Source data coding method, coding device and corresponding computer program.
By optimizing the parity matrix H with a structured base matrix BG' that evenly distributes connections and minimizes short cycles, the decoding complexity and energy consumption of LDPC codes are reduced, improving error correction efficiency in 5G communications.
Patent Information
- Application Number
- FR2023004320
- 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, reducing the number of decoding iterations and energy consumption by optimizing the kernel and extension matrices A, B, C, and D.
The proposed method significantly reduces the decoding complexity and energy consumption of LDPC codes, enhancing the performance of error correction while maintaining compatibility with the 5G standard.
Smart Images

Figure 00000000_0000_ABST
Abstract
Description
Title of the invention: Method for encoding source data, corresponding encoding device and 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, error correcting codes of the LDPC type (in English “Low Density Parity Check”).
[0003] The invention finds applications in particular 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 electric cable).
[0004] In particular, the invention finds applications in all fields where it is sought to offer good transmission reliability, for example for radio transmissions (Wifi®, 5G, 6G, etc.). 2. Prior art
[0005] [Fig.l] illustrates a transmitter of a digital transmission chain. Such a transmitter uses conventional signal processing modules.
[0006] Thus, the source data 11, for example binary data coming from a source generating video type data (for example a source of animated images, virtual or augmented reality images, an image sequence source), audio type (for example voice), flow control type whether video or audio or other, or coming from a sensor, an actuator, etc., are coded in a channel coding block 12 to introduce redundancy. The coded data are then interleaved in an interleaving block 13 to mix the coded data. A mapping block 14 makes it possible to convert the coded data into constellation points (BPSK, QPSK, 16QAM, etc.). The associated symbols are placed in frame 15 and modulated in a single-carrier or multi-carrier modulation block 16, for example of the OFDM (“Orthogonal Frequency-Division Mul-tiplexing”) type.
[0007] In order to improve the robustness of communications via a communication channel, it is known to use error-correcting codes in the channel coding block 12. For example, LDPC codes, turbo codes, polar codes, or even Reed-Salomon codes offer good performance in terms of error correction.
[0008] Turbo codes have been chosen in particular for the 4G mobile telephony standard, and LDPCs for the 5G mobile telephony standard.
[0009] We will now recall the main characteristics of LDPC codes, as defined in the 5G standard in particular.
[0010] Conventionally, LDPC codes are defined by three parameters K, M and N; - K corresponds to the size of the source data, for example in number of useful bits if we consider binary data, - M corresponds to the size of the redundancy data, for example in 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 relationships 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. of “1” in the parity matrix H), the greater the complexity of the decoder will be.
[0015] As an illustration, following the example of the parity matrix H of figure 2, the parity equation of the first line 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 basic matrix BG1 and 52 matrices constructed from the basic 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 basic matrices BG1 or BG2, from an expansion factor Z and a circular rotation of a diagonal matrix. Such a construction is known as "Protograph".
[0023] According to this technique, a parity matrix is obtained by replacing each non-zero element of the basic matrix BG1 or BG2 by 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 unaffected elements of the base matrix).
[0024] An advantage of this LDPC encoder family is to easily obtain the set of H parity matrices while limiting memory usage.
[0025] For example, for an expansion factor Z — 8 and a circular rotation V — 2, a non-zero element of the base matrix BG1 or BG2 is replaced by the matrix next:
[0026] '0 0 1 0 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 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 the ETSI TS 138 212 V15.2.0 document, for each basic matrix BG1 or BG2, it is thus possible to construct 52 matrices, classified into eight sub-families according to the value of the expansion factor Z (between 2 and 384), according to the index “ils”: Index ils 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 basic matrix BG1 or BG2 is made according to the size of the source data K and the desired coding rate R. Thus, as illustrated in [Fig.3], the basic matrix BG1 is mainly used for high coding rates and large sizes of the source data. These two basic matrices are notably designed to cover the different 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 deleted before transmission, which means that the basic efficiencies are respectively RBGl = 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] Has 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] From 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, in particular thanks to the possibility of parallelizing operations at the decoder level, a disadvantage of these codes is the complexity linked to decoding. Indeed, the greater the number of decoding iterations, the greater the number of operations to be carried out in reception and the more resource-consuming they are in terms of resources. The decoding block can thus use up to 90% of the receiver's resources. It can be considered the most energy-consuming part of the physical layer, particularly for 5G.
[0041] There is therefore a need for a new coding technique seeking to reduce the energy consumption of the decoder. 3. Statement of the invention
[0042] The invention proposes a new solution in the form of a method for coding at least one block of KZ source data, delivering at least one code word of size NZ formed from the KZ source data and MZ redundancy data, N = K + M, with Z an integer expansion factor, Z > 1.
