KR20100064180A - Maximum likelihood decoding apparatus and method - Google Patents

Abstract

PURPOSE: A maximum similarity decryption apparatus and a method are provided to increase a transferring speed of a data by reducing a time to be required in a process of decoding a raptor code through the maximum similarity decryption. CONSTITUTION: A blocked matrix U generator(1010) generates the blocked matrix U by implementing maximum score search about the matrix of a MxL size including a raptor code. A modified matrix generator(1020) generates the modified matrix including only one on each row and column by implementing a Gaussian elimination to the blocked matrix U. A vector calculator(1030) calculates the vector of length L using the vector of length R with the modified matrix. The blocked matrix U generator generates an initial matrix by only storing a location of '1' in each row and column of the matrix of the MxL size.

Description

Maximum Likelihood Decoding Apparatus and Method {Maximum Likelihood Decoding Apparatus and Method}

The present invention relates to a maximum similarity decoding apparatus and method. More specifically, maximum likelihood decoding is used in decoding the Raptor Code, which is an error correction code adopted in The 3rd Generation Partnership Project (3GPP) Multimedia Broadcast and Multicast Service (MBMS). The present invention relates to an apparatus and method for decoding that can reduce the time required for decoding.

Raptor code, which is an error correction code adopted in 3GPP MBMS, can be efficiently decoded by the maximum likelihood decoding method. Here, the maximum similarity decoding means the best method for restoring original information. Since the time and space required for decoding are very small, the raptor code can be efficiently decoded.

3GPP TS 26.346 V7.4.0, "Technical specification group services and system aspects; multimedia broadcast / multicast services; protocols and codecs, June 2007", is a 3GPP standard document that describes the Raptor Code. The maximum likelihood decoding method proposed in this standard document is divided into four phases. Of the four phases proposed by this standard, the first phase is the most important part, and the time spent in the first and second phases accounts for more than 95% of the total decoding time, which is required in two phases. As described above, there is a problem of requiring a large amount of time and greatly delaying the decoding time.

In order to solve the above problems, the present invention can reduce the time required when decoding the Raptor code using the maximum similarity decoding, thereby increasing the data transmission speed and reducing the power consumption required for decoding. There is a main purpose.

According to an aspect of the present invention, there is provided an apparatus for maximum similarity decoding of a raptor code, comprising: a blocked matrix U generator for performing a maximum score search on an MxL sized matrix having a raptor code to generate a blocked matrix U; A modified matrix generator for performing a Gaussian elimination method on the blocked matrix U to generate a modified matrix having only one '1' in each row and each column; And a vector calculator for calculating a vector having a length L using the transform matrix and the vector having a length r.

According to another object of the present invention, there is provided a method of maximum similarity decoding of a raptor code, the method comprising: generating a blocked matrix U by performing a maximum score search on an MxL sized matrix having a raptor code; A modified matrix generation step of performing a Gaussian elimination method on the blocked matrix U to generate a modified matrix having only one '1' in each row and each column; And a vector calculation step of calculating a vector having a length L using the transform matrix and the vector having a length r.

As described above, according to the present invention, since the time required for decoding the Raptor code using the maximum similarity decoding can be reduced, the data transmission speed can be increased, and the power consumption required for decoding can be reduced. have.

Hereinafter, some embodiments of the present invention will be described in detail through exemplary drawings. In adding reference numerals to the components of each drawing, it should be noted that the same reference numerals are used to refer to the same components as much as possible even if displayed on different drawings. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

In addition, in describing the component of this invention, terms, such as 1st, 2nd, A, B, (a), (b), can be used. These terms are only for distinguishing the components from other components, and the nature, order or order of the components are not limited by the terms. If a component is described as being "connected", "coupled" or "connected" to another component, that component may be directly connected or connected to that other component, but between components It will be understood that may be "connected", "coupled" or "connected".

A. Maximum Similarity Decoding According to 3GPP Standard

According to The 3rd Generation Partnership Project (3GPP) standard, the maximum likelihood decoding of Raptor Code is to solve the equation 'Ac = r'. Here, A represents an MxL (where M? L) matrix, c is a vector of length L, and r is a vector of length M. A and r are already known, and it can be said that the problem of maximum similarity decoding is to find c for which the above equation is valid.

