Source data coding method, coding device and corresponding computer program.
The modified parity matrix H for LDPC codes in 5G communications addresses high decoding complexity by distributing connections and minimizing cycles, enhancing performance and reducing energy consumption.
Patent Information
- Application Number
- FR2023004316
- Authority / Receiving Office
- FR · FR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-04-28
- Publication Date
- 2024-11-01
AI Technical Summary
Existing LDPC codes used in 5G communications face high decoding complexity and energy consumption due to the large number of decoding iterations, which can consume up to 90% of the receiver's resources.
A modified parity matrix H is constructed using a base matrix BG' with a specific structure, distributing connections homogeneously and minimizing short cycles by employing prime number or squared prime number blocks in the extension matrix C, and applying circular rotations to identity matrices, reducing the number of decoding iterations.
This approach significantly reduces the decoding complexity and energy consumption of LDPC decoders, improving performance and efficiency in error correction, especially for smaller code lengths.
Smart Images

Figure 00000000_0000_ABST
Abstract
Description
Title of the invention: Source data coding method, coding device and corresponding computer program. 1. Field of the invention
[0001] The field of the invention is that of digital communications.
[0002] More specifically, the invention relates to error-correcting codes, in particular, but not exclusively, LDPC (Low Density Parity Check) type error-correcting codes.
[0003] The invention finds particular applications in the field of data storage or transmission, for example via wireless communications (for example by radio waves or unguided optical waves) or wired communications (for example by optical fiber or electrical cable).
[0004] In particular, the invention finds applications in all areas where it is sought to offer good transmission reliability, for example for radio transmissions (Wifi®, 5G, 6G ...). 2. Prior art
[0005] Figure 1 illustrates a transmitter in a digital transmission chain. Such a transmitter uses conventional signal processing modules.
[0006] Thus, the source data 11, for example binary data from a video-type data generator (e.g., a source of moving images, virtual or augmented reality images, an image sequence source), audio-type data (e.g., voice), flow control data, whether video, audio, or other, or data from a sensor, actuator, etc., is encoded in a channel coding block 12 to introduce redundancy. The encoded data is then interleaved in an interleaving block 13 to mix the encoded data. A mapping block 14 converts the encoded data into constellation points (BPSK, QPSK, 16QAM, etc.). The associated symbols are framed 15 and modulated in a single-carrier or multi-carrier modulation block 16, for example of the OFDM type (“Orthogonal Frequency-Division Multiplexing”).
[0007] In order to improve the robustness of communications via a communication channel, it is known to use error-correcting codes in the channel 12 coding block. For example, LDPC codes, turbo codes, polar codes, or even Reed-Salomon codes offer good performance in terms of error correction.
[0008] Turbo codes were notably chosen for the 4G mobile telephony standard, and LDPCs for the 5G mobile telephony standard.
[0009] We will now focus on recalling the main characteristics of LDPC codes, as defined in the 5G standard in particular.
[0010] Classically, LDPC codes are defined by three parameters K, M and N; - K corresponds to the size of the source data, for example in the number of useful bits if we consider binary data, - M corresponds to the size of the redundancy data, for example in the number of redundancy bits if we consider binary data, and - N corresponds to the size of the data from the LDC encoder, for example in number of useful bits and redundancy.
[0011] The coding efficiency is classically given by R = K / N, with the relation N = K + M.
[0012] An LDPC code can in particular be defined by a parity matrix H, which gives the parity relations between the source data and the redundancy data.
[0013] An example of a parity matrix H is given in [Fig.2].
[0014] The values equal to "1" in the parity matrix H correspond to the connections (parity equation) between the useful data and the redundancy data. The greater the number of connections (i.e., "1"s in the parity matrix H), the greater the complexity of the decoder will be.
[0015] By way of illustration, following the example of the parity matrix H in Figure 2, the parity equation of the first row gives the first redundancy data r(O)
[0016] r(0) ©6(1) ©6(2) ©b(3) ©6(4) ®b{6) ®b(7) © b(9) ©10) ©b( 11) ©6(12) ©6(13)
[0017] with:
[0018] / X0 = 0 or 1, and
[0019] ® the "exclusive or" operation.
[0020] In the ETSI TS 138 212 V15.2.0 document, the 3GPP standardization consortium defined two basic structures or matrices BG1 and BG2 (“Base Graph” in English) allowing the construction of 104 parity matrices H (52 matrices constructed from the base matrix BG1 and 52 matrices constructed from the base matrix BG2).
[0021] For example, for 5G, the size of the basic matrices BG1 and BG2 is: BG1: K = 22, M = 46, N = 68 BG2: either K = 6, M = 26, N = 32, or K = 8, M = 34, N = 42, or K = 9, M = 38, N = 47, or K = 10, M = 42, N = 52.
[0022] The parity matrices H are obtained from the basis matrices BG1 or BG2, from an expansion factor Z and a circular rotation of a diagonal matrix. Such a construction is known as the "Protograph".
[0023] According to this technique, a parity matrix is obtained by replacing each non-zero element of the basis matrix BG1 or BG2 with a diagonal matrix of size Z x Z to which a circular rotation V has been applied (which is not necessarily the same for all non-nuised elements of the basis matrix).
[0024] One advantage of this family of LDPC encoders is to easily obtain the set of parity matrices H while limiting the use of memory.
[0025] For example, for an expansion factor Z — 8 and a circular rotation V — 2, a a non-zero element of the basis matrix BG1 or BG2 is replaced by the matrix next:
[0026] '0 0 1 0 0 0 0 0' 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 .0 1 0 0 0 0 0 0.
[0027] According to ETSI TS 138 212 V15.2.0, for each basic matrix BG1 or BG2, it is thus possible to construct 52 matrices, classified into eight subfamilies according to the value of the expansion factor Z (between 2 and 384), according to the index "they": Index | Expansion Factor Z | 0 | 2, 4, 8, 16, 32, 128, 256 | 1 | 3, 6, 12, 24, 48, 96, 192, 384 | 2 | 5, 10, 20, 40, 80, 160, 320 | 3 | 7, 14, 28, 56, 112, 224 | 4 | 9, 18, 36, 72, 144, 288 | 5 | 11, 22, 44, 88, 176, 352 | 6 | 13, 26, 52, 104, 208 | 7 | 15, 30, 60, 120, 240
[0028] Thus, it is possible to encode source data blocks of size KZ, ranging from 12 bits (from the basic matrix BG2 with K = 6 and the minimum expansion factor Z — 2) to 8448 bits (from the basic matrix BG1 with K — 22 and the maximum expansion factor Z = 384).
[0029] The choice between using the BG1 or BG2 base matrix depends on the size of the source data K and the desired coding efficiency R. Thus, as illustrated in [Fig. 3], the BG1 base matrix is mainly used for high coding efficiency and large source data sizes. These two base matrices are specifically designed to cover the various uses of 5G. which may require high throughput and / or high reliability (robustness).
