A genetic optimization method for block-structured LDPC codes

Through the genetic algorithm, the basic matrix of the block structure LDPC code is optimized, which solves the shortcomings of the existing LDPC code in terms of bit error rate and bit error level, and realizes lower frame error rate and higher flexibility codeword performance Improved, suitable for a variety of code lengths and code rates.

CN113992210BActive Publication Date: 2024-11-26SOUTHEAST UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111264759.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-28
Publication Date
2024-11-26
Estimated Expiration
2041-10-28

AI Technical Summary

Technical Problem

The existing LDPC codes have shortcomings in bit error rate and bit error level, especially in high signal-to-noise environments, which are difficult to achieve extremely low bit error rate and frame error rate. The existing optimization algorithms lack flexibility and are difficult to adapt to multiple A design environment.

Method used

The genetic algorithm is used to optimize the base matrix of the block structure LDPC code. By stating the base matrix information, setting the test information, initializing the population, performing mutation and cross-operation, row vectors are optimized generation by generation to reduce the frame error rate, and the optimized base is formed. Matrix to improve codeword performance.

Benefits of technology

It significantly improves codeword performance, is suitable for the optimization of multiple code lengths and code rates, and achieves lower frame error rates and higher flexibility, and is suitable for not only binary domains, but also multivariate domains.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113992210B_ABST
    Figure CN113992210B_ABST
Patent Text Reader

Abstract

The present invention provides a genetic optimization method applicable to block-structured LDPC codes, comprising the following steps: S01 statistically analyzing the information of the base matrix; S02 setting test information; S03 initializing rows; S04 analyzing the initial vector of the row to be optimized; S05 initializing genetic evolution; S06 initializing the population; S07 calculating the fitness of the k-th generation population; S08 performing a selection operation; S09 performing a mutation operation; S10 performing a crossover operation; S11 updating the number of loop generations: k = k + 1; if k < K, then proceed to step S07, otherwise proceed to step S13; S13 temporarily storing the optimal result; S14 updating the number of optimized rows: m = m + 1; S15 if m < M2, then proceed to step S04, otherwise proceed to step S16; S16 outputting the optimal result; S17 aborting the algorithm. The optimization method of the present invention can not only achieve the construction optimization of a single codeword, but also jointly optimize the construction of multiple codeword matrices under the condition of being compatible with multiple code lengths and multiple code rates.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of channel coding and decoding, and relates to the optimization of low-density parity check codes, and in particular to a genetic optimization method suitable for block-structured LDPC codes. Background Art

[0002] LDPC code, or Low-Density Parity-Check (LDPC) code, is a type of linear block code with a sparse check matrix first proposed by Dr. Robert G. Gallager in 1963 and re-proposed by DJC Mackay and RW Neal in 1995. As one of the modern error-correcting codes, it is widely used in various communication systems such as Digital Video Broadcasting (DVB), Advanced Television Systems Committee (ATSC) and 5G NR due to its high throughput, error correction performance close to capacity and low encoding and decoding complexity.

[0003] As one of the most widely used LDPC codes, different from general LDPC codes, the check matrix of block-structured LDPC codes can be regarded as composed of block sub-matrices, each sub-block matrix can be marked with an integer, so that the entire block LDPC code check matrix can be represented by a lower-dimensional base matrix, which greatly reduces the complexity of encoding and decoding and is more convenient for hardware implementation. It has a structured check matrix, supports the use of different expansion factors to achieve the length scalable construction of LDPC codes, and also supports different degrees of truncation on the original base matrix to achieve multi-rate construction, so it can realize the encoding and decoding of a series of codewords with different code lengths or different code rates on a fixed codec.

[0004] Taking 5G LDPC code as an example, the 3GPP protocol provides two base matrices BG1 and BG2, and their matrix structures are as follows: Figure 1 As shown in the figure, the entire matrix has a raptor-like structure, consisting of five parts: A, B, C, D, and E. Each element in the base matrix can be replaced by a Z×Z cyclic shift matrix or an all-zero matrix. Since the expansion factor specified in the 3GPP protocol can range from 2 to 384, the codeword length corresponding to its BG1 matrix can reach 8848, and that of BG2 can reach 3840.