Diagonalizing the first L rows of matrix A yields c. The maximum similarity decoding method proposed by the 3GPP standard describes a method of efficiently diagonalizing the first L rows of the matrix A.

1 is an exemplary diagram illustrating an MxL matrix that is a result of performing a first phase on matrix A.

First, when the first phase is performed according to the 3GPP standard, the original matrix A is changed into an MxL matrix as shown in FIG. 1. In FIG. 1, I represents an identity matrix, and all 'all zero' below the I matrix represent a matrix having elements of '0'. The two submatrices U1 and U2 on the right are the remaining matrices obtained by creating the left submatrix I and all zeros, respectively. You can get this type of matrix through elementary row operations and column exchanges. If the submatrices U1 and U2 occupy u columns, it is assumed that the Mxu submatrix occupied by the two submatrices U1 and U2 is the submatrix U.

2 is an exemplary diagram illustrating an LxL matrix that is a result of performing a second phase on an MxL matrix.

The second phase of the 3GPP standard transforms the submatrix U2 into an identity matrix after erasing the last M-L row through a general Gaussian elimination. When the second phase is performed, the matrix of the form shown in FIG. 1 is transformed into an LxL matrix as shown in FIG.

The third and fourth phases of the 3GPP standard use the I matrix in the lower right corner to convert U1 into a matrix with all elements '0'. Once this conversion is complete, the original matrix A is converted to an identity matrix, resulting in c.

B. Problems of Maximum Similarity Decoding According to 3GPP Standard

According to the maximum similarity decoding method proposed by the 3GPP standard, the maximum component search (MCS) must be executed in the first phase, which takes a considerable time. In addition, in the process of transforming the submatrix U2 into an identity matrix using Gaussian elimination in the second phase, there is room for optimizing the implementation. Finally, the original matrix A is transformed into an identity matrix to find the vector c, which is not the best way to find the vector c in terms of time.

C. Maximum similarity decoding method according to an embodiment of the present invention

The maximum similarity decoding method of the raptor code proposed in the present invention is based on the following five points.

1. Since matrix A has a sparse property of the number of '1's, the matrix A does not decode using the matrix itself, and stores and decodes only the position of' 1 'existing in each row / column. This not only significantly reduces the amount of space required for decoding, but also significantly reduces the time required to search for rows that satisfy certain conditions, which are very frequent during the decoding process, greatly improving the overall decoding time. You can.

3 is an exemplary view showing a matrix A expressed according to an embodiment of the present invention.

Referring to FIG. 3, as shown in FIG. 3, the matrix A in the Ac = r equation, Cidx [1],. , Cidx [L], Ridx [1],... , Ridx [M]. Where Cidx [i] represents the position of the row of '1' in the i th column, and Ridx [j] represents the position of the column of '1' in the j th row. In addition, Cidx [i] .Deg represents the number of '1's in column i, Cidx [i] .Edge represents the position of every' 1 'row in column i, and Cidx [i] .Edge [k] Denotes the position of the k th '1' row of the i column. Ridx [j] .Deg represents the number of '1's in j rows, Ridx [j] .iDeg represents the number of' 1's in j rows before decoding starts, and Ridx [j] .Edge represents j rows Ridx [j] .Edge [l] represents the position of the lth '1' column in row j, and Ridx [j] .score represents the score of row j. Represents information about. A score is demonstrated in the process mentioned later.

2. The purpose of maximum similarity decoding is to find the vector c. To this end, as in the 3GPP standard, matrix A may be converted into an identity matrix to obtain c, but in this case, decoding takes much time. In an embodiment of the present invention, the matrix A is transformed into a matrix in which only one '1' exists in each row / column. In this manner, even when the matrix is deformed, the vector c can be obtained.

3. Searching with Ridx / Cidx can save you more time than searching on a matrix. However, if there are deletions in Ridx / Cidx, it takes a lot of time to keep the Ridx / Cidx aligned. Therefore, we use Cidx.Eidx and Ridx.iDeg. Cidx.Eidx is defined as in Equation 1, and is made in advance as in Equation 1 before starting decoding.