[0030] Furthermore, according to the 5G standard, the first two columns of the basic matrix BG1 or BG2 are removed before transmission, which implies that the basic efficiencies are respectively RBG1 = K / (N - 2) = 22 / 66 = 1 / 3 and RBG2 = 1 / 50 = V 5
[0031] It is also noted that the basic matrices BG1 or BG2 used to construct the parity matrix H of an LDPC code according to the 5G standard can be represented in the form of six sub-matrices A, B, O, C, D and I, as described in the document "High Area-Efficient Parallel Encoder with compatible architecture for 5G LDPC codes", Y. Zhu et al., and illustrated in [Fig.4]:
[0032] [AB 0' LC D I.
[0033] with:
[0034] A kernel matrix of size 4 x K
[0035] B a matrix having at least one double diagonal of size 4x4
[0036] C an extension matrix of size (M - 4) x K
[0037] D an extension matrix of size (M - 4) x 4
[0038] 0 a zero matrix of size (N - K-4) x4
[0039] Z an identity matrix of size (Af-4) x (N - K - 4 ) •
[0040] Although LDPC codes are recognized for their ability to guarantee high-speed transmissions, particularly due to the possibility of parallelizing operations at the decoder level, a drawback of these codes is the complexity associated with decoding. Indeed, the greater the number of decoding iterations, the greater the number of operations to be performed at the receiver, and the more resource-intensive they become. The decoding block can thus use up to 90% of the receiver's resources. It can be considered the most energy-intensive part of the physical layer, especially for 5G.
[0041] There is therefore a need for a new coding technique that seeks to reduce the energy consumption of the decoder. 3. Statement of the invention
[0042] The invention proposes a novel solution in the form of a method for encoding at least one block of source data KZ, delivering at least one codeword of size NZ formed from the source data KZ and redundancy data MZ, N = K + M, with Z an integer expansion factor, Z > 1.
[0043] The method according to the invention implements a step of encoding said source data KZ using a parity matrix H of size (MZ XNZ) obtained from a modified basis matrix BGr of size M x N_ by replacing each element of said modified basis matrix BG' by an expansion matrix of size ZxZ,
[0044]
[0045] said modified basis matrix BG' expressed in the form: BG'~ AB 0' CD I.
[0046] with:
[0047] A kernel matrix of size J x K
[0048] B a matrix having at least one double diagonal of size J x J
[0049] C is an extension matrix of size (M - J) x K
[0050] D an extension matrix of size (M - J) x J
[0051] 0 a zero matrix of size (N - K - J) XJ
[0052] Z an identity matrix of size (M -J) X (N - K - J ) .
[0053] According to the invention, said extension matrix C comprises at least two row blocks, each comprising a different number of rows, equal to a prime number or the square of a prime number, each row block comprising at least one occurrence of a diagonal matrix.
[0054] In other words, the extension matrix C comprises at least a first block of El rows comprising one or more diagonal matrices of size El x L1 and a second block of L2 rows comprising one or more diagonal matrices of size L2 x L2, with El E2, El and E2 prime numbers or squared prime numbers.
[0055] The number of lines per block (which also corresponds to the size of the diagonals) is thus chosen from prime numbers (2, 3, 5, 7, 11, 13, etc.) or prime square numbers (4, 9, 25, etc.).
[0056] The aim is thus to avoid the repetition of the same pattern / of a diagonal matrix of the same size across the different row blocks, so as to obtain independence between the rows of the extension matrix C and to avoid, or at least limit, short cycles. The aim is also to distribute the non-nuisance elements of the extension matrix C across the different rows and / or columns of the extension matrix C.
[0057] The proposed solution seeks in particular to improve the performance of error-correcting codes, in particular the LDPC decoder, by reducing the number of decoder iterations and therefore energy consumption.
[0058] In particular, the coded data (code word formed from KZ source data and MZ redundancy data) can be stored in a memory and / or transmitted from a transmitter to a receiver, via a transmission channel.
[0059] According to a particular embodiment, at least one of said row blocks comprises at least two occurrences of said diagonal matrix.
[0060] By choosing the size of the diagonals (for example E1 and E2), it is possible to make Varying the number of connections per line is crucial. It's worth noting that the greater the number of connections (i.e., the number of undamaged elements in the basic matrix, and consequently in the parity matrix H), the greater the decoder's complexity. Therefore, one embodiment seeks to achieve a compromise between the number of connections and decoding performance.
[0061] According to a particular embodiment, for at least one of said row blocks, said extension matrix C comprises a portion of said diagonal matrix, and said extension matrix D comprises the extension of said portion.
[0062] In other words, the matrix formed by said extension matrix C and said extension matrix D includes at least one complete motif representing said diagonal matrix and possibly a partial motif.
[0063] According to a particular embodiment, the number of undamaged elements per column and / or per row relative to the total number of elements per column and / or per row in said kernel matrix A, excluding the first two columns, is between 50 and 100%.
[0064] In particular, the number of undamaged elements per column and / or per row relative to the total number of elements per column and / or per row in said kernel matrix A, excluding the first two columns, is between 75 and 100%.
[0065] The kernel matrix A thus has many connections per row and / or per column, compared to the rest of the modified matrix BG, which allows the source data to be recovered during decoding. For the 5G standard, the kernel matrix A, for example, has on the order of 3 / 4 undamaged elements per row.
[0066] According to a particular embodiment, the first two columns of the kernel matrix A and / or of said extension matrix C comprise an alternation of undamaged and damaged elements per row and per column.
[0067] We thus seek an equitable distribution of connections in the first two columns of the kernel matrix A and / or of said extension matrix C with a staggered structure. As indicated in the preamble, these first two columns may not be transmitted, particularly in 5G.
[0068] According to a particular embodiment J = 4, K = 22 and M — 46.
[0069] The modified basic matrix thus has a format similar to the BG1 matrix with a decomposition into six sub-matrices (A, B, C, D, Q and / ) and a size identical to the basic BG1 matrix. It can therefore be used in any system compatible with the 5G standard.