[0043] The method according to the invention implements a step of coding said KZ source data using a parity matrix H of size (MZ XNZ) obtained from a modified basic matrix BGr of size M x N_ by replacing each element of said modified base matrix BG' by an expansion matrix of size ZxZ,
[0044]
[0045] said modified basic matrix BG' being expressed in the form: BG'~ AB 0' CD I.
[0046] with:
[0047] Has 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 an extension matrix of size (M - J) x K
[0050] From 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 blocks of lines, each comprising the same number of lines, each block of lines comprising a diagonal matrix.
[0054] In other words, the extension matrix C comprises at least two blocks of ZI rows each comprising a diagonal matrix of size IA x IA, and preferably a single diagonal matrix of size El x Ll.
[0055] For example, El = 4 if K = 6, El = 6 if K = 8, El = 7 if K = 9, IA - 8 if K — 10
[0056] By choosing the size of the diagonals, it is possible to vary the number of connections per line, and in particular to have only one connection per line for the extension matrix C, excluding the first two columns. It is recalled that the greater the number of connections (i.e. unaffected elements in the basic matrix, and consequently in the parity matrix H), the greater the complexity of the decoder will be. We therefore seek, according to one embodiment, to achieve a compromise between the number of connections and decoding performance.
[0057] The use of diagonal matrices thus makes it possible to distribute the unaffected elements of the extension matrix C over the different rows and / or columns of the extension matrix C.
[0058] The proposed solution seeks in particular to improve the performance of error-correcting codes, in particular of the LDPC decoder, by reducing the number of iterations of the decoder and therefore the energy consumption.
[0059] In particular, the coded data (code word formed from the 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.
[0060] In particular, the number of rows of matrices C and D depends on the value of E, the basic efficiency (1 / 5 for BG2) and the number of punched columns.
[0061] In the general case, N = K + M and the efficiency R — K / N. For 5G, the yield R = K / (TV - 2) with the first two columns punched.
[0062] Si = 6, M ~ 4xK + 2 and N — 5xK + 2.
[0063] It is thus possible to calculate the size of matrices C and D; if the kernel matrix A has a size of J x K, then the size of the extension matrix C is (M - J) x K and the size of the extension matrix is (M - J) x J. For 5G, J -4.
[0064] According to a particular embodiment, said extension matrix D comprises at least two blocks of lines, each comprising the same number of lines, each block of lines comprising a diagonal matrix.
[0065] In other words, the extension matrix D comprises at least two blocks of L2 rows each comprising a diagonal matrix of size L2 x L2, with L1 = L2 or El * E2, and preferably a single diagonal matrix of size E2 x L2.
[0066] For example, L2 = 4 whatever the value of K.
[0067] The use of diagonal matrices makes it possible to distribute the unaffected elements of the extension matrix D over the different rows and / or columns of the extension matrix D.
[0068] According to a particular embodiment, the number of unaffected 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%.
[0069] In particular, the number of unaffected 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%.
[0070] The kernel matrix A thus has many connections per row and / or per column, compared to the rest of the modified matrix BG', which makes it possible to find the source data during decoding. For the 5G standard, the kernel matrix A has, for example, around 3K / 4 undamaged elements per row.
[0071] 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 non-harmful elements and harmed elements per row and per column.
[0072] We are thus seeking an equitable distribution of the connections in the first two columns of the core 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.
[0073] According to a particular embodiment 2 = 4, and - K = 6 and M = 26, or ^ = 8 and M = 34, or K = 9 and M = 38 or K = 10 and M = 42.
[0074] The modified basic matrix thus has a format similar to the BG2 matrix with a decomposition into six sub-matrices (A, B, C, D, 0 and Z) and a size identical to the BG2 basic matrix. It can therefore be used in any system compatible with the 5G standard.
[0075] According to a particular embodiment, depending on the value of K, said kernel matrix A is equal to: for K= 10 for K=9 for K=8 for K=6 VI VO V3 V2 VI V4 V3 V2 V4 VO V3 V2 V4 V2 V3 vo V2 VO V3 V4 V2 V3 VI VI V3 V4 V4
[0076] said matrix B having at least one double diagonal is equal to: VO vo vo vo VI vo vo vo vo
[0077] said extension matrix C is equal to: for K= 10 for K=9 for K=8 for K=6 V4 vo V2 VI V3 V2 VI V3 V2 V4 VO VO VI VI V2 V2 vo VI VI V2 V4 V3 VI V4 V2 VO vo VI VI V2 V4 V3 V2 V2 V3 VO V4 VI VO V4 VI vo V2 V2 V3 VO V4 VI V3 V3 V4 VI VO V2 VI vo V2 VI V3 VI V4 V2 VO V4 V4 V2 VO V4 VI vo V2 VI V3 vo V4
[0078] said extension matrix D is equal to:
[0079] The empty elements of the above matrices correspond to zero values.