Cidx [i] .Edge [j] = k, Ridx [k] .Edge [x] = i → Cidx [i] .Eidx [j] = x

When decoding starts and Cidx.Eidx is used, an index of a row can be deleted within a constant time as shown in Equation 2 without searching Ridx [k] .Edge.

Cidx [i] .Edge [j] = k, Cidx [i] .Eidx [j] = x → Ridx [k] .Edge [x] =-1

4 is an exemplary diagram illustrating an arbitrary Ridx [k] in a state where decoding is performed according to an embodiment of the present invention.

Referring to FIG. 4, there is '-1' in Ridx [k] .Edge, which is a deleted index, and it can be seen that Ridx [k] .Deg is smaller than Ridx [k] .iDeg. When searching Ridx.Edge, you can search as much as iDeg and find the desired index by checking whether the index is valid with '-1'. Since the search time in the index is much smaller than the search time in the matrix, and the deletion time can be made to O (1), the speed of the program can be greatly improved. This method can only be used where the deletion only occurs, and you must use a different method for matrix U.

4. Run Largest Score Search (LSS) instead of Maximum Component Search. The existing maximum component search reduces the size of the matrix U, but takes too much time to search, and consequently slows down the overall program. The maximum score search method reduces the size of the matrix U smaller than the maximum component search method, but dramatically improves the speed of searching for rows having a degree of '2', thereby improving the overall program speed. The maximum score search is implemented in three parts.

(1) Part 1: Initial Score

In Part 1, as a preprocessing step for searching for the maximum score, a score table (Score-) that calculates the values of the scores of all the rows before starting decryption, and has the score as the key for the row whose degree is '2'. Table).

1a. In the initial state, the score for each row is the sum of the degrees of the column at the index of the row.

5 is an exemplary diagram for explaining a process of calculating a value of a score of each row of an initial matrix.

5, Ridx [2] .Deg = 3, Ridx [2] .Edge [0] = 1, Ridx [2] .Edge [1] = 4, Ridx [2] .Edge [2. ] = 8, the value of the score of the third row of the initial matrix as shown in FIG. 5 is calculated as the sum of the degrees of the column at the index of the third row, as shown in equation (3).

Ridx [2] .score = Cidx [1] .Deg + Cidx [4] .Deg + Cidx [8] .Deg

1b. When the row's degree is '2', insert the row's index into the score table using Ridx.score as the key. Then, the largest score among the rows with a degree of '2' is stored in the maximum score.

6 is an exemplary view showing a score table according to an embodiment of the present invention.

Referring to FIG. 6, in 13 (2, 2) of the first row, '13' means score or key, the first '2' is the number of rows having the row degree '2', and the second '2' 'Is the number of rows in the first line. When generating the score table for the first time, two numbers d1 and d2 in the horizontal direction have the same value. '31' and '78' are the indices of the row having a score of '13' and the row whose degree is '2'. In FIG. 6 the maximum score is '30'.

(2) Part 2: Ridx.score & Score-Table Update

In the first phase, when selecting and processing a row whose degree is '2', the number of rows among the rows whose degree was '2' before processing decreases after processing. Also, many of the rows whose degree was greater than '2' prior to processing reduced the degree, resulting in the degree being '2'. For this reason, an index whose Ridx [i] .Deg is not '2' is generated among the indexes of the rows in the score table.

Therefore, in order to keep the degree of rows corresponding to all the indexes in the score table to be '2', there is a problem in that inserting and deleting takes a long time. When updating to solve this problem, when doing the maximum score search instead of deleting the index from the score table for the rows that change from '2' to '1', Ridx [i] .Deg (where , i is an index (2) in the score table) can significantly reduce the time required to maintain the score table.