[0070] According to a particular mode of realization, the core matrix A is equal to : vo vo VI V 3 V 4 V 3 V2 V 1 V 1 V 1 VI V4 VO V4 V 2 V2 V 3 V2 V2 V2 V2 V 3 V 0 V 0 V 3 V 4 V3 in VI V2 V 1 VO V3 V 4 in VI V4 VO V 4 V 0 V2 V 2 VO VO V3 V 4 V3 VI in V4 VI V 2 V 1 V3 V 4 V 0 V 4 V2 V2 V2 VO VO V 0
[0071] ladite matrix B presenting au moins une double diagonale est égale à : VI V 0 vo V 0 V 0 V 0 vo VI vo
[0072] ladite matrix d'extension C est égale à : vo V3 V 0 V 1 V2 V3 V3 VI V 2 V 3 V4 VI VI V 3 V 4 VI VO V 3 V3 V 1 V 2 V3 V 2 VI vo V2 V 3 V4 V 1 V4 V4 V 1 VO V3 VI V 0 V 3 V2 VI V3 V 2 V 1 V4 V 3 VO VV vo VI 4 3 V4 VI V 0 V3 V4 V3 V 2 V 1 V4 V 1 V 4 V3 VO V 3 VO VI V2 V 1 V2 V 3 VI V 3 V4 VI VI V 1 VO VI V3 V 2 V 1 V3 VI V4 V3 VI V 3 VO VI V 1 V2 vo V 3 V4 V4 V 1 VO V4 V 3 V 2 vo V 1 V4 V3 V 3 VO V3 VI V4 V2 V3 VO VI V 1 V2 V2 V 3 V4 V2 V 1 V 0 V3 V 3 V2 vo V 1 V4 V3 V 3 V 0 V4 V 1 V2 V2 V 3 V2 VI VI VO V3 V3 V2 VI VI V 4 V 3 V4 V 0 VI V4 V 2 V3 V2 V 4 V 1
[0073] said extension matrix D is equal to: V4 V 0 V 2 V2 V 0 V 2 V4 VI V 3 V 1 V3 V 2 V 4 vo V3 VI V3 V1 V3 V1 V3 V4 V0
[0074] The empty elements of the above matrices correspond to zero values.
[0075] According to this embodiment: each empty or zero element of said modified basis matrix BG' is replaced by a zero matrix of size Z x Z, Each non-zero element of said modified basis matrix BG' is replaced by a circular permutation matrix obtained by applying a circular rotation Vi to an identity matrix of size ZXZ, such that: Circular rotation Vz if Z belongs to {2,4,8,16,32,128,256}: V0 = 0 + 4y, yg [0:63] Vl=l+4y, yg [0:63] V2=2 + 4y, yG[0:63] V3 = 3 + 4y, ye [0:63] V4 = 0 + 4y, ye [0:63] if Z belongs to {3,6, 12,24,48,96,192,384} V0 = 0 + 5x + 6y, xe [0:1] and yg [0: 63] Vl=l+6y, ye [0:63] V2 = 2 + 6y, ye [0:63] V3 = 3 + 6y, yg [0:63] y4 —4 + 6y; ye [0:63] if Z belongs to {5,10,20,40,80,160,320} V0 = 0+5y, ye [0:63] Vl=l + 5y, yg [0:63] V2 = 2+5y, yg [0:63] V3=3 + 5y, yg [0:63] V4 = 4 + 5y, yg [0:63] if Z belongs to {7,14,28,56,112,224}: VO = 0 + 5x + 7y, xg [0: 1] and yg [0: 31] VI = 1 + 5.x + 7y, xg [0: 1] and y G [0:31] V2 = 2 + 7y, yg [0:31] V3 = 3 + 7y, yg [0:31] V4 = 4+7y, yg [0:31] if Z belongs to {9,18,36,72,144,288}: V0 = 0+5x + 9y, xg[0: 1] and yg [0: 31] VI = 1 + 5x + 9y, e [0 ; 1]^ ye [0:31] V2 = 2 + 5x + 9v xg [0:1] and yg [0:31] V3 = 3 + 5a + 9y, x G [0:1] and yg [0:31] V4 = 4 + 9y, yg [0:31] if Z belongs to {11,22,44,88,176,352}: VO = 0 + 5x + 11 y, xg [0:2] and yg [0:31] VI = 1 + 5x + 1 ly, xe[0:1] and yg[0:31] V2 = 2 + 5x + 1 ly, xg[0:1] and yg[0:31] V3 = 3 + 5x + 1 ly, xg [0:1] and yg [0:31] V4 = 4 + 5x + 1 ly, xg [0:1] and yg [0:31] if Z belongs to {13,26,52,104,208}: V0 = 0+5x + 13v. xg[0: 2]^yG[0:15] VI = 1+ 5x + 13y, xg [0: 2] and yg[0: 15] V2 = 2 + 5x + 13y, xg [0: 2] and yg [0: 15] V3 = 3 + 5x + 13y, xG [0: 1] and yg [0: 15] V4 = 4 + 5x + 13y, xg [0: 1] and yg [0: 15] if Z belongs to {15,30,60,120,240} : V0 = 0 + 5x + 15y. xg [0:2] and yg [0:15] VI = 1+ 5x + 15y, xg [0:2] and yg [0:15] V2= 2+5x+ 15y; xg[0:2] and yg[0:15] V3 = 3 + 5x + 15y, x G[0:2] and y G[0:15] V4 = 4 + 5x+15y, xg[0:2] and ye[0:15]
[0076] with x and y being random variables.
[0077] In another embodiment, the invention relates to a coding device for at least one block of KZ source data, delivering at least one codeword of size NZ formed from said KZ source data and MZ redundancy data, N = K + M, with Z an integer expansion factor, Z > 1,
[0078] comprising at least one processing unit configured to encode said KZ source data using a parity matrix H of size (MZ xNZ), said parity matrix H being obtained from a modified basis matrix BG' of size M x N as described above.
[0079] Such a coding device, also called an encoder, is particularly suitable for implementing the coding process described above. It may, of course, include the various features relating to the process according to the invention, which may be combined or taken individually. Thus, the features and advantages of the encoder are the same as those of the process described above. Consequently, they are not described in further detail.
[0080] The invention further relates to one or more computer programs comprising instructions for implementing a process as described above when this or these programs are executed by at least one processor.
[0081] The invention also relates to a computer-readable information medium, and containing instructions from a computer program as mentioned above. 4. List of figures
[0082] Other features and advantages of the invention will become more apparent from the following description of a particular embodiment, given by way of simple illustrative and non-limiting example, and the accompanying drawings, among which: - [Fig.1] illustrates a transmitter of a conventional digital transmission chain; - Figure 2 shows an example of an H parity matrix for an LDPC code; - [Fig.3] illustrates selection criteria for the basic matrix BG1 or BG2; - [Fig.4] illustrates the decomposition of the basic matrix BG1 or BG2 into sub-matrices; - [Fig.5] presents the main steps implemented by a coding process according to one embodiment of the invention; - Fig. 6A and Fig. 6B illustrate the structure of the BG1 matrix; - [Fig.7A] and [Fig.7B] illustrate the concept of a cycle; - Figures 8 to 15 present different structures of matrices A, C and / or D allowing optimization of the modified basic matrix according to an embodiment of the invention; - Figures 16 and 17 illustrate performance curves of an LDPC decoder implementing a decoding of a code word obtained according to an embodiment of the invention; - [Fig. 18] presents the simplified structure of an encoder implementing a coding process according to an embodiment of the invention.