[0080] The matrices AC and D have a nested structure depending on the value of K.
[0081] Thus, for K = 6, the kernel matrix A(K = 6) has a size J x K, or 4x6 according to the example presented. For K = 8, the kernel matrix A (K = 8) has a size .7 x K, or 4x8 according to the example presented. It is formed from the kernel matrix A(K — 6) to which two columns are added. For K = 9, the kernel matrix A (K = 9) has a size J x K, or 4x9 according to the example presented. It is formed from the kernel matrix A(K = 8) to which one column is added. For K = 10, the kernel matrix A (K = 10) has a size J x K, or 4x10 according to the example presented. It is formed from the kernel matrix A(K = 9) to which one column is added.
[0082] Similarly, for K — 6, the extension matrix C (K = 6) has a size ( M - J ) x K, or 22x6 according to the example presented. For K — 8, the extension matrix C(K = 8) has a size ( M - J) x K, or 30 x 8 according to the example presented. It is formed from the extension matrix C( K = 6) to which two columns and eight rows are added. For K = 9, the extension matrix C(K = 9) has a size ( M - J ) x K, or 34 x 9 according to the example presented. It is formed from the extension matrix C(K = 8) to which one column and four rows are added. For K — 10, the extension matrix C( K = 10) has a size ( M - J) X K, or 38x10 according to the example presented. It is formed from the extension matrix C(K = 9) to which one column and four rows are added.
[0083] Finally, for K = 6, the extension matrix D ( K = 6 ) has a size ( M - J ) xJ, or 22 x 4 according to the example presented. For K = 8, the extension matrix D(K = 8) has a size [M -J) xJ, or 30 x 4 according to the example presented. It is formed from the extension matrix D ( K = 6 ) to which eight lines are added. For K = 9, the extension matrix D(K = 9) has a size ( M - J) x J, or 34 x 4 according to the example presented. It is formed from the extension matrix D(K = 8) to which four lines are added. For K = 10, the extension matrix D( K = 10) has a size (M - J) x J, or 38 x 4 according to the example presented. It is formed from the extension matrix D( K = 9) to which four lines are added.
[0084] According to this embodiment: - each empty or null element of said modified base matrix BG' is replaced by a null matrix of size Z x Z, - each non-zero element of said modified base 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, ye [0:63] VI = l + 4y, ye [0:63] V2 = 2 + 4y, ye [0:63] V3 = 3 + 4y, yg [0:63] V4 = 0 + 4y, ye [0:63] if Z belongs to {3,6 12,24,48,96,192,384} V0 = 0 + 5x + 6y, % g [0:1] and ye [0: 63] VI = l+6y, ye [0:63] V2 = 2 + 6y; ye [0:63] V3 = 3 + 6y, yg [0:63] V4 = 4 + 6y, yg [0:63] if Z belongs to {5,10,20,40,80,160,320} V0 = 0+5y, yg [0:63] Vl=l + 5y, ye [0:63] V2 = 2 + 5y, ye [0:63] V3 = 3 + 5y, yg [0:63] V4 = 4 + 5y, ye [0:63] if Z belongs to {7,14,28,56,112,224} : VO = 0 + 5x + ly, xg [0: 1] and yg [0: 31] VI = 1 + 5x + ly, xg [0: 1] and yg [0: 31] V2 = 2 + 7y, ye [0: 31] V3 = 3 + 7y, ye [0: 31] V4 = 4 + 7y, yg [0: 31] if Z belongs to {9,18,36,72,144,288} : V0 = 0 + 5x + 9y, .xe [0: 1] and ye [0: 31] V1 = 1 + 5x + 9y, xg [ 0: 1 ] and yg [0: 31] V2 = 2 + 5.x + 9y, xe [0:1] and ye [0: 31] V3 = 3 + 5x+9y, xg[0: l]ezye[0:31] V4 = 4 + 9y, yG[0:31] if Z belongs to {11,22,44,88,176,352}: V0 = 0 + 5x + 1 ly, x G [0:2] and ye [0:31] VI = 1 + 5x + 1 ly xg [0: 1] and yg [0: 31] V2 = 2 + 5x + 1 ly xg [0:1] and ye [0: 31] V3 = 3 + 5x + 1 ly, xg [0: 1] and yg [0: 31] V4 = 4 + 5x + 1 ly, xe [0:1] and ye [0:31] if Z belongs to {13,26,52,104,208} : VO = 0+ 5x + 13y xg [0:2] and 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]éOg[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, xe [0: 2] and yg [0:15] V4 = 4 +5x+15y, xG[0:2] and yg[0:15].
[0085] with x and random variables.