[0005] In many important applications, such as data storage and high-speed digital communications, extremely low bit error rates (BER) or frame error rates (FER) are usually required. Theoretically, the bit error rate of LDPC code decoding will decrease as the signal-to-noise ratio increases, and can be close to the Shannon limit, but in reality, LDPC codes usually have an error floor, so in this scenario, the bit error rate and error floor are required to be as low as possible. There are some existing constructions for LDPC codes, most of which use some classic tools, such as density evolution, EXIT, etc. Most of these algorithms take the basic structural properties of LDPC codes, such as code distance characteristics, circle length, etc. as design targets, and can design and optimize codewords to a certain extent, but the algorithms are relatively inflexible and difficult to adapt to more design environments. Summary of the invention

[0006] The purpose of the present invention is to provide a genetic optimization method suitable for block structure LDPC codes, which takes the bit error rate as the optimization guide target, uses a genetic algorithm for evolution, and optimizes the LDPC code based on an existing base matrix. The optimized codewords do not change the basic structure of the original base matrix, and the optimization and improvement of the codeword performance are more obvious. For the optimization compatible with multiple codewords, better performance is obtained not only in general but also in each code length than before the optimization.

[0007] To achieve the above object, the present invention adopts the following technical solution:

[0008] A genetic optimization method applicable to block structured LDPC codes comprises the following steps:

[0009] Step S01, statistical basis matrix information:

[0010] The base matrix B of the initial block structure LDPC code to be optimized has M rows b and the number of columns N b and its expansion factor group Z involved in the optimization;

[0011] Step S02, setting test information:

[0012] Including the signal-to-noise ratio Γ0 for evaluating performance and the frame error rate Θ0 of the initial codeword under the corresponding signal-to-noise ratio;

[0013] Step S03, row initialization:

[0014] Initialize the row to be optimized m=M1;

[0015] Step S04, initial vector analysis of the row to be optimized:

[0016] Analyze the line weight ρ of the line to be optimized within the evaluation region; according to the properties of the rows of the block-structured base matrix, form the position vector p and the parameter vector e from the positions of its non-zero elements and their corresponding shift parameters, and then form the description vector d = {p, e} of this line as the initial vector for optimizing this line;

[0017] Step S05, Genetic evolution initialization:

[0018] Initialize the loop generation number k = 0, specify the maximum loop generation number K, and specify the number of population individuals I, where I is a positive even number;

[0019] Step S06, Initialize the population;

[0020] Step S07, Calculate the fitness of the k-th generation population, k = 0, 1,..., K;

[0021] Step S08, Selection operation;

[0022] Step S09, Mutation operation;

[0023] Step S10, Crossover operation;

[0024] Step S11, Update the loop generation number: k = k + 1;

[0025] Step S12, If k < K, then go to Step S07, otherwise go to Step S13;

[0026] Step S13, Temporarily store the optimal result:

[0027] Restore this line according to the block structure through the position vector and parameter vector information of the saved individuals, and replace the original base matrix row to form the base matrix B m , and save it;

[0028] Step S14, Update the number of optimized lines: m = m + 1;

[0029] Step S15, If m < M2, then go to Step S04, otherwise go to Step S16;

[0030] Step S16, Output the optimal result:

[0031] Output the optimal base matrix result B of the last optimized line M2 Output;

[0032] Step S17, Abort the algorithm.

[0033] In the said Step S01, for a block-structured LDPC code, its parity-check matrix is represented by the base matrix B with M b rows and N b columns and the expansion factor Z. To obtain the final LDPC code, each element b of the base matrix B mnReplaced by a Z×Z all-zero matrix (b mn =-1) or take the value b of the non-negative element in the basis matrix mn The mapped unit matrix is ​​cyclically shifted (b mn ≥0), 1≤m≤M b ,1≤n≤N b ; The final LDPC code matrix size is defined as M×N, where M=M b ×Z,N=N b ×Z;

[0034] First, read the base matrix B of the initial block structure LDPC code to be optimized = [b mn ], get the number of rows M b and the number of columns N b And the expansion factor group Z involved in the optimization is {Z1, Z2, ..., Z S}, where the element b in the basis matrix mn The maximum expansion factor Z involved in the optimization S Correspondingly, Z1 <Z2<,…,<Z S ; and obtain the region m∈[M1,M2], n∈[N1,N2] of the basis matrix to be optimized, where 1≤M1 <M2≤M b , 1≤N1 <N2≤N b .

[0035] In step S02, for each LDPC code check matrix to be optimized, that is, the group of extension factors to be optimized Z={Z1, Z2, ..., Z S}, set the signal-to-noise ratio Γ0 to evaluate the performance = {γ1,γ2,…,γ S} and the frame error rate of the initial codeword under the corresponding signal-to-noise ratio Θ0={θ1,θ2,…,θ S}.

