Patent Application
20110200149
The present invention relates to a preprocessing apparatus and a method thereof for group-based decoding; and, more particularly, to a preprocessing apparatus and for group-based decoding which improves a convergence speed by allocating a different group according to a log likelihood ratio (LLR) of a received signal unlike a conventional technology, which repeats a decoding operation using a group identically allocated for all received signals.
This work was supported by the IT R&D program of MIC/IITA [2007-S-008-01, “Development of 21 GHz Band Satellite Broadcasting Transmission Technology”].
Lately, among channel coding schemes, a Low Density Parity Check (LDPC) coding scheme is actually being applied to real systems due to its superior performance. The coding scheme using an LDPC code has an advantage of a fast decoding rate because the LDPC code allows easy parallel decoding unlike a turbo code. Although turbo code has been known as a code with good performance, it is difficult to realize a high speed system with a turbo code because of high implementation complexity.
Since a channel code for designing a broadcasting MODEM has a considerably long frame length, it is difficult to apply it to actual systems despite the general advantages of an LDPC code. As LDPC code has an enormous structure, it is necessary to develop a processing structure that allows high-speed decoding in order to design a channel CODEC supporting diverse code rates.
A ‘Sum-Product’ algorithm and a ‘Min-Sum’ algorithm have been introduced for decoding an LDPC coded signal. Such decoding algorithms repeatedly calculate a log likelihood ratio (LLR) for decoding signal. The LLR denotes a probabilistic reliability degree of a received signal.
The ‘Sum-Product’ algorithm has a high calculation cost because it is required to perform calculation using a mathematical function such as a ‘tan h’ function (real number addition and function evaluation). A table having corresponding function values was used to reduce the calculation cost. On the contrary, the ‘Sum-Product’ algorithm has an advantage of a high decoding performance because of accurate calculation.
Unlike the ‘Sum-Product’ algorithm, the ‘Min-Sum’ algorithm uses approximated equations that are equivalent to the mathematical functions of the ‘Sum-Product’ algorithm, for example, addition, minimum, determination of positive and negative, and multiplication of positive and negative. Therefore, the ‘Min-Sum’ algorithm performs simpler calculation than the ‘Sum-Product’ algorithm. However, the decoding performance deteriorates due to the influence of the approximated equations.
The decoding algorithms for an LDPC code repeatedly decode signals based on propagation of probabilistic information. That is, the ‘Sum-Product’ algorithm or the ‘Min-Sum’ algorithm processes all of columns for all bits of a corresponding coded signal after finishing processing all of rows of the coded signal in one repetition time of decoding.
For example, if a check matrix of an LDPC code to decode is a two way m×n matrix H=[Hm,n] where n is an integer larger than 0 smaller than n and m is an integer larger than 0 smaller than M, the decoding algorithm processes rows of all sets (m,n) that satisfies Hm,n=1 in an order of m=1, 2, 3, . . . , M. After processing the rows, the decoding algorithm processes columns of all sets (m,n) that satisfies Hm,n=1 in an order of n=1, 2, 3, . . . , N.
Accordingly, the decoding algorithm has a problem that decoding latency increases in proportion to the repetition times of a decoding process as the repetition number of a decoding process increases.
In order to overcome the decoding latency problem, a Shuffled Belief Propagation (BP) algorithm was introduced to reduce the repetition number of a decoding operation.
The Shuffled BP algorithm calculates and updates probabilistic information one bit by one bit through processing rows and columns of LDPC coded signals. As a result, the propagation of the probabilistic information is effectively performed, and the convergence is advantageously accelerated.
At first, a repetition time i is set to 1 (i=1), a maximum repetition number is set to Imax, and an initial value zm,n(0) of a log likelihood ration (LLR) is set to Fn(zm,n(0):=Fn).
Also, a check matrix H of an LDPC code, which is a target matrix, is a two way m×n matrix H=[Hm,n] where n is an integer larger than 0 and smaller than N and m is an integer larger than 0 and smaller than M. Hm,n denotes an element of a mth row and a nth column of the matrix H.