[0086] In another embodiment, the invention relates to a 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,
[0087] comprising at least one processing unit configured to encode said KZ source data using a parity matrix H of size ( A£Z x NZ ), said parity matrix H being obtained from a modified basis matrix BG' of size M x N as previously described.
[0088] Such a coding device, also called an encoder, is particularly suitable for implementing the coding method described above. It may of course include the various characteristics relating to the method according to the invention, which may be combined or taken in isolation. Thus, the characteristics and advantages of the encoder are the same as those of the method described above. Consequently, they are not detailed further.
[0089] The invention also relates to one or more computer programs comprising instructions for implementing a method as described above when this or these programs are executed by at least one processor.
[0090] The invention also relates to a computer-readable information medium, comprising instructions of a computer program as mentioned above. 4. List of figures
[0091] Other characteristics and advantages of the invention will appear more clearly on reading the following description of a particular embodiment, given as a simple illustrative and non-limiting example, and the appended drawings, among which: - [Fig.l] illustrates a transmitter of a conventional digital transmission chain; - Figure 2 shows an example of a parity matrix H 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 method according to an embodiment of the invention; - Figures 6A and 6B illustrate the structure of the BG1 matrix; - figures 7A and 7B illustrate the concept of cycle; - Figures 8 to 13D show different structures of matrices A, C and / or D allowing to optimize the modified basic matrix according to an embodiment of the invention; - [Fig. 14] shows the simplified structure of an encoder implementing a coding method according to one embodiment of the invention.
[0092] 5. Description of an embodiment of the invention
[0093] 5.7 General principle
[0094] The general principle of the invention is based on a clever distribution of the connections (i.e. of the unharmed elements) in a modified basic matrix BG\ seeking 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 basic 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.
[0095] [Fig.5] illustrates the main steps of a coding method according to one embodiment of the invention.
[0096] During a first step 51, the basic matrix BG ' of size M^N is obtained.
[0097] Such a modified basis matrix BG ' can be decomposed into six sub-matrices A , B, 0, C, D, Z, and expressed in the form:
[0098] [AB 01 BG=ïc D / 1
[0099] with:
[0100] Has a kernel matrix of size J x K
[0101] B a matrix having at least one double diagonal of size JXJ
[0102] C an extension matrix of size (M - J) x K
[0103] From an extension matrix of size (M - J) x J
[0104] 0 a zero matrix of size (N - K -J) XJ
[0105] Z an identity matrix of size ( M -J) X ( N - K - J)
[0106] In particular, the extension matrix C has a particular structure and comprises at least two blocks of rows, each comprising the same number of rows, each block of rows comprising a diagonal matrix, and preferably a single complete diagonal matrix.
[0107] During a second step 52, a parity matrix H is obtained from the modified base matrix BG. To do this, each element of the modified base matrix BG' is replaced by an expansion matrix of size Z x Z, with Z an integer expansion factor, Z > 1.
[0108] During a third step 53, at least one block of KZ source data is coded using the parity matrix H of size (MZ X NZ), so as to obtain at least one code word of size NZ formed of the KZ source data and MZ redundancy data, N — K + M.
[0109] Such a code word may in particular be stored in a memory of the encoder or transmitted by a transmitter to a receiver.
[0110] In a particular embodiment, the number of unharmed elements per column in said kernel matrix A and said extension matrix C, excluding the first two columns, is between / 7- 3 and fi + 3, with fi an integer. The number of unharmed elements per row in said extension matrix C and said extension matrix Z), excluding the first two columns, is between "- 2 and a + 2, with a an integer.
[0111] In this way, we seek to distribute the number of connections per column and / or per row in the modified basic matrix evenly.
[0112] For example, in 5G, is between 7 and 10, and the number of undamaged elements per column in the matrix formed from the core matrix A and the extension matrix C is between 4 and 13, for example between 7 and 8.
[0113] According to another example, the number of unaffected elements per column in the extension matrix C is between 2 and 8, for example equal to 4 or 5.
[0114] According to yet another example, the number of unaffected elements per column in said kernel matrix A is between 2 and 4, for example equal to 3.
[0115] For example, in 5G, a is between 2 and 4, and the number of unaffected elements per row in the matrix formed by the extension matrix C (excluding the first two columns) and the extension matrix D is between 0 and 6, for example equal to 2.
[0116] According to another example, the number of unaffected elements per row in the extension matrix C (excluding the first two columns) is equal to 1.
[0117] According to yet another example, the number of undamaged elements per row in the extension matrix D is equal to 1.
[0118] In particular, if J = 4, K = {6, 8, 9 or 10} and M = {26^ 34, 38o»42}, the modified basis matrix BG ' has a size identical to the basis matrix BG2, and can be used in any system using the 5G standard.