[0083] 5. Description of an embodiment of the invention
[0084] 5.7 General Principle
[0085] The general principle of the invention is based on a clever distribution of connections (i.e., undamaged elements) in a modified basis matrix BG', aiming to distribute the connections in a substantially homogeneous manner and / or to limit the number of short cycles. Consequently, a parity matrix H obtained from the modified basis matrix BG' has a particular structure. Such a structure makes it possible to improve the performance of error-correcting codes, in particular by reducing the complexity of decoding.
[0086] Figure 5 illustrates the main steps of a coding process according to one embodiment of the invention.
[0087] In a first step 51, we obtain the basis matrix BG ' of size MxN.
[0088] Such a modified basis matrix BG' can be decomposed into six submatrices A, B, O, G, D, I, and expressed in the form:
[0089] [AB 01 BG IC D / 1
[0090] with:
[0091] A kernel matrix of size J x K
[0092] B a matrix having at least one double diagonal of size JXJ
[0093] C is an extension matrix of size (M -J)x K
[0094] D an extension matrix of size (M -J)xJ
[0095] 0 a zero matrix of size (N - K - J) XJ
[0096] / an identity matrix of size (M -J) X (N - K - J)
[0097] In particular, the extension matrix C has a particular structure and comprises at least two row blocks, each comprising a different number of rows, equal to a prime number or the square of a prime number, each row block comprising at least one occurrence of a diagonal matrix.
[0098] In a second step 52, a parity matrix H is obtained from the modified basis matrix BG. To do this, each element of the modified basis matrix BG' is replaced by an expansion matrix of size ZXZ, with Z an integer expansion factor, Z > 1.
[0099] In a third step 53, at least one block of source data KZ is coded using the parity matrix H of size (MZ XNZ), so as to obtain at least one codeword of size NZ formed from the source data KZ and redundancy data MZ, N — K + M.
[0100] Such a code word can in particular be stored in a memory of the encoder or transmitted by a transmitter to a receiver.
[0101] In a particular embodiment, the number of undamaged elements per column in the matrix formed by the kernel matrix A and the extension matrix C, excluding the first two columns, is between 1 / 3 and 1 / 3 + 3, with 1 / 3 an integer. In the matrix formed by the extension matrix C and the extension matrix D, the number of undamaged elements per row, excluding the first two columns, is between -4 and +4, with a an integer.
[0102] In this way, we seek to distribute the number of connections homogeneously per column and / or per row in the modified basic matrix.
[0103] For example, in 5G, fi is between 7 and 10, and the number of non-nuised elements per column in the matrix formed by the kernel matrix A and the extension matrix C, excluding the first two columns, is between 5 and 13, for example between 7 and 10.
[0104] According to another example, the number of undamaged elements per column in the extension matrix C, excluding the first two columns, is between 2 and 8, for example equal to 5 or 6.
[0105] According to yet another example, the number of undamaged elements per column in said kernel matrix A, excluding the first two columns, is between 2 and 4, for example equal to 3.
[0106] For example, in 5G, a is between 5 and 8, and the number of non-nuisance elements per row in the matrix formed by the extension matrix C and the extension matrix D is between 1 and 12, for example between 2 and 9.
[0107] According to another example, the number of unharmed elements per row in the extension matrix C is between 1 and 7.
[0108] In particular, if J ~ 4, K = 22 and M = 46, the modified basis matrix BG' has the same size as the basis matrix BG1, and can be used in any system using the 5G standard.
[0109] 5.2 Basic matrix BG1
[0110] In order to better understand the invention, Figure 6A shows the structure of the basic matrix BG1 as defined in the document ETSI TS 138 212 V15.2.0, formed of matrices A, B, C and D. The sub-matrices ï and O are omitted for the sake of simplification.
[0111] The "1" elements of the basis matrix BG1 represent the connections of the different parity equations. The number of connections per column and per row has been added in [Fig. 6A].
[0112] Thus, the first column of matrices A and C comprises 30 connections, the second column comprises 28 connections, the third column comprises 7 connections, etc.
[0113] The first column of matrices B and D comprises 12 connections, the second column comprises 5 connections, etc.
[0114] The first four rows of matrices A and B each comprise 19 connections.
[0115] The first row of matrices C and D includes 2 connections, the second row includes 7 connections, etc.
[0116] If we disregard the null matrix O and the identity matrix I, the minimum number of connections per row is 2 and the maximum number of connections per row is 19. The minimum number of connections per column is 4 and the maximum number of connections per column is 30. There is therefore a large disparity in the number of connections per row and / or per column in the matrix BG1.
[0117] The kernel matrix A has, proportionally, many connections, i.e., many undamaged elements, because it is the kernel of the BG1 matrix. This allows the decoder, at the receiver, to retrieve this data. It should also be noted that the first two columns of the BG1 matrix, which correspond to the untransmitted data, also contain many "1"s.
[0118] The upper part of the BG1 matrix also has more connections than its The lower section ensures good performance for high yields. This is because the lower part of the BG1 die can be punched to achieve the desired yield. Therefore, it is preferable to select connections in the upper part of the BG1 die.
[0119] By way of example, Figure 6B illustrates the circular rotation factors V for the matrix BG1 of Figure 6A, for an expansion factor Z = 256.
[0120] The parity matrix H is constructed by replacing each non-zero element of the matrix BG1 with an identity matrix of size ZXZ, and applying a circular rotation V according to the factors illustrated in Figure 6B. Thus, the element "1" of the first row / first column of the matrix BG1 in Figure 6A is replaced by an identity matrix of size 256 x 256 to which a circular rotation V = 250 is applied. The element "1" of the first row / second column of the matrix BG1 in Figure 6A is replaced by an identity matrix of size 256 x 256 to which a circular rotation V = 69 is applied, and so on.
[0121] 5.5 Modified Basic Matrix
[0122] The position of the connections in the BG1 basic matrix, as well as the rotation factors to be used, are specified in the 5G 3GPP standard. However, as previously mentioned, it remains desirable to improve the performance of the error-correcting codes, in particular by reducing the complexity of the decoding, for example by reducing the number of connections.
[0123] A modified basis matrix is therefore proposed below, aiming to reduce the complexity of decoding. New connection structures are thus proposed, according to different embodiments of the invention. Depending on the embodiment considered, such structures are efficient and / or easy to reproduce depending on the dimensions of the parity matrix H.
