In this embodiment, the γ value can be set to a value of 2 or more. Further, the waveform output value after the identification point of the impulse response can be set in advance and maintained by performing adaptive equalization on the feedforward filter (FF) 1. The state transition diagram for maximum likelihood detection may be created using these values.

Next, the operation of the above-described embodiment will be described more specifically taking the case of the tree depth γ = 3 as an example.

FIG. 4 is a diagram showing the equalization process of the reproduced waveform corresponding to the recording data "1" (equalization process of impulse response), and the equalization process will be described in detail with reference to FIG.

The solid line waveform is the impulse response of the feedforward filter (FF) 1. The time (identification point) at which the output value rises by the feedforward filter (FF) 1 is set to "1", and the leading edge is equalized to 0 from that point. At this time, the equalization is not performed on the trailing edge portion from the identification point. Since the feedback filter (FB) 6 in the present embodiment outputs the compensation value with a delay of γ bits, the compensation can be performed from the point of γ bit delay from the discrimination point. The arrow at each time point indicates the compensation amount, and the dotted line indicates the equalized waveform after compensation by the feedback filter (FB) 6. The values of a and b are arbitrary and are determined by the characteristics of the reproduced waveform and the like. Further, the forms of the feedforward filter (FF) 1 and the feedback filter (FB) 6 are also arbitrary.

Feedforward filter (FF) 1
Since it suffices that the reproduced waveform be equalized as shown by the solid line in FIG. 4, an analog filter or a transversal filter can be used. Feedback filter (FB) 6
Since it suffices that the compensation value for equalization can be output from the detected data string, it is possible to use a transversal filter or a RAM table.

The metric calculation may be performed based on the state transition diagram shown in FIG. 3B, and is basically the same as that of the conventional ACS circuit. The difference is that, in addition to the surviving path P _0-3 corresponding to each state, the path metric value L _0-3 corresponding to the four surviving paths and the equalization difference E _0-3 are output.

FIG. 5 shows the configuration of the comparator 3 which finds the smallest path metric from these path metrics L _0-3 and finds the corresponding state number S and the equalization error ε for adaptive equalization. The comparator 3 is composed of a comparator 10 and a selector (S) 9. The comparator 10 compares the four path metrics L _0-3 and outputs the state number S that gives the minimum value. The selector (S) 9 outputs Es as an equalization error εi from four input values (equalization error E _0-3 ) based on the state number S.

FIG. 6 shows the path memory 4 and the selector 5.
Shows the configuration of. The structure of the path memory is basically the same as that of the conventional PRML system. However, state transition diagram 3 (b)
According to the above, since the selectors in the first stage are selected from the same value, the delay elements and selectors in the first stage are omitted. Further, the data string D of n stages is output to the outside in correspondence with the input data number n bits of the feedback filter (FB) 6. The data D from each stage (each time point) is four pieces corresponding to the state, and is the determination data corresponding to the surviving path of each state.

The selector 5 selects the data at each time point according to the state number S and outputs the most probable temporary judgment data string d ^* at the present time. The final detection data is output from the rear end of the path memory 4. At this time, the path memory length is the feedback filter (FB).
If it is longer than the number of input data of 6, the configuration shown in FIG. 6 may be used, but if it is shorter, the detection data may be held by using a delay element and input to the feedback filter (FB) 6. Such a configuration may be adopted.

FIG. 7 illustrates the process of determining the temporary determination data string.
FIG. 7A is a diagram for clarifying the trellis diagram.
Here is an example. The surviving path at time i is shown as a solid line
It When such a survival path is required,
In the memory, data as shown in Fig. 7 (b) is recorded.
Has been. When the surviving paths merge, the current
The judgment data all match. Here, i-3
Merged with, and the detection data before that has been confirmed.
It Path metric L of the surviving path at the present moment _0-3of
Of these, the smallest value is L₁If it was,
The enclosed data string corresponding to state S1 is selected and
Used as input to the feedback filter (FB) 6
It

As described above, feedback can be applied with a delay of γ clock as in the FDTS / DF method. FIG. 8 shows equalization reference values and branch metrics for performing metric calculation. Based on these equations, branch metric and path metric are calculated and maximum likelihood detection is performed. The equalization reference value R and the constant derived from R can be calculated in advance to reduce the calculation amount.