Assuming that the degree of the i-th row has decreased in the first phase, update Ridx [i] .score for the row that has decreased. When Ridx [i] .Deg decreases to '2', i is inserted into the score table using Ridx [i] .score as the key. In this case, i is inserted into the end of the array in the score table to reduce the insertion time. At this time, 'd1' and 'd2' of the score table increase by '1'. Also, compare the maximum score with Ridx [i] .score and update the maximum score if Ridx [i] .score is larger. When Ridx [i] .Deg decreases to '1'. Decrease 'd1' in the score table by '1' using Ridx [i] .score as the key.

7 is an exemplary view showing an updated score table according to an embodiment of the present invention.

Referring to FIG. 7, no deletion occurs in the score table. We no longer have a 'X' in the row index where the degree is not '2', but we don't do anything in the actual implementation. In addition, the newly added index (which is stored at the end of the array) in the score table is more likely to be valid (preferably '2') than the previous index. How to use this is described in Part 3 which will be described later.

(3) Part 3: Exploring the Maximum Score

In the maximum score search, to speed up by reducing the number of U columns, when selecting a row with a degree of '2', the row with the highest score is selected. At this time, by using the value of the maximum score and the score table, it is possible to find the row having the largest score at an early time. However, the closest thing to look for in the maximum score search is that the degree of the row corresponding to the updated maximum score decreases as the first phase progresses. In this case, the maximum score does not coincide with the 'largest score value among the rows having the degree 2'. However, since the maximum score is always greater than or equal to 'the largest score value among the rows having a degree of 2', as shown in FIG. 8, the score table is used to determine the 'largest score value among the rows having a degree of 2'. You can find it.

8 is an exemplary view illustrating a process of searching for a maximum score according to an embodiment of the present invention.

In Fig. 8,? Indicates the number of indices d1 of which the actual degree is '2' of the score table corresponding to the maximum score. Since there are no rows with a score of '28' and a degree of '2', the maximum score is reduced by '1'. In FIG. 8, ② has one row with a score of '27' and a degree of '2'. Use the total number of indexes (d2) to search from the back of the array to find valid index values. Line 119 has the largest score among rows with the actual degree '2'.

5. In matrix U, a new data structure is needed because index insertion and deletion occur and decoding increases density. Therefore, we use Blocked Matrix U, a new data structure that represents 32 columns in total for matrix U. The additional position of the blocked matrix U and the value to be added may be calculated as in Equations 4 and 5 below.

Additional position (blocked matrix columns) = largest integer less than (number of columns in current matrix U / block size)

Blocked Element Variable = 2 ^{(number of columns in current matrix U-1% (block size))}

One element of the matrix has only '0' and '1' as values, but one element of the blocked matrix has an integer value. Thus, one element of the blocked matrix can represent several elements of the matrix, and how many elements of the matrix each element of the blocked matrix represents is determined by the size of the block. As the size of the block is larger, the number of loops required to process the matrix U may be reduced, thereby reducing the decoding time. For example, the block size of the program is '32'. This is the largest block that can be used in an unsigned type.

According to an embodiment of the present invention, the method of decoding the maximum similarity decoding raptor code is as follows.

(1) first phase

a. When the minimum Ridx.Deg is '1',

First, Ridx [i] (1 ≦ i ≦ M) is sequentially searched for rows in which Ridx [i] .Deg is '1'. Search for a valid column in Ridx [i] .Edge that corresponds to the searched row, search Cidx [j] .Edge again, and determine the values of Ridx, Cidx, and the blocked matrix U. At this time, Ridx.score is calculated for the row that Ridx.Deg decreases, and if the value of the reduced Ridx.Deg is '1', it is put in an array called 'DIO (Degree is One)'. When using the DIO, if the changed Ridx.Deg value is '1', the DIO is put in the DIO to help reduce the time for searching for the value of the Ridx.Deg '1'. After processing the found rows, DIO searches for rows with Ridx, Deg equal to '1', searches for valid columns in Ridx [i] .Edge corresponding to the searched rows, searches Cidx [j] .Edge, Repeat the process of determining the values of Ridx, Cidx, and the blocked matrix U. If there are no more rows with Ridx.Deg '1' in the DIO during the iteration, Ridx [i corresponding to the row searched for rows with Ridx.Deg '1' again from rows (i + 1) to M ] .Edge searches for valid columns, Cidx [j] .Edge, and repeats the process of determining the values of Ridx, Cidx, and the blocked matrix U.