[0124] According to a particular embodiment, the structure of the modified basis matrix is defined so as to have a low disparity in the number of connections per row and / or per column. Indeed, if we consider an LDPC encoder, for example, it should be recalled that LDPC decoding is based on the probabilities of having transmitted a value equal to "1" or "0". It is therefore desirable to distribute the number of connections homogeneously, so that the number of connections in the row, and especially in the column, is substantially constant or evenly distributed.
[0125] According to another particular embodiment, the structure of the modified basis matrix is defined so as to limit the number of short cycles, and if possible to avoid short cycles (known as "Girth"). Indeed, such short cycles limit the performance of error-correcting codes, particularly LDPC codes, especially for small K values, for example on the order of 256 bits or less.
[0126] To better understand this principle, figures 7A and 7B illustrate the phenomena of loops with an expansion factor Z = 4. According to Figure 7A, an expansion factor Z = 4 with the same rotation factor V = 0 is applied to the "1" elements located at positions (i, j), (i + 7, j), (i, j + 7), and (i + 7, j + 7). This results in a loop, illustrated by the arrows. According to Figure 7B, an expansion factor Z = 4 with the same rotation factor V = 0 is applied to the "1" elements located at positions (4j), (z + 7, j), and (f + 7, 7 + 7), and a rotation factor V = 1 is applied to the element located at position (47 + 7). In this case, there is no loop. Changing the rotation factor of one of the matrix elements therefore eliminates loops.
[0127] Different solutions are presented below for obtaining a low disparity in the number of connections per row and / or per column of the modified basic matrix and / or for limiting (or even avoiding) short cycles.
[0128] 5.3.1 Optimization of the kernel matrix A
[0129] According to a first example, it is possible to distribute the connections fairly in the core matrix A, while seeking to have a large number of connections as defined in the 5G standard.
[0130] For example, in the kernel matrix A, the number of undamaged elements per column, excluding the first two columns, is between 1 and J.
[0131] Figure 8 illustrates an example of kernel matrix A, in which the number of undamaged elements per column is constant and equal to 3.
[0132] 5.3.2 Optimization of the first two columns of the kernel matrix A and / or of the extension matrix C
[0133] According to a second example, the first two columns (which are not normally transmitted) can be chosen to ensure a good distribution of connections.
[0134] For example, the first two columns of the kernel matrix A and / or the extension matrix C comprise an alternation of unaffected and affected elements per row and per column. In other words, the connections are staggered, as illustrated in [Fig.9].
[0135] 5.3.3 Optimization of the C extension matrix
[0136] According to a third example, it is possible to distribute the connections equally in the extension matrix C.
[0137] The extension matrix C is generally the largest in terms of dimension. For the basis matrix BG1, for example, the extension matrix C, excluding its first two columns, has a size of 40 x 20.
[0138] According to this third example, the connections are distributed along diagonals to ensure a good distribution of connections.
[0139] For example, if we consider a block of 10 rows and 20 columns, as illustrated In figure 10, it is possible to have two occurrences of a 10 x 10 diagonal matrix on the block of 10 rows, i.e. two row connections and one column connection.
[0140] According to another example, if we consider a block of 5 rows and 20 columns as illustrated in figure 11, it is possible to have four occurrences of a 5x5 diagonal matrix on the 5-row block, i.e., four row connections and one column connection.
[0141] It is therefore possible to vary the number of connections per line according to the size of the diagonals or blocks of lines.
[0142] To guarantee independence between rows, and consequently avoid short cycles, the proposed solution relies on choosing a size for the diagonals that corresponds to prime numbers or squared primes. In particular, we seek to select prime or squared primes that avoid reproducing the same patterns in the extension matrix C,
[0143] Thus, according to the invention, the extension matrix C comprises at least two row blocks, each comprising a different number of rows, equal to a prime number (2, 3, 5, 7, 11, 13 ...) or to the square of a prime number (4 = 22, 9 = 32), each row block comprising at least one occurrence of a diagonal matrix (i.e. at least one complete pattern).
[0144] In this way, all the parity equations are distinct from each other.
[0145] In particular, at least one of the row blocks comprises at least two occurrences of the diagonal matrix (i.e. at least two complete patterns).
[0146] The structure of the extension matrix C can, in particular, be extended into the extension matrix D. Thus, for at least one block of rows, the extension matrix C comprises a portion of a diagonal matrix, and the extension matrix D comprises the extension of the portion of the diagonal matrix. For example, returning to Figure 11, if we consider a block of 5 rows and 20 columns, with the first 18 columns belonging to the extension matrix C and the last 2 columns belonging to the extension matrix D, the extension matrix C comprises three "complete" patterns, each corresponding to the 5x5 diagonal matrix. The fourth pattern is partial and extends into the extension matrix D. The matrix formed by the extension matrices C and D comprises four "complete" patterns, each corresponding to the 5x5 diagonal matrix.
[0147] By way of example, Figure 12 shows an example structure for the extension C and extension D matrices, which can be used in 5G. According to this example, the extension C matrix has a size (M - J) x K = 42 x 22 and the extension D matrix has a size (M - J) x J = 42 x 4. The first two columns of the extension C matrix are not shown.
[0148] Since we seek to select prime numbers that avoid reproducing the same patterns for the extension matrix C, we eliminate the prime numbers 2 and 3 according to this example. For example, it is possible to define a first block of rows with diagonal matrices of size 4, a second block of rows with diagonal matrices of size 5, a third block of rows with diagonal matrices of size 7, a fourth block of rows with diagonal matrices of size 9, a fifth block of rows with diagonal matrices of size 11, etc.
[0149] By noting n the number of a column of the extension matrix C, with n ranging from 1 to ( K - 2 ) (if we do not take into account the first two columns of the extension matrix C), and m the number of a row of the extension matrix C, with m ranging from 1 to (M -J ), the location of the connections can be obtained in a deterministic way.
[0150] For example, for the rows of the extension matrix C, the location of the connections can be obtained by the following algorithm:
[0151] For m ranging from 1 to 4:
[0152] For n ranging from 1 to (K - 2):
[0153] If (n mod 4) = m then BG Impî] = 1
[0154] For m ranging from 5 to 9:
[0155] Pourn ranging from 1 to (K - 2):
[0156] If (n mod 5) = m then BG pHp7] = 1
[0157] For m ranging from 10 to 16:
[0158] For n ranging from 1 to (K - 2):
[0159] If (n mod 7) = m then BG pnpz] — 1
[0160] For m ranging from 17 to 25:
[0161] For n ranging from 1 to (K-2):
[0162] If (n mod 9) = m then BG pnpî] - 1
[0163] For m ranging from 26 to 36:
[0164] For n ranging from 1 to (K - 2):
[0165] If (n mod 11) = m then BG p«pz] = 1
[0166] For m ranging from 37 to (M - J) = 42:
[0167] For n ranging from 1 to (K - 2):
[0168] If (n mod 13) = m then BG pnpï] - 1
[0169] The extension matrix D allows the diagonal matrices to be extended.
