JPS5818683B2 - Error correction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION 1. Object of the Invention The invention is applied to an error correction system whose purpose is to correct concentrated errors on a recording medium (eg, magnetic disk, drum, tape).

2. Prior Art Referring to FIGS. 1 to 3, an explanation will be given of how error correction is applied.

Figure 1 shows a recording format typically used in magnetic disks, etc., in which a count field 1, a data field 3, and an error correction code 2 are added to the trailing end of each track to form a record. Consecutive pieces of information are separated by gaps 4 and recorded.

FIG. 2 shows an enlarged view of the data field 3 and error correction code 2 shown in FIG.

Following the 15-byte data information, a 7-byte error correction code (hereinafter referred to as ECC)
tion Cade) 2 is recorded.

With this ECC2, for example, if a concentrated error within 11 consecutive bits occurs between bytes 5 to 7, the error displacement that displays the error pattern and error position is calculated, and the actual error is calculated from these two contents. It is possible to obtain correct data from data that contains errors that have been read.

As an FCC that corrects intensive errors, for example, FI
A KE code or a modified version thereof is used, in which case the ECC pattern is calculated as a function of the previous data 3 pattern and recorded on the recording medium.

When reading, FCC is calculated from the read data,
The calculated FCC (C) and the FCC (K) actually read from the medium are compared, and if they match, it is determined that the read data is error-free.

If the FCC(C) and ECC(R) do not match, it indicates that some error exists.

FIG. 3 shows a flowchart of error correction according to the prior art. If FCC(C) and FCC(R) match, block 11 becomes No, and the read data is used as is.

If ECC (C'l) and ECC (R) do not match, block 11 will be YES, and the next block 12 will check whether the error can be corrected.

If the error can be corrected here, block 12
ES, an error pattern and error displacement are calculated, errors are corrected based on these two contents, and the corrected data is used.

If the error cannot be corrected, the answer to block 12 is NO. In this case, the FCC states that it is impossible to calculate correct data, and blocks 10, 13, 14. The data is reread until the correct data is found.

When this number reaches N. This data is considered unreadable.

The reason why error correction is necessary is that unless fixed errors such as scratches on the medium are tolerated, high-density recording would be extremely expensive due to media selection and other costs.

Even if it is said that scratches and the like are allowed, it is limited to one place or less with a size of less than the scratch number ζ usually C bits.

(Tolerating more scratches will cause errors that will become uncorrectable.

) Even under such restrictions, yields can be greatly improved by producing recording media with no scratches.

If the error is caused only by temporary noise, the data can be read during several rereads, and there is no particular reason why ECC should not be used.

3 Problems with the Prior Art The concentrated error correction method is an error correction method that is based on the following assumptions.

1) If the spread of the error (the length including both ends of the error bit) is within a certain range (for example, within 11 bits), the error pattern and displacement can be determined correctly.

(This is referred to as error correction ability
This spread is expressed by C bits.

)2) If the spread of the error is C+1 bits or more and within a certain number of bits, it is determined that it is erroneously correctable by the operation of block 12 in FIG.
There is a range in which there is no such thing as "correction".

Detectability
It is called A (hereinafter abbreviated as DA), and a value of 22 bits is used as an example.

3) When the error range further expands, there is an error pattern in which the operation of block 11 in FIG. 3 incorrectly determines that there was no error even though there was an error.

Detectability B (hereinafter referred to as D) is the spread of errors that do not cause false detection.
It is abbreviated as B.

) in the example using the FI RE code, it matches the length of the FCC, and in the example of FIG. 2 it matches 7×8=56 bits.

Now, the problem is that DA=22 bits, DB=
This is the meaning of the two 56-bit numbers.

For both DA and DB, if it can be assumed that the error occurs in one field and from one cause, that is, only one place,
One of these is the spread of errors, which indicates how far there is a probability that incorrect data will be used.If we compare this with the number of error patterns in which no errors are detected, we can see the following: It is expected that there will be a wide difference.

For example, if the error spread is 57 bits, 1) the third
The number of patterns for which a NO decision is made in block 11 of the figure is only one because there is only one pattern whose error pattern matches the generating polynomial.

2) Although it is generally difficult to calculate the number of patterns for which YES is determined in block 12 of Fig. 3, the error length is set to 57.
Even within bits, it is thought that there are patterns corresponding to some power of 10.

This means that DA is so much worse than DB that it is comparable to some power of 10.

The above is a case where the error is concentrated in one place caused by one cause, but in reality, the error is caused by two or more causes (for example, one is a fixed error caused by scratches on the medium, etc.),
On the other hand, it cannot be ignored that the error occurs due to a temporary error caused by noise, and in this case, the spread of the error is from one end of these two errors to the other, and the error exceeds both DA and DB. Even in the case where DA has far more error patterns than DB, M i 5corr
It is expected that this will occur.