b. If at least Ridx.Deg is '2',

If the minimum Ridx.Deg is '2', processing for the selected row is started through the maximum score search. The maximum score search continues until there are no rows with Ridx.Deg equal to '2'. The matrix U is generated in the process of processing the rows where the selected Ridx.Deg is '2' through the maximum score search. Blocked matrix U part needs to be processed separately because matrix insert U is frequently inserted and deleted, and as the decryption proceeds, the density becomes high and index cannot be used. Thus, matrix U is processed using a blocked matrix U. Ridx.Deg decreases during the process of 'adding the blocked matrix U'. At this time, Ridx.score is calculated and if the reduced Ridx.Deg is '1', it is put in the DIO. On the other hand, when the addition of the blocked matrix U is completed, the matrix V may be processed as if the minimum Ridx.Deg is '1'.

(2) second phase

At the beginning of the first phase, Ridx stores information about the processed rows.

9 is an exemplary diagram illustrating a blocked matrix U before and after performing Gaussian removal according to an embodiment of the present invention.

9A shows the blocked matrix U of the rows before performing Gaussian elimination, and 9B shows the blocked matrix U after performing the Gaussian elimination. You can easily get 9B from 9A by following the usual Gaussian removal method.

10 is a block diagram illustrating a decoder of a raptor code according to an embodiment of the present invention.

One embodiment of the present invention described above may be implemented as a maximum similarity decoding apparatus of the raptor code. That is, the maximum similarity decoding apparatus 1000 of the raptor code according to another embodiment of the present invention is a maximum similarity decoding apparatus of the raptor code, and performs a maximum score search on a matrix of size MxL having the raptor code and blocks the matrix U. Blocked matrix U generator 1010 for generating a transform matrix matrix 1010 and a modified matrix and length for performing a Gaussian elimination on the blocked matrix U to generate a modified matrix having only one '1' in each row and each column. and a vector calculator 1030 that calculates a vector of length L using the vector r.

Here, the blocked matrix U generator 1010 may generate an initial matrix by storing only a position of '1' in each row and each column of the MxL sized matrix having the raptor code to be decoded. In addition, the blocked matrix U generator 1010 may calculate an initial score of each row of the initial matrix and generate a score table having a score as a key for the row whose degree is '2'.

Here, the initial score of each row may be the sum of the degrees of the column at the index of each row, and the blocked matrix U generator 1010 scores the rows of which the degree is '2' when the degree of each row is '2'. You can use as a key to insert the index of the row whose degree is '2' into the score table.

In addition, the blocked matrix U may be generated by adding one or more columns to the matrix U generated by performing a maximum score search on the MxL matrix, and the one or more columns to be added may be the size of the predetermined block based on the number of columns of the matrix U. It can be added at the position of the largest integer smaller than the number divided by.

11 is a flowchart illustrating a method of maximum similarity decoding of a raptor code according to another embodiment of the present invention.

The maximum similarity decoding apparatus 1000 of the raptor code according to the embodiment of the present invention performs a maximum score search on an MxL sized matrix having the raptor code to generate a blocked matrix U (S1110), and blocks the matrix U. A Gaussian elimination method is performed to generate a modified matrix having only one '1' in each row and each column (S1120), and a vector having a length L is calculated using the modified matrix and a vector having a length r (S1130).

As described above, according to an embodiment of the present invention, the maximum similarity decoding of the 3GPP MBMS Raptor code can be performed very quickly. Faster decryption can increase the data transfer speed. In addition, when the maximum similarity decoding is performed in the same manner as in the embodiment of the present invention, the terminal using the battery requires less battery consumption than the conventional decoding method, thereby increasing the battery usage time.

While the present invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiments. In other words, within the scope of the present invention, all of the components may be selectively operated in combination with one or more. In addition, although all of the components may be implemented in one independent hardware, each or all of the components may be selectively combined to perform some or all functions combined in one or a plurality of hardware. It may be implemented as a computer program having a. Codes and code segments constituting the computer program may be easily inferred by those skilled in the art. Such a computer program may be stored in a computer readable storage medium and read and executed by a computer, thereby implementing embodiments of the present invention. The storage medium of the computer program may include a magnetic recording medium, an optical recording medium, a carrier wave medium, and the like.