[0119] 5.2 Basic matrix BG1
[0120] In order to better understand the invention, we recall in figure 6A 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 I and O are omitted for the sake of simplification.
[0121] The “1” elements of the basic 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].
[0122] 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.
[0123] The first column of matrices B and D includes 12 connections, the second column includes 5 connections, etc.
[0124] The first four rows of matrices A and B each include 19 connections.
[0125] The first row of matrices C and D includes 2 connections, the second row includes 7 connections, etc.
[0126] If we do not take into account the zero matrices O and identity 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 strong disparity in the number of connections per row and / or per column in the matrix BG1.
[0127] The kernel matrix A has, in proportion, many connections, i.e. many unharmed elements, because it is the kernel of the BG1 matrix. This allows the decoder, on reception, to find this data. We also note that the first two columns of the BG1 matrix, which correspond to the untransmitted data, also have many "1s".
[0128] The upper part of the BG1 die also has more connections than its lower part, which helps ensure good performance for high yields. Indeed, the lower part of the BG1 die can be punched to achieve the desired yield. It is therefore preferable to choose the connections in the upper part of the BG1 die.
[0129] As an example, Figure 6B illustrates the circular rotation factors V for the matrix BG1 of Figure 6A, for an expansion factor Z = 256.
[0130] 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 of 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 of 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.
[0131] 5.3 BG2 base matrix
[0132] Similarly, ETSI TS 138 212 V15.2.0 describes the structure of the basic matrix BG2, which can be decomposed into sub-matrices A, B, C, D, I and O. Sub-matrices I and O are omitted hereinafter for the sake of simplification.
[0133] 5.4 Modified base matrix
[0134] The position of the connections in the BG2 base matrix, as well as the rotation factors to be used, are specified in the 5G 3GPP standard. However, as previously indicated, 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.
[0135] A modified basic matrix is therefore proposed below, seeking to reduce the complexity of decoding. New connection structures are thus proposed, according to different embodiments of the invention. According to the embodiment considered, such structures are efficient and / or easy to reproduce depending on the dimensions of the parity matrix H.
[0136] According to a particular embodiment, the structure of the modified basic 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 coder for example, we recall that LDPC decoding is based on the bases of 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 rows, and especially in columns, is substantially constant or equally distributed.
[0137] According to another particular embodiment, the structure of the modified basic matrix is defined so as to limit the number of short cycles, and if possible to avoid short cycles ("Girth" in English). Indeed, such short cycles limit the performance of error-correcting codes, in particular LDPC codes, especially for small K values, for example of the order of 256 bits or less.
[0138] To better understand this principle, Figures 7A and 7B illustrate the loop phenomena with an expansion factor Z = 4. According to Figure 7A, an expansion factor Z = 4 is applied with the same rotation factor ^ = 0 to the elements "1" located in position ( / , J), ( / + 7, j), (z, j + 7) and (i + 7, j + 7). A loop is obtained, illustrated by the arrows. According to Figure 7B, an expansion factor Z = 4 is applied with the same rotation factor V = 0 to the elements "1" located in position (i, j), ( / + 7, / ) and (7 + 7, j + 7), and a rotation factor V = 1 to the element located in position ( / , j + 7). In this case, there is no loop. Changing the rotation factor of one of the matrix elements therefore makes it possible to eliminate loops.
[0139] We present below different solutions making it possible to obtain a low disparity in the number of connections per row and / or per column of the modified basic matrix and / or to limit (or even avoid) short cycles.
[0140] 5.4.1 Optimization of the kernel matrix A
[0141] 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.
[0142] For example, in the kernel matrix A the number of unharmed elements per column, excluding the first two columns, is between 1 and J.
[0143] Figures 8A to 8D illustrate an example of a kernel matrix A, in which the number of unaffected elements per column is constant and equal to 3, respectively for K = 6, K = 8, K = 9 and K = 10.
[0144] 5.4.2 Optimization of the first two columns of the kernel matrix A and / or of the C expansion matrix
[0145] According to a second example, the first two columns (which are generally not transmitted) can be chosen to guarantee a good distribution of connections.
[0146] For example, the first two columns of the kernel matrix A and / or the extension matrix C comprise an alternation of unharmed and harmed elements per row and per column. In other words, the connections are distributed in a staggered manner, as illustrated in [Fig.9].
[0147] 5.4.3 Optimization of the extension matrix C and the extension matrix D
[0148] According to a third example, it is possible to distribute the connections fairly in the extension matrix C (excluding the first two columns) and / or in the extension matrix D.
[0149] The extension matrix C is generally the largest in terms of its dimension. For the basic matrix BG2, for example, the extension matrix C, excluding its first two columns, has a size:
[0150] 22x4 for K = 6
[0151] 30x6for^ = 8
[0152] 34 x 7 for K — 9 and
[0153] 38 x 8 for K = 10.