[0036] In step S07, for each individual in the population, according to the population individual d ki , where i = 1, 2, ..., I, respectively restore the row according to the block structure and replace the original basis matrix row to form the test basis matrix B ki On this basis, through each optimization expansion factor Z s ,s=1,2,…,S expansion, get the corresponding test code word H ki (s), for different expansion factors Z s ,s=1,2,…,S-1, its basis matrix elements are obtained through Z S The basis matrix element b corresponding to the factor mn Modulo different expansion factors mod(b mn ,Z s )get;

[0037] Through decoding simulation test in γ s The frame error rate θ under ki (s), and then calculate the comprehensive test parameters:

[0038]

[0039] Among them, θ ki (s) is the i-th individual of the k-th generation, corresponding to the expansion factor Z s The matrix of the signal-to-noise ratio γ s The frame error rate under s The expansion factor before optimization is Z s The original matrix in the corresponding evaluation signal-to-noise ratio γ s The frame error rate is obtained by the following test, s = 1, 2, ..., S;

[0040] Then calculate its fitness f ki =1 / σ ki ,i=1,2,…,I。

[0041] In step S08, for the individuals in the population, the selection probability corresponding to each individual is formed according to the population fitness. Then construct the roulette probability RP of each individual k [i], set RP k [0] = 0, using the iterative formula RP k [i] = RP k [i-1]+SP k [i], i = 1, 2, ..., I, forming the roulette probability RP k [1],RP k [2],…,RP k [I], where RP k [I] = 1;

[0042] Generate a random number w within 1, 0≤w<1, and determine the roulette area where w is located. If RP k [i-1]≤w <RP k [i], then select the i-th individual and describe its vector d ki ={p ki ,e ki}, including the position vector p ki and parameter vector e ki Information saved.

[0043] In step S09, for each individual in the population, whether to mutate is determined according to the mutation probability MP, and a mutation point α∈[1,ρ] is randomly generated. ki ={p ki ,e ki} position vector and parameter vector The position and parameters at the mutation point α are mutated to positions different from other position coordinates in the range [N1, N2] and in [0,Z S -1] range of parameters Get individual in

[0044] If the position vector p ki There are positions α other than the mutation point p The value at If the original position vector is the same as α, then p Swap the values ​​at .

[0045] In step S10, the population is divided into two parts, and the individuals of the population are paired, that is, the 2v-1th individual is paired with the 2vth individual, v = 1, 2, ..., I / 2; for each pair of individuals, whether to crossover is determined according to the crossover probability CP, and a crossover point β∈[1,ρ] is randomly generated, and the two individuals to be crossed d k(2v-1) and d k(2v) The position vector and and the parameter vector and The values ​​at β, β+1, …, ρ are exchanged respectively to obtain two individuals after crossover. and in,

[0046] If after the crossover, a position value is in the position vectors of two individuals or If it exists more than once in d, skip this pair. k(2v-1) and d k(2v) Crossover operation.

[0047] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0048] 1. Most of the existing LDPC codeword optimization structures are optimized based on the basic structural properties of LDPC codes. Even if the evolutionary algorithm is used for optimization, the main purpose is to optimize its structural properties. The present invention uses the bit error rate as the optimization guide target, which has a clear goal and strong feasibility.

[0049] 2. The present invention has certain flexibility. It can not only optimize a single codeword, but also can comprehensively optimize multiple codewords with compatible code lengths and compatible code rates based on the same base matrix. By using a unified evaluation standard, the overall codeword performance can be optimized, and even the performance of multiple codewords that are optimized together can be optimized.

[0050] 3. The present invention has certain mobility and is applicable not only to binary domains but also to multi-domains; it is applicable not only to LDPC codes but also to codewords in which each row can be described as two unrelated vectors. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a schematic diagram of the 5G LDPC code matrix structure;

[0052] Figure 2 A flow chart of the genetic optimization method for block structured LDPC codes provided by the present invention;

[0053] Figure 3 Schematic diagram of parameter values ​​of D blocks in the check matrix of 5G_BG1_LDPC code with code rate R=1 / 2 and expansion factor Z=48;

[0054] Figure 4 A schematic diagram of parameter values ​​of D blocks in a check matrix with a code rate of R=1 / 2 and an expansion factor of Z=48 using the optimized 5G_BG1_LDPC code of the present invention;

[0055] Figure 5 Frame error rate comparison curves of 5G_BG1_LDPC code before and after optimization using the present invention at code rate R=1 / 2 and expansion factors Z=48, 24, 12 and 6. DETAILED DESCRIPTION

[0056] The present invention will be further explained below in conjunction with the accompanying drawings.

[0057] like Figure 2 As shown, the present invention is a genetic optimization method suitable for block structure LDPC codes, comprising the following steps:

[0058] S01: Statistical basis matrix information

[0059] For a block structure LDPC code, its check matrix can usually be expressed as M b Row N b The column basis matrix B and expansion factor Z represent that, in order to obtain the final LDPC code, each element b of the basis matrix B mn can be replaced by