In addition, the terms "comprise", "comprise" or "having" described above mean that the corresponding component may be included unless otherwise stated, and thus, excludes other components. It should be construed that it may further include other components. All terms, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. Terms used generally, such as terms defined in a dictionary, should be interpreted to coincide with the contextual meaning of the related art, and shall not be interpreted in an ideal or excessively formal sense unless explicitly defined in the present invention.

The foregoing description is merely illustrative of the technical idea of the present invention, and various changes and modifications may be made by those skilled in the art without departing from the essential characteristics of the present invention. Therefore, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention but to describe the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of the present invention.

As described above, the present invention is applied to a field related to an apparatus and a method for decoding a raptor code, which is an error correction code adopted in 3GPP MBMS, to reduce the time required when decoding the raptor code using maximum similarity decoding. As a result, it is possible to increase the data transmission speed and to reduce the power consumption required for decryption.

1 is an exemplary diagram illustrating an MxL matrix that is a result of performing a first phase on matrix A;

2 is an exemplary diagram illustrating an LxL matrix that is a result of performing a second phase on an MxL matrix.

3 is an exemplary view showing a matrix A expressed according to an embodiment of the present invention;

4 is an exemplary diagram illustrating an arbitrary Ridx [k] in a state where decoding is performed according to an embodiment of the present invention;

5 is an exemplary diagram for explaining a process of calculating a value of a score of each row of an initial matrix;

6 is an exemplary view showing a score table according to an embodiment of the present invention;

7 is an exemplary view showing an updated score table according to an embodiment of the present invention;

8 is an exemplary view illustrating a process of searching for a maximum score according to an embodiment of the present invention;

9 is an exemplary diagram showing a blocked matrix U before and after performing Gaussian removal according to an embodiment of the present invention.

10 is a block diagram illustrating a decoder of a raptor code according to an embodiment of the present invention.

<Description of Symbols for Main Parts of Drawings>

1010: Blocked Matrix U Generator 1020: Deformation Matrix Generator

1030: Vector Calculator

Claims (8)

In the maximum similarity decoding apparatus of the raptor code, A blocked matrix U generator that performs a maximum score search on an MxL sized matrix having the raptor code to produce a blocked matrix U; A modified matrix generator for performing a Gaussian elimination method on the blocked matrix U to generate a modified matrix having only one '1' in each row and each column; And A vector calculator for calculating a vector of length L using the transformation matrix and the vector of length r. Maximum similarity decoding apparatus comprising a. The method of claim 1, wherein the blocked matrix U generator, And an initial matrix is generated by storing only a position of '1' in each row and each column of the MxL sized matrix. The method of claim 1, wherein the blocked matrix U generator, A maximum similarity decoding apparatus, comprising: calculating an initial score of each row of an initial matrix and generating a score table having a score as a key for a row having a degree of '2'. The method of claim 3, wherein the initial score of each row, And the sum of the degrees of the columns at the index of each row. 4. The method of claim 3, wherein the blocked matrix U generator, When the degree of each row is '2', the maximum similarity decoding is performed by inserting an index of the row having the degree of '2' into the score table using the score of the row having the degree of '2' as a key. Device. The method of claim 1, wherein the blocked matrix U, And at least one column of the matrix U generated by performing a maximum score search on the MxL matrix. The method of claim 6, wherein the one or more columns, And the maximum similarity decoding apparatus is added to the position of the largest integer smaller than the number divided by the size of the predetermined block. In the maximum similarity decoding method of the raptor code, Generating a blocked matrix U by performing a maximum score search on an MxL sized matrix having the raptor code to produce a blocked matrix U; A modified matrix generation step of performing a Gaussian elimination method on the blocked matrix U to generate a modified matrix having only one '1' in each row and each column; And A vector calculation step of calculating a vector having a length L using the deformation matrix and the vector having a length r. Maximum similarity decoding method comprising a.

Images