[0154] According to this third example, the connections in the extension matrix C are distributed on diagonals to ensure a good distribution of the connections. In particular, the connections are distributed to have one connection per row and one connection per column for a block of rows.
[0155] According to the invention, the extension matrix C therefore comprises at least two blocks of lines, each comprising the same number of lines, each block of lines comprising a diagonal matrix.
[0156] Figure 10A illustrates an example of an extension matrix C for K - 6, excluding the first two columns. The extension matrix C according to this example comprises at least two blocks of rows (101A, 102A, 103A), each comprising the same number of rows (L1 = 4), each block of rows comprising a diagonal matrix (of size L1 x L1). One or more rows of nuisance elements may be interposed between two blocks of rows. For example, four rows of nuisance elements are present between two blocks of rows comprising a diagonal matrix.
[0157] Figure 10B illustrates another example of an extension matrix C for K = 8, excluding the first two columns. The extension matrix C according to this example comprises at least two blocks of rows (101B, 102B, 103B, 104B), each comprising the same number of rows (LL - 6), each block of rows comprising a diagonal matrix (of size LL x LL). One or more rows of nuisance elements may be interposed between two blocks of rows. For example, two rows of nuisance elements are present between two blocks of lines comprising a diagonal matrix.
[0158] Similarly, the extension matrix D may comprise at least two blocks of lines, each comprising the same number of lines, each block of lines comprising a diagonal matrix.
[0159] The connections in the extension matrix D can thus be equally distributed on diagonals to ensure a good distribution of the connections. In particular, the connections are distributed to have one connection per row and one connection per column for a block of rows.
[0160] Figure 1 1A illustrates an example of an extension matrix D for K = 6. The extension matrix D according to this example comprises at least two blocks of lines (11 1A, 112A, 113A, 114A, 115A), each comprising the same number of lines (L2 = 4), each block of lines comprising a diagonal matrix (of size L2 x L2). The blocks of lines are consecutive, i.e. no line of null elements is interposed between two blocks of lines.
[0161] Figure 1 IB illustrates another example of an extension matrix D for K = 8. The extension matrix D according to this example comprises at least two blocks of lines (111B, 112B, 113B, 114B, 115B, 116B, 117B), each comprising the same number of lines (L2 = 4), each block of lines comprising a diagonal matrix (of size L2 x L2). The blocks of lines are consecutive, i.e. no line of null elements is interposed between two blocks of lines.
[0162] In particular, all line blocks have an identical size.
[0163] According to this particular embodiment, since the connections are distributed to have one connection per row and one connection per column for a block of rows in the extension matrices C (excluding the first two columns) and extension D, the number of connections per row is equal to 1 (if we consider a row of null elements in the extension matrix C) or 2.
[0164] The number of connections per column depends on the size of the variable K. In the example illustrated in Figures 10A and 11A, for K = 6, it is equal to 3 for the extension matrix C and equal to 5 or 6 for the extension matrix D. In the example illustrated in Figures 10B and 11B, for K = 8, it is equal to 4 for the extension matrix C and equal to 7 or 8 for the extension matrix D.
[0165] As an example, Figure 12 proposes an example of structure for the kernel matrices A, double diagonal B, extension C and extension D, for different values of K, using the matrices of Figures 10A and 11A for K = 6, and the matrices of Figures 10B and 11B for K - 8. The structure is extended to other values of K, in particular K - 9 and K = 10. As illustrated in Figure 12, the matrices A, C and D have a nested structure according to the value of K. Such a structure is in particular compatible with the 5G standard.
[0166] In the matrix formed from the extension matrix C and the extension matrix D, the number of unaffected elements per row, excluding the first two columns, is between a - 2 and « + 2, with a an integer.
[0167] If we place ourselves in the context of 5G, a is for example between 2 and 4.
[0168] 5.4.4 Global optimization of the modified basis matrix BG'
[0169] In the examples above, the extension matrices C (except the first two columns) and extension D are optimized to reduce short cycles.
[0170] However, if we take into account the matrix formed by the kernel A, double diagonal B, spreading C and extension D matrices proposed in the examples above, short cycles may remain.
[0171] Short cycles can for example result from the structure proposed for the kernel matrix A or for the first two columns of the spreading matrix C.
[0172] As illustrated in [Fig.7B], taking rotation factors into account makes it possible in particular to reduce, or even avoid, short cycles.
[0173] 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.
[0174] If we place ourselves 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 base matrix can be replaced by 1 0 for a circular rotation V - 0 or by [0 11 for a circular rotation 0 11 11 01 V= 1.