[0170] Figure 12 illustrates in particular the number of connections per row and per column for the C extension and D extension matrices.
[0171] It is observed that in the matrix formed by the extension matrix C and the extension matrix D, the number of non-nuised elements per row, excluding the first two columns, is between "-4 and "+4", with a an integer.
[0172] If we place ourselves in the context of 5G, a is for example between 5 and 8.
[0173] In the example illustrated in Figure 12, the number of undamaged elements per line in the matrix formed by extension matrix C and extension matrix D is between 2 and 6.
[0174] The number of non-nuisance elements per column in the extension matrix C is between 5 and 6, and in the extension matrix D is equal to 4.
[0175] Since the lower part of the extension matrix C and the extension matrix D can be punched, the largest number of connections is chosen on the upper part of the extension matrix C and the extension matrix D.
[0176] Another example of optimization of C extension matrices (including the first two columns) and D extension matrices is illustrated in [Fig. 13].
[0177] The proposed structure is slightly different, but the extension matrix C still includes at least two row blocks, each comprising a different number of rows, equal to a prime number (2, 3, 5, 7, 11, 13 ...) or to the square of a prime number (4 = 22, 9 = 32), each row block comprising at least one occurrence of a diagonal matrix (i.e. at least one complete pattern).
[0178] It retains good properties in terms of connection distribution and / or reduction / elimination of short cycles.
[0179] According to the example in [Fig.13], a first block of rows is defined with diagonal matrices of size 4, a second block of rows with diagonal matrices of size 5, a third block of rows with on the first row undamaged elements in positions 1 to 4 (without taking into account the first two columns), on the second row undamaged elements in positions 5 to 8, on the third row undamaged elements in positions 9 to 12, on the fourth row undamaged elements in positions 13 to 16, and on the fifth row undamaged elements in positions 17 to 20, a fourth block of rows with diagonal matrices of size 7, a fifth block of rows with diagonal matrices of size 9, etc.
[0180] 5.3.4 Global optimization of the modified basis matrix BG'
[0181] In the examples above, the extension matrices C (except for the first two columns) and extension matrix D are optimized to reduce short cycles.
[0182] However, if we take into account the matrix formed by the kernel matrix A, double diagonal matrix B, spreading matrix C and extension matrix D proposed in the examples above, short cycles may remain.
[0183] Your short cycles can for example result from the proposed structure for the kernel matrix ri or for the first two columns of the spreading matrix C.
[0184] As illustrated in [Fig.7B], taking into account rotation factors makes it possible in particular to reduce, or even avoid, short cycles.
[0185] It is thus possible to increase the diversity of the modified basic matrix and therefore to reduce the number of short cycles by taking into account the rotation factors.
[0186] In the context of the 5G standard, the expansion factor Z is at least equal to 2. This means that there are at least two possible rotation factors: a non-zero element of the modified basis matrix can be replaced by [10 for a circular rotation V - 0 or by [0 11 for a circular rotation 10 11 11 oJ V = 1.
[0187] Figure 14 illustrates an example of a structure for matrices A, B, C and B taking into account at least two distinct rotation factors, for example an even rotation factor P and an odd rotation factor I.
[0188] In this example, the extension matrix C always includes at least two row blocks, each comprising a different number of rows, equal to a prime number (2, 3, 5, 7, 11, 13 ...) or to the square of a prime number (4 = 22, 9 = 32), each row block comprising at least one occurrence of a diagonal matrix (i.e. at least one complete pattern).
[0189] In addition, a block of lines is associated with at least two distinct rotation factors.
[0190] The structure illustrated in [Fig. 14] makes it possible in particular to reduce the cycles by reorganizing the rotation factors into even values “P” and odd values “I”.
[0191] As illustrated in Figure 15, it is possible to increase the number of rotation factors to increase the diversity of the modified basis matrix BG. For example, five rotation factors, denoted V0 to V4, are chosen. Of course, a different number of rotation factors can be chosen, which leads to a different structure for the matrices A, B, C, and D.
[0192] It is in particular possible to define the equations which define the rotation factors V0 to V4 as a function of the index "they" according to the 5G 3GPP standard. Index “they” Rotation Circular V 0 V0 = 0 + 4y, ys [0:63] Vl=l + 4y, ye [0:63] V2=2 + 4y, ye[0:63] V3 = 3 + 4y, ys [0:63] V4 = 0 + 4y, ye [0:63] 1 vo = 0 + 5x + 6y, xe [0: 1] et ye [0: 63] Vl=l + 6y, ye [0:63] V2 = 24-6y, ye [0:63] V3 = 3 + 6y, ye [0: 63] V4 = 4 + 6y, ye [0:63] 2 V0 = 0 + 5y, ye [0:63] VI = l + 5y, ye [0:63] V2 = 2 + 5y, ye [0:63] V3 = 3 + 5y, ye [0:63] V4 = 4 + 5y, ye [0:63] 3 V0 = 0+5x + 7y, xe[0:1] et ye[0: 31] VI = 1 + 5x + 7y, xe [0:1] et ye [0:31] V2=2 + 7y, ye[0:31] V3=3 + 7y, ye[0:31] V4 = 4 + 7y, ye [0:31] 4 VO - 0+5x + 9y, .X e [0:1] et ye [0: 31] VI = l + 5x + 9y, xe [0: l]e / ye[0:31] V2 = 2 + 5x + 9y, xe [0:1] et ye [0: 31] V3 = 3 4- 5x + 9y, xe [0: 1] et ye [0: 31] V4 = 4 + 9y, ye [0:31] 5 VO = 0 + 5x + 1 ly, xe [0:2] et ye [0: 31] V1 = 14- 5x 4-1 ly, .xe[0:1] and ye[0:31] V2 = 2 + 5x + 1 ly, xe [0:1] and ye [0:31] V3 = 3 + 5x + 1 ly, xe [0:1] and ye [0:31] V4 = 4 + 5x + 1 ly, xe [0:1] and ye [0:31] 6 VO = 0+5x 4- 13y, xe [0: 2] and ye [0:15] VI = 1 + Sx + 13y, xe [0: 2] and ye[0:15] V2 - 2 + 5x + 13y, xe [0: 2] and ye [0: 15] V3 = 3 + 5x 4-13y, xe [0:1] and yes [0:15] V4 = 4 + 5.x + 13y, xe [0:1] and ye [0:15] . 7 VO = O + 5x + 15y, xs [0: 2] and ye [0:15] VI = 1 + 5x + 15y, xe [0: 2] and ye [0: 15] V2 = 2 + 5x + 15y, % e [0: 2] and ye [0: 15] V3 = 3 + 5x + 15y, xe [0 : 2] and ye [0: 15] V4 = 4+ 5x + 15y, xe [0: 2] and ys [0:15]
[0193] with x and 3' random variables whose interval is given in the table above.