Next, a second embodiment of the present invention will be described with reference to FIG.
And it demonstrates with reference to FIG.

In the present embodiment, the comparator 3, the path memory 4 and the selector 5 in the first embodiment are replaced by a comparator 903, a path memory 904 and a selector 905 having different configurations from those of the first embodiment, and FIG. 10 is a diagram showing a configuration of a comparator 903 in the second embodiment of the present invention, and FIG.
FIG. 6 is a diagram showing configurations of a path memory 904 and a selector 905 according to a second embodiment of the present invention. Since the other configurations are similar to those of the first embodiment, only these configurations will be described.

In the embodiment shown in FIGS. 5 and 6, the number of clocks required for the signal to propagate through the feedback loop is only one clock. With this one clock,
It is necessary to calculate the equalized waveform, calculate the branch metric and the path metric, select the surviving path, detect the minimum path metric, select the temporary detection data, and further calculate the compensation amount with the feedback filter 6 (FB). ,
The circuit load is very large. Therefore, in the second embodiment, it is possible to increase the number of clocks required for the signal to propagate through the feedback loop.

FIG. 9 shows the configuration and operation of this embodiment.
Further, with reference to FIG. 10 and FIG. In the comparator 903 of this embodiment, the delay element (D) 907 ₁ ,
907 ₂ is provided, and the path memory 904 is provided with a delay element (D) 907 ₃ . Each of these delay elements (D) delays the path metric L, the equalization error E, and the selected branch (surviving path) P, which are the outputs of the ACS circuit 2, by 1 bit, and instead, the path memory 904.
The data string D output from is shifted by 1 bit (1
Output earlier). As described above, the calculation of the feedback loop can be performed with a time margin of one clock. However, in this method, the provisional determination data d ^* has low accuracy because it uses the output from the stage number lower by 1 bit in the path memory.

In order to obtain a time margin of k bits (clock), the output position of the data D shown in FIG. 10 is shifted toward the lower stage number of the path memory, and a delay element is inserted somewhere instead. Good. At that time, the timing of each calculation should not be shifted. This causes the feedback loop to operate with 1 + k clocks.
Further, since the equalization error for performing adaptive equalization may be deviated from the time point i, adaptive equalization is performed accordingly. In this method, the feedback loop can generally be operated with a maximum γ-1 bit delay.

[0064]

Since the present invention is constructed as described above, it has the following effects.

In the FDTS / DF method, the maximum likelihood detection method (Viterbi detection method) used in the PRML method is used instead of the FDTS detection, which has the effect of improving the error rate.

[Brief description of drawings]

FIG. 1 is a block diagram showing the overall configuration of a detector of the present invention.

FIG. 2 is a diagram showing an impulse response for each γ in the present invention.

FIG. 3 is a state transition diagram for each γ according to the present invention.

FIG. 4 is a diagram showing a waveform equalization method for γ = 3 in the present invention.

5 is a diagram showing a detailed configuration of the comparator 3 of FIG. 1 when γ = 3.

6 is a diagram showing a detailed configuration of the path memory 4 and the selector 5 of FIG. 1 when γ = 3.

7 is a diagram showing an example of operations of the path memory 4 and the selector 5 of FIG.

8 is a diagram showing a metric calculation formula when γ = 3 in the ACS circuit 2 of FIG. 1;

FIG. 9 is a diagram showing a configuration of a comparator 3 when γ = 3 according to another configuration of the present invention.

FIG. 10 is a diagram showing a configuration of a path memory 4 and a selector 5 when γ = 3 in another configuration of the present invention.

FIG. 11 is a block diagram showing a configuration of an FDTS / DF method as a conventional example.

FIG. 12 is a diagram showing a tree search (FDTS) detection method of the FDTS / DF method as a conventional example.