Then, a row process that calculates Eq. 2 and Eq. 3 is performed with conditions of Eq. 1.
where N(m) and M(n) are subsets of a set[1,N];
N(m) is defined as N(m):={n:Hm,n=1}; and
M(n) is defined as M(n):={m:Hm,n=1}.
That is, N(m) is a set of row indexes having 1 among m rows of a matrix H, and M(n) denotes a set of column indexes having 1 in n columns of a matrix H.
A\a denotes a set obtained by eliminating an element a from a set A. That is, N(m)\n is a set of column indexes with a nth column removed from a set N(m), and M(n)\m is a set of row indexes with a mth row removed from a set M(n).
Zm,n′(i) denotes an LLR updated at an ith repetition, and εm,n(i) denotes a LLR of an ith repetition that is sent from a check node to a variable node.
Then, a column process that calculates Eq. 4 and Eq. 5 is performed with conditions of Eq. 1.
where zm,n(i) denotes a LLR of an ith repetition that is sent from a check node to a variable node; and
zn(i) denotes an after value of the ith repetition.
Then, a hard decision is performed on the after value zn(i), and a decoding series is generated at step 1. Eq. 6 shows the decoding series.
w=[wn] Eq. 6
where Wn denotes elements of a decoding series w from 1 to M.
After the step 1, if Eq. 6 satisfies conditions of Eq. 7, the decoding series of Eq. 6 is outputted. If one of conditions of Eq. 7 is not satisfied, the repetition number i is added by one, and the step 1 is performed again.
Parity Check:OK(H*w=0)
Or
Repetition number is Maximum I=Imax Eq. 7
As described above, the shuffled BP algorithm effectively performs the propagation of probability information by performing the row-process using zm,n′(i) which is an updated LLR at the same ith repetition.
If probability information is calculated and updated one bit by one bit through the row process and the column process of a received signal (Ng=1), it is the shuffled BP algorithm. If probability information is calculated and updated by a unit of predetermined bits (group) through the row process and the column process of a received signal (1<Ng<N), it is a group shuffled BP algorithm.
That is, the group shuffled BP algorithm divides columns of a parity check matrix H used for coding and decoding an LDPC code into several groups and repeatedly performs decoding based on the groups.
In general, the LDPC code can be expressed as a bipartite graph. The bipartite graph expresses the LDPC code with variable nodes, check nodes, and edges connecting the variable nodes and the check nodes.
The group shuffled BP algorithm also performs decoding through updating a probability value between a check node and a variable node in a bipartite graph like LDPC code decoding.
However, the group shuffled BP algorithm characteristically updates the probability value by a predetermined group while updating a probability value from the check node to the variable node.
After a probability value is updated for one group, the updated probability value is used to update a probability value for a next group in order to use further reliable a probability value for a decoding operation, that is, the update operation of the probability value.
That is, a previously updated probability value is not uniformly used in updating a probability value from a check node to a variable node. The previously updated probability value is differently used at each group. The reliability of the probability value between the check node and the variable node is improved using a further reliable probability value for updating a next probability value when the updating of the probability value is repeated. Finally, the performance of a decoder is improved.
Such a group shuffled BP algorithm has an advantage of redwing decoding latency, which is caused by sequentially performing decoding operations, by dividing a variable node into groups and updating messages of variable nodes in each group in parallel.
However, the group shuffled BP algorithm disadvantageously cannot utilize characteristics of a received signal because the group shuffled BP algorithm repeatedly performs decoding operations through groups identically allocated for all of received signals.
That is, the group shuffled BP algorithm has problems of a large computation amount for decoding and slowing down a convergence speed because the group shuffled BP algorithm sequentially groups for all of received signal.
An embodiment of the present invention is directed to providing a preprocessing apparatus and a method thereof for group based decoding, which improve a convergence speed for group-based decoding by calculating log likelihood ratios (LLR) of received signals and sorting and grouping the received signals based on the calculated LLRs or generating a distribution chart of the received signals and grouping the received signals based on the generated distribution chart.
Other objects and advantages of the present invention can be understood by the following description, and become apparent with reference to the embodiments of the present invention. Also, it is obvious to those skilled in the art of the present invention that the objects and advantages of the present invention can be realized by the means as claimed and combinations thereof.