[0194] The structure illustrated in Figure 15 makes it possible in particular to reduce short cycles for small matrices, for example for K < 256.
[0195] In particular, it is observed that in the matrix formed by the kernel matrix A and the extension matrix C, the number of unaffected elements per column, excluding the first two columns, is between -3 and +3, where fl is an integer. In the matrix formed by the extension matrix C and the continuation matrix D, the number of unaffected elements per row, excluding the first two columns, is between a - 4 and a + 4, where a is an integer.
[0196] If we place ourselves in the context of 5G, is for example between 7 and 10 and has between 5 and 8. In the example illustrated in figure 15, the number of non-nuisance elements per column in the matrix formed by the kernel matrix A and the extension matrix C is between 7 and 10, and the number of non-nuisance elements per row in the matrix formed by the extension matrix C and the extension matrix D is between 2 and 9.
[0197] The number of connections per row and / or per column is therefore quite homogeneous.
[0198] 5.3.5 Obtaining a parity matrix
[0199] From the modified basis matrix BG' thus obtained, it is possible to obtain a parity matrix H, by replacing each empty or zero element of the modified basis matrix BG' with a zero matrix of size ZXZ, and each non-zero element of the modified basis matrix BG with a circular permutation matrix obtained by applying a circular rotation Vi to an identity matrix of size ZXZ, !E [0,1, ...}•
[0200] The parity matrix H thus obtained can be used by an encoder to encode at least one block of KZ source data.
[0201] 5.3.6 Performance Curves
[0202] The decoding performance obtained in terms of error rate (ER) as a function of the signal-to-noise ratio (SNR in dB) for LDPC decoders, after encoding source data using a parity matrix obtained either from the basic matrix BG1 such that
[0203]
[0204]
[0205]
[0206]
[0207]
[0208]
[0209]
[0210]
[0211] defined by 3GPP for the 5G standard, either from a modified basic matrix BG' according to an embodiment of the invention (having the same size as the basic matrix BG1 for comparison). The performance obtained for code lengths of 330 bits (K = 22, Z = 15) is a code efficiency of 1 / 3 for [Fig. 16] and 3 / 4 for [Fig. 17]. The decoder algorithm in this example is MinSum, as described in the document "A fully parallel LDPC decoder architecture using probabilistic minsum algorithm for high-throughput applications", C.-C. Cheng et al., IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 61, no. 9, pp. 2738-2746, Sep. 2014, with an alpha parameter equal to 0.7. Curves 161 and 171 respectively illustrate the decoding performance using the modified basis matrix BG' after 20 decoder iterations. Curves 162 and 172 respectively illustrate the decoding performance using the basis matrix BG1 after 20 decoder iterations. Curves 163 and 173 respectively illustrate the decoding performance using the modified basis matrix BG' after 5 decoder iterations. Curves 164 and 174 respectively illustrate the decoding performance using the basis matrix BG1 after 5 decoder iterations. It is noted that the proposed structure, according to at least one embodiment of the invention, improves the performance of the LDPC decoder according to the 5G standard. The difference is even greater with a smaller number of decoder iterations. 5.3.7 Variants We have described above an example in which the modified basis matrix BG' has the same size as the basis matrix BG1, with J = 4, K = 22 and M = 46. This makes it possible to use the modified basis matrix BG' in any communication system according to the 5G standard. However, this is just one example, and other values for J, K and M can be considered, especially for other transmission standards (Wifi®, 6G, etc.). Similarly, a modified basis matrix BG' with five rotation factors V0 to V4 has been proposed. However, a different number of rotation factors can be chosen. We have also given examples with a matrix B of the form ' 1 0 0' 1 0 1 1 0 1. 0 Other forms of matrix B with at least one double diagonal can be used. In this case, the position of the connections and / or the values of the factors of rotations can be updated to take into account the structure of matrix B.
[0212] An LDPC code was also taken as an example. The invention can also be extended to other error-correcting codes using a parity matrix.
[0213] 5.4 Coding device
[0214] Finally, in relation to [Fig.18], we present the simplified structure of an encoder according to at least one embodiment described above.
[0215] As illustrated in [Fig. 18], an encoder comprises at least one memory 181, at least one processing unit 182, equipped for example with a programmable computing machine or a dedicated computing machine, for example a processor P, and controlled by the computer program 183, implementing steps of the coding process according to at least one embodiment of the invention.
[0216] At initialization, the code instructions of the computer program 183 are, for example, loaded into a RAM memory before being executed by the processor of the processing unit 182.
[0217] The processor of the processing unit 182 implements steps of the coding process described above, according to the instructions of the computer program 183, to encode the source data KZ using a parity matrix H of size (MZ XNZ), and deliver at least one codeword of size NZ formed from the source data KZ and redundancy data MZ.
Claims
Claims
1. Method for coding at least one block of KZ source data, delivering at least one code word of size NZ formed from said KZ source data and MZ redundancy data, N — K + M, with Z an integer expansion factor, Z > 1, said method implementing a step of coding (53) said KZ source data using a parity matrix H of size (MZxNZ), characterized in that said parity matrix H is obtained (52) from a modified base matrix BG' of size M x N, by replacing each element of said modified base matrix BG' with an expansion matrix of size ZXZ,said modified base matrix BG ' being expressed in the form: [AB 01 SG =lc D / J with: A a kernel matrix of size J x KB a matrix having at least one double diagonal of size J x JC an extension matrix of size (M - J) x KD an extension matrix of size (M - J) x J 0 a zero matrix of size (N- K - J) XJ / an identity matrix of size (M - J) X (N - K - J) and in that said extension matrix C comprises at least two blocks of rows, each comprising a different number of rows, equal to a prime number or to the square of a prime number, each block of rows comprising at least one occurrence of a diagonal matrix.,
2. Coding method according to claim 1, characterized in that at least one of said blocks of lines comprises at least two occurrences of said diagonal matrix.
3. Method according to any one of the preceding claims characterized in that for at least one of said line blocks, said extension matrix C comprises a portion of said diagonal matrix, and in that said extension matrix D comprises the extension of said portion.
4. Method according to any one of the preceding claims, characterized in that the number of undamaged elements per column and / or per row relative to the total number of elements per column and / or per row
5.
6.
7.