[0175] Figures 13A to 13D illustrate examples of structure for matrices A, B, C and D. for different values of K. respectively K = 6, K = 8, K = 9 and K = 10, taking into account at least two distinct rotation factors. As illustrated in these figures, it is possible to increase the number of rotation factors to increase the diversity of the modified basic 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 matrices A, B, C and D.
[0176] As already indicated, the matrices A, C and D have a nested structure according to the value of AJ| it is in particular possible to define the equations which define the rotation factors V0 to V4 as a function of the index "ils" according to the 5G 3GPP standard. Index "they" Circular Rotation V 0 V0=0 + 4y, ye [0:63] Vl=l+4y, yg [0:63] V2 = 2 + 4y, ye [0:63] V3 = 3 + 4y, ye [0:63] V4 = 0+4y, ye [0:63] 1 VO - 0 + 5x + 6y, xe [0: 1] et ye [0: 63] VI = l + 6y, ye [0:63] V2 = 2 + 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, yg [0:63] V2 = 2 + 5y, yg [0:63] V3 = 3 + 5y, yg [0:63] V4 = 4 + 5y, yg [0:63] 3 VO - 0 + 5x + 7y, xg [0: 1] et yg [0: 31] VI = 1 + 5x + 7y, xg [0:1] et yg [0: 31] V2=2 + 7y, >-g[0:31] V3 = 3 + 7y, yg [0:31] V4 = 4 + 7y, yg [0:31] 4 VO = 0 + Sx + 9y x G [0 : 1] et y G [0: 31] VI = l + 5x + 9y, xg [0 : yg [0:31] V2 = 2 + 5x + 9y, xg [0:1] et yg [0: 31] V3 = 3 + 5x + 9y, xg [0: 1] et yg [0: 31] V4 = 4 + 9y, yg [0:31] 5 VO = 0 + 5x + 1 ly, xg [0: 2] et yg [0: 31] VI = 1 + 5x + 1 ly, xg [0: 1] et yg [0: 31] V2 - 2 + 5x 4-1 ly, xg [0: 1] et yg [0: 31] V3 = 3 + 5x + 1 ly, xg [0: 1] et yg [0: 31] V4 = 4 + 5x + 1 ly, xg [0:1] et yg [0:31] 6 VO=O+5x+13y, xc [0: 1] and ye [0: 15] P'4 = 4 + 5x+ 13}, xe [0: 1] and yg [0: 15] 7 VO = 0+ 5x + 15}; xs [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 + 15}, xg [0: 2] and yg [0: 15]
[0177] with x and 2' random variables whose interval is given in the table above.
[0178] The structures illustrated in Figure 13A to 13D make it possible in particular to reduce short cycles for small sized matrices, for example for K < 256.
[0179] It is noted in particular that in the matrix formed from 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 0 4- 3, with / 5 an integer. In the matrix formed from the extension matrix C and the extension matrix D, the number of unaffected elements per row, excluding the first two columns, is between a - 2 and « + 2, with a an integer.
[0180] If we place ourselves in the context of 5G, is for example between 7 and 10, for example equal to 9, and a between 2 and 4, for example equal to 3. In the examples illustrated in Figures 13A to 13D, the number of unharmed elements per column in the matrix formed by the kernel matrix A and the extension matrix C is between 6 and 12, depending on the value of K. and the number of unharmed elements per row in the matrix formed by the extension matrix C and the extension matrix D is between 1 and 2.
[0181] The number of connections per row and / or per column is therefore fairly homogeneous.
[0182] 5.4.5 Obtaining a parity matrix
[0183] From the modified base matrix BG' thus obtained, it is possible to obtain a parity matrix H, by replacing each empty or zero element of the modified base matrix BG' by a zero matrix of size ZXZ, and each non-zero element of the modified base matrix BG' by a circular permutation matrix obtained by applying a circular rotation Vi to an identity matrix of size Z x Z, z G {0,1, ...} •
[0184]
[0185]
[0186]
[0187]
[0188]
[0189]
[0190]
[0191]
[0192]
[0193]
[0194]
[0195] The parity matrix H thus obtained can be used by an encoder to encode at least one block of KZ source data. 5.4.6 Variants An example has been described above in which the modified basis matrix BG' has the same size as the basis matrix BG2, with J — 4, K- {6, 8, 9owl0} andM= {26, 34, 38 or 42}. This allows in particular to be able to use the modified basic matrix BG' in any communications system according to the 5G standard. However, this is just an example, and other values for J, K and M can be considered, particularly 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 1 0 .1 0 0 0” 1 0 1 1 0 1. Other forms of the B matrix with at least one double diagonal can be used. In this case, the position of the connections and / or the values of the rotation factors can be updated to take into account the structure of the B matrix. An LDPC code was also taken as an example. The invention can also be extended to other error-correcting codes using a parity matrix. 5.5 Coding Device Finally, in relation to [Fig.14], we present the simplified structure of an encoder according to at least one embodiment described above. As illustrated in [Fig. 14], an encoder comprises at least one memory 141, at least one processing unit 142, equipped for example with a programmable computing machine or a dedicated computing machine, for example a processor P, and controlled by the computer program 143, implementing steps of the coding method according to at least one embodiment of the invention. Upon initialization, the code instructions of the computer program 143 are for example loaded into a RAM memory before being executed by the processor of the processing unit 142. The processor of the processing unit 142 implements steps of the coding method described previously, according to the instructions of the computer program 143, to code the KZ source data using a parity matrix H of size (MZ XNZ), and deliver at least one code word of size NZ formed of the KZ source data and MZ redundancy data.