(In other words, incorrect data will be used without error detection.

) In general, when miscorrection occurs, a C-bit error pattern and error displacement that are unrelated to the actual error are obtained, but these two pieces of information are determined as a function of the actual error. Distinguish whether there is an actual correctable error corresponding to these two pieces of information that resulted in such a result, or whether there was an error that exceeded your ability and resulted in such a result by mistake. I have no means at all.

Therefore, when reading a track with fixed scratches, for example, if a single noise occurs, it means that an error with a possibility of M i 5 correction has occurred immediately, and the DA's ability becomes a big problem.

4. Despite the occurrence of an error that cannot be corrected in accordance with the purpose of the invention,
The purpose is to prevent erroneous error correction.

5 Key Points of the Invention When a correctable error occurs, immediately reread without using correction data and confirm that the displacement occurs at a fixed location (there is indeed a fixed error). Then use the corrected data.

6 Embodiment of the Invention FIG. 4 shows a flowchart of error correction according to an embodiment of the invention.

In FIG. 4, blocks 16 to 18 are operations added for the present invention compared to FIG.
It will be explained that the probability of scorrection is greatly improved.

Case 1: An error occurs due to a single scratch.

Case 2: A temporary error occurs due to noise.

Case 3: An error occurs due to scratches (one place) and noise.

Case 1 If NO is determined in block 12 of the first read, data is read again.

If YES is determined in block 12, it is the first c.
It is determined in block 16 whether the data is orrectable, and if so, the displacement is stored in block 17 and the data is read again.

Since the error is caused by a fixed error due to scratches, the result is YES even in block 11 of two readings.

If the next block 12 determines YES again, this is not the first time, so the determination in block 16 becomes NO.

Therefore, the process goes to block 18, where the previously stored displacement and the currently calculated displacement are compared. There is a high possibility that they will match with an error (for example, ±1 byte), and in this case, correct data is determined from the second read data, error pattern, and displacement, and the corrected data is used.

If the comparison in block 18 does not result in a match, rereading is repeated up to a predetermined number of times (N) until a match is found.

Case 2 If the determination is No in block 12 of the first reading, the data is read again.

If YES is determined in block 12, it is the first C.
It is determined in block 16 whether orrgctble, and if so, the displacement is stored in block 17 and the data is read again.

Since this is caused by a temporary error due to noise, there is a high possibility that the second reading will result in No in block 11 and correct data will be obtained.

However, if the result in block 11 is YES again, rereading is repeated up to a predetermined number of times (N) until the result in block 11 is NO.

Case 3 If it is judged as No in block 12 of the first reading,
Reread the data.

If YES is determined in block 12, it is the first c.
It is determined in block 16 if it is an orrec table, and if so, the displacement is stored in block 17 and the data is read again.

In the second reading, the noise disappears, the fixed error due to the scratch remains, and block 11 also returns YES, but block 12 returns YES again, and block 1
8, the previously stored displacement and the currently calculated displacement C5 do not match.

This is because the previous one was calculated from errors due to scratches and noise, whereas this time, the noise disappeared in the second reading and was calculated from errors only due to scratches. It is.

As a result, the new displacement is stored in the second block 17 and read out again.

If YES is determined again in block 12 of the third reading, the displacement stored last time and the displacement calculated this time are compared in block 18.
There is a high possibility that the comparison results will match to the same degree as in case 1, and in this case, the data read the third time,
Correct data is obtained from the error pattern and displacement, and the corrected data is used.

If the comparison in block 18 does not result in a match, rereading is repeated up to a predetermined number of times (N) until a match is found.

In any of the above cases, temporary errors such as noise may be introduced during trials, but the noise will definitely disappear after several trials, and eventually correct error correction information will be obtained. Can be done.

7 Effects of the Invention According to the present invention, error correction can be performed based on the reliability of the DB.

Also, given the required reliability, the goal can be achieved with a smaller error correction code length than the FCC.

Note that according to the present invention, the number of retries increases, but since the frequency of occurrence of scratches and noise is usually extremely small, the increase in overhead due to this is negligible.

[Brief explanation of drawings]

Fig. 1 is a diagram showing the currently used data recording format, Fig. 2 is a partially enlarged view of Fig. 1, Fig. 3 is a flowchart of error correction according to the prior art, and Fig. 4 is a diagram showing the present invention. 3 is a flowchart of error correction according to the embodiment. (5)

Claims (1)

[Claims] 1. When a correctable error is detected when reading data on a recording track, the same data is read four times again before using the correction data for the error,
An error correction method characterized in that the correction data is used after confirming that the position of the error occurring in the reading is the same as before.