In accordance with an aspect of the present invention, there is provided a preprocessing apparatus for group-based decoding including a log likelihood ratio (LLR) calculator for calculating LLRs of received signals; a signal sorter for sorting the received signals based on the calculated LLRs; and a grouping unit for grouping the sorted signals.
In accordance with another aspect of the present invention, there is provided a preprocessing apparatus for group-based decoding including: an LLR calculator for calculating LLRs of received signals; a distribution generator for generating a distribution chart of the received signals based on the calculated LLRs; and a grouping unit for grouping the received signals using the generated distribution chart.
In accordance with another aspect of the present invention, there is provided a preprocessing method for group-based decoding including: calculating log likelihood ratios (LLR) of received signals; sorting the received signals based on the calculated LLRs; and grouping the sorted signals.
In accordance with another aspect of the present invention, there is provided a preprocessing method for group-based decoding including: calculating LLRs of received signals; generating a distribution chart of the received signals based on the calculated LLRs; and grouping the received signals using the generated distribution chart.
The present invention relates to a group allocation method for improving performance of a grouping shuffled BP algorithm which is a reliability-based decoding method among decoding algorithms of a low density parity check (LDPC) code.
A preprocessing apparatus and a method thereof can improve a convergence speed for group-based decoding by calculating log likelihood ratios (LLR) of received signals and sorting and grouping the received signals based on the calculated LLRs or generating a distribution chart of the received signals and grouping the received signals based on the generated distribution chart.
The advantages, features and aspects of the invention will become apparent from the following description of the embodiments with reference to the accompanying drawings, which is set forth hereinafter.
As shown in
The decoder 200 may use a “Group Shuffled BP” algorithm for decoding.
As shown in
The signal sorter 120 sorts the received signal based on the calculated LLRs of the received signal in an ascending order and enables the decoder 200 to perform a variable node update operation and a check node update operation sequentially from a group of signals having lowest reliability to a group of signals having highest reliability. Here, the reliability is equivalent to the size of LLR. That is, the reliability of low reliable signals is greatly improved according to the characteristics of a ‘tangent hyperbolic’ function. Therefore, a convergence speed is improved in overall.
The grouping unit 130 groups all of sorted signals from the signal sorter 120 at the same ratio according to a predetermined number of groups. For example, each group includes four signals if 12 signals (a, b, c, l) are sorted according to the LLRs and are grouped into four groups.
As shown in
The distribution generator 320 generates a distribution chart of received signals according to the LLRs in order to reduce a calculating amount for sorting the received signals based on the LLRs.
The grouping unit 330 groups the received signals of the distribution chart at the same ratio according to a predetermined number of groups.
Here, it was assumed that decoding rate correctly converge to a transmitted code when the BER is dropped to below 10−5.
As shown in
The graph of
At step S701, log likelihood ratios (LLR) of received signals are calculated.
At step S702, the received signals are sorted based on the calculated LLRs. Here, it is preferable to sort the received signals based on the calculated LLRs in an ascending order.
At step S703, the sorted signals are grouped. Here, all of the sorted signals are grouped at the same ratio according to a predetermined number of groups.
At step S801, the LLRs of received signals are calculated.
At step S802, a distribution chart of the received signals is generated based on the calculated LLRs.
At step S803, the received signals are grouped using the generated distribution chart. Here, it is preferable to sort the received signals of the distribution chart at the same ratio according to a predetermined number of groups.
The method of the present invention described above may be programmed for a computer. Codes and code segments constituting the computer program may be easily inferred by a computer programmer of ordinary skill in the art to which the present invention pertains. The computer program may be stored in a computer-readable recording medium, i.e., data storage, and it may be read and executed by a computer to realize the method of the present invention. The recording medium includes all types of computer-readable recording media.
While the present invention has been described with respect to the specific embodiments, it will be apparent to those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the following claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 10-2007-0132301 | Dec 2007 | KR | national |
| 10-2008-0043580 | May 2008 | KR | national |
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/KR2008/007448 | 12/16/2008 | WO | 00 | 6/17/2010 |