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 the same number of rows, each block of rows comprising a diagonal matrix.,
2. Coding method according to claim 1, characterized in that said extension matrix D comprises at least two blocks of lines, each comprising the same number of lines, each block of lines comprising a diagonal matrix.
3. 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 50 and 100%.
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. 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 / = 4 and: ^ = 6etA / = 26, or K — 8 and M — 34, or K = 9 and M = 38 or K = 10 and M = 42. Method according to any one of the preceding claims, characterized in that, depending on the value of K, said kernel matrix A is equal to: for K= 10 for K=9 for K=8 for K=6 VI VO V3 V2 VI V4 V3 V2 V4 VO V3 V2 V4 V2 V3 vo V2 VO V3 V4 V2 V3 VI VI V3 V4 V4 said matrix B having at least one double diagonal is equal to: VO vo vo vo VI vo vo vo vo said extension matrix C is equal to: for K= 10 for K= :9 for K=8 for K= 6 V4 vo V2 VI V3 V2 VI V3 V2 V4 VO vo VI VI V2 V2 vo VI VI V2 V4 V3 VI V4 V2 VO vo VI VI V2 V4 V3 V2 V2 V3 VO V4 VI VO V4 VI vo V2 V2 V3 VO V4 VI V3 V3 V4 VI VO V2 VI vo V2 VI V3 VI V4 V2 VO V4 V4 V2 VO V4 VI vo V2 VI V3 vo V4 said extension matrix D is equal to: each empty or zero element of said modified basis matrix BG' is replaced by a zero matrix of size Zx 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: Rotation Circulaire Vi si Z appartient à {2,4, 8,16,32,128,256} V0 = 0 + 4y, ye [0:63] Vl=l + 4y, ye [0:63] V2 = 2 + 4y, ye [0:63] V3 = 3 + 4y, ye [0:63] V4 = 0 + 4y, ye [0:63] si Z appartient à {3,6, 12,24,48,96,192,384} VO = 0 + 5x + 6y, _x e [0: 1] et ye [0: 63] VI = l + 6y, ye [0:63] V2 = 2 + 6y, ye [0:63] V3 = 3 + 6y, ye [0:63] V4 = 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 + si [ 5y, 4 ] [ 40: 6 ] V [ 40:6 = Z belongs to {7,14,28,56,112,224} V0 = 0 + 5x + 7y, x G [0: 1] and y G [0 : 31] VI = 1 + 5x + 7y, xg [0: 1] and yg [0: 31] V2 = y = 2+7 yg [0:31] V4 = 4 + 7y, yg [0:31] if Z belongs to {9,18,36,72,144,288} V0 = 0 + 5x + 9y, xe[0:1] er ye[0:31] VI - l + 5x2] [03 +9y, V =2]. 5x + 9y, xg [0 : 1] and y G [0: 31] V3 = 3 + 5x + 9y, x G [0: 1] and yg [0: 31] V4 = 4 + 9y, ye [0:31] if Z belongs to VO = 0+5x + y [1:2]: 3 {11,22,44,88,176,352} Vl = l + 5x+ lly, xe[0: 1] and yg[0:31] V2 = 2 + 5x+lly, xe[0: l]^ye[0:31] V3 = 3 + 5x +1 ly,+ 1,+ [0:1] 1 xe [0:1] and yg [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]
8. 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), 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 Z, said modified base matrix BG ' being expressed in the form: [AB 0] BG =l _ , with: A a kernel matrix of size J xK B a matrix having at least one double diagonal of size JXJC 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 the same number of rows, each block of rows comprising a diagonal matrix.
9. Computer program comprising instructions for implementing a method according to any one of claims 1 to 7 when this program is executed by a processor.
Citation Information
Patent Citations
Quasi-cyclic-low density parity check code construction method and device
CN115276670A
Encoding method and apparatus, and computer storage medium
EP3648357A1
Low density parity check decoder, electronic device, and method therefor
WO2021233788A1