FIG. 13 is a diagram showing an impulse response waveform of the FDTS / DF method as a conventional example.

[Explanation of symbols]

1 FF (feed forward filter) 2 ACS circuit (metric calculator) 3 comparator 4 path memory 5 selector 6 FB (feedback filter) 7 1-bit delay element 8 selector (2 inputs, 1 output) 9 selector (4 inputs, 1 Output 10 Comparator (minimum path metric detector) 11 FDTS detector ai FF output yi Equalized waveform bi FB output P Survival path for each state L Path metric for each surviving path E Equalization error for each surviving path D Each survivor Detected data string for a path d ^* Temporary detected data string (corresponding to the path having the smallest path metric among the surviving paths) d Detected data string (output of the last stage of the path memory) γ tree depth (Definition is the FDTS / DF method N) Number of input data of n FB m Path memory length εi Equalization error (provisional judgment data The corresponding) S selected number (corresponding to the path having the smallest path metric among the survivor path) i times a, b, c impulse response waveform output value R equalizing reference value B branch metric

Claims (6)

(57) [Claims] 1. A feedforward filter which equalizes a leading edge portion of an impulse response of a reproduced waveform and sets an output value at a discrimination point to "1" and an output value before that to "0", and the discrimination. From the feedback filter for compensating the trailing edge after the point of tree depth γ bit delay from the point, and the equalized waveform in which the feedforward filter output and the feedback filter output are superposed, the tree depth γ and the impulse response Equalization error E for each equalization reference level value, branch metric for each branch, and path metric are obtained based on the state transition diagram derived from the waveform output, the path metric for each state is compared, and the path metric is the minimum. A metric calculator that selects a path that becomes a surviving path to the state, and Enter the-option has been survivor path for current i, from i-γ i-γ-n + 1
Data string D with n pieces of data up to the point in time corresponding to each state
And a path memory that outputs a final-stage value as detection data, a comparator that determines a path having a minimum path metric from among surviving paths to each state obtained by the metric calculator, and the path memory A selector for inputting a data string D output by the selector, selecting a data string corresponding to a path that gives the minimum path metric obtained by the comparator, and outputting the selected data string to the feedback filter. An information reproducing apparatus using a feedback filter, characterized in that a compensation amount for equalization is obtained based on a data string selected by the selector. 2. The information reproducing apparatus using the feedback filter according to claim 1, wherein the path memory length is shorter than the number of input data of the feedback filter, and the detection data output from the path memory is held and fed back. An information reproducing apparatus using a feedback filter having a delay element for outputting to a filter. 3. An information reproducing apparatus using the feedback filter according to claim 1, wherein the selector corresponds to a path having a minimum path metric among equalization errors corresponding to each surviving path. An information reproducing apparatus using a feedback filter, characterized in that an equalization error is selected and a feedforward filter and a feedback filter are adaptively equalized using the selected equalization error. 4. The information reproducing apparatus using the feedback filter according to claim 1, wherein the feedback filter and the metric calculator have a tree depth γ of 2, and the metric calculator uses the feed. An information reproducing apparatus using a feedback filter, wherein maximum likelihood detection is performed by making a state transition diagram of two states correspond to a reproduced waveform in which a forward filter output and a feedback filter output are superposed. 5. The information reproducing apparatus using the feedback filter according to claim 1, wherein the feedback filter and the metric calculator have a tree depth γ of 3, and the metric calculator uses the tree depth γ. An information reproducing apparatus using a feedback filter characterized by performing maximum likelihood detection by associating a state transition diagram of four states with a reproduced waveform in which a feedforward filter output and a feedback filter output are superposed. 6. The information reproducing apparatus using the feedback filter according to claim 1, wherein the feedback filter and the metric calculator have a tree depth γ of 4, and the metric calculator uses the tree depth γ. An information reproducing apparatus using a feedback filter, wherein maximum likelihood detection is performed by associating a state transition diagram of eight states with a reproduced waveform in which a feedforward filter output and a feedback filter output are superposed.