8. in said kernel matrix A, excluding the first two columns, is between 50 and 100%. Method according to any one of the preceding claims, characterized in that the number of unharmed elements per column and / or per row relative to the total number of elements per column and / or per row in said kernel matrix A, excluding the first two columns, is between 75 and 100%. Method according to any one of the preceding claims, characterized in that said first two columns of the kernel matrix A and / or of said extension matrix C comprise an alternation of non-harmed elements and harmed elements per row and per column. Method according to any one of the preceding claims, characterized in that J = 4, K = 22 and M — 46. Method according to any one of the preceding claims, characterized in that said kernel matrix A is equal to: vo V 0 V 1 V 3 V 4 V 3 V 2 V 1 V 1 V 1 V 1 V 4 V 0 V 4 V 2 V 2 V 3 V 2 V2 V 2 V 2 V 3 V 0 V 0 V 3 V 4 V 3 V 0 V 1 V 2 V 1 V 0 V 3 V 4 V 0 VI V 4 V 0 V 4 V 0 V 2 V 2 V 0 V 0 V 3 V 4 V 3 V 1 VO V 4 V 1 V 2 V 1 V 3 V 4 V 0 V 4 V 2 V 2 V 2 V 0 V 0 V 0 ladite matrix B presenting au moins une double diagonale est égale à : VI V 0 vo V 0 V 0 V 0 vo VI vo ladite matrice d'extension C est égale à : in VVVVV 3 0 1 2 3 V3 V 1 V 2 V 3 V 4 V 1 VI V 3 V 4 V 1 V 0 V 3 V3 V 1 V 2 V 3 V 2 V 1 vo V 2 V 3 V 4 V 1 V4 V 4 V 1 V 0 V 3 VI V 0 V 3 V 2 V 1 V3 V 2 V 1 V 4 V 3 vo V 4 V 3 V 0 V 1 V4 V 1 V 0 V 3 V4 V 3 V 2 V 1 V4 V 1 V 4 V 3 VO V 3 V 0 V 1 V2 V 1 V 2 V 3 VI V 3 V 4 V 1 VI V 1 V 0 VI V 3 V 2 V 1 V3 VVV 1 4 3 VI V 3 V 0 VI V 1 V 2 vo V 3 V 4 V4 V 1 V 0 V4 V 3 V 2 vo V 1 V 4 V3 V 3 V 0 V3 V 1 V 4 V2 V 3 V 0 VI V 1 V 2 V2 V 3 V 4 V2 V 1 V 0 V3 V 3 V 2 VO V 1 V 4 V3 V 3 V 0 V4 V 1 V 2 V2 V V4 V 0 V 2 V2 V 0 V 2 V4 VI V 3 V 1 V3 V 2 V 4 vo V3 VI V 3 V 1 V3 V 1 V3 V4 V 0 and in that: each empty or zero element of said modified basis matrix BG' is replaced by a zero matrix of size Z x Z, each non-zero element of said modified basis matrix BG' is replaced by a circular permutation matrix obtained by applying a circular rotation Vi to an identity matrix of size Z x Z, such that: Circular rotation Vi if Z belongs to {2,4,8,16,32,128,256} V0 = 0 + 4y,yg [0:63] Vl=l + 4y,yg [0:63] V2 = 2 + 4y,yg [0:63] V3 = 3 + 4y,yg [0:63] V4 = 0 + 4y,y G [0:63] if Z belongs to {3,6,12,24,48,96,192,384} VO = 0 + 5x + 6y, xg [0:1] and yg [0:63] Vl=l + 6y, yg [0:63] V2 = 2 + 6y, yg [0:63] V3 = 3+6y, yg [0:63] V4 = 4 + 6>; yg [0:63] if Z belongs to {5,10,20,40,80,160,320} V0 = 0 + 5y, yg [0:63] Vl=l + 5y, yg [0:63] V2 = 2+5y, yg [0:63] V3=3 + 5y, yg [0:63] V4 = 4 + 5y, yg [0:63] if Z belongs to {7,14,28,56,112,224} VO = 0 + 5.x + 7y, xe [0: 1] and yc [0:31] VI = l + 5x + 7y, xe[0: l]^yG[0:31] V2 = 2 + 7y, yg [0:31] V3 = 3 + 7y, yg [0:31] V4-4 si + Z [0:3ent, yg] {9,18,36,72,144,288} VO — 0 + 5x + 9y, xg [0 : 1] and yg [0: 31] Vl = l+5x+9y, xe [0: l]etye[0:31] V2 = 2 + 5x + 9y [0:] = 3 [0 g 5x + 9y, xg [0: 1] and yg [0 : 31] V4 = 4 + 9y, yg [0:31] if Z belongs to {11,22,44,88,176,352} VO = 0+5x +1 ly, xe [0:21] + ly] x 5,x0:3 [0: 1] and yg [0: 31] V2 - 2 + 5x +1 ly, xe [0:1] et ye [0:31] V3-3 + 5x+ lly, xe[0:1] etye[0:31] V4 = 4+5x + 1 ly, xe [0:1] et ye [0:31]. if Z belongs to {13,26,52,104,208}: V0 - 0 + 5x + 13y, .xg [0:2] and yg [0:15] VI = 1 + 5.x + 13y, xg [0:2] and yg [0:15] V2 = 2+5x+ 13y, xg [0:2] and yg [0:15] V3 = 3+5.x + 13y, xg [0:1] and yg [0:15] V4 = 4+5.x + 13¾ xg [0:1] and ye [0:15] if Z belongs to {15,30,60,120,240}: V0 = 0+5x+15y, xe [0:2] and yg [0:15] Vl = l + 5x+ 15); xG[0:2]ef yg[0:15] V2 = 2 + 5.x + 15y, xg [0:2] and yg [0:15] V3 = 3 + 5x + 15y. xg [0:2] and yg [0:15] V4-4 + 5x+l5y, xg [0:2] and yg [0:15]
9. with x and 3' random variables. Device for coding at least one block of KZ source data, delivering at least one code word of size NZ formed from said KZ source data and MZ redundancy data, N — K + M, with Z an integer expansion factor, Z > 1, comprising at least one processing unit configured to code (53) said KZ source data using a parity matrix H of size (MZxNZ), said parity matrix H being obtained (52) from a modified base matrix BG' of size M x N, by replacing each element of said modified base matrix BG' by an expansion matrix of size ZXZ, said modified base matrix BG' being expressed in the form: [AB 0] BG =l _ , with : Has a kernel matrix of size J xK B a matrix with at least one double diagonal of size JXJC an extension matrix of size (M - J) x K From an extension matrix of size (M - J) x J 0 a zero matrix of size (N- K - J) xJ / an identity matrix of size ( M - J ) X ( N - K - J ) said extension matrix C comprising at least two blocks of rows, each comprising a different number of rows, equal to a prime number or to the square of a prime number, each block of rows comprising at least one occurrence of a diagonal matrix.
10. Computer program comprising instructions for implementing a method according to any one of claims 1 to 8 when this program is executed by a processor.
Citation Information
Patent Citations
Information processing method, apparatus and communication device
EP3582398A1
Low density parity check decoder, electronic device, and method therefor
WO2021233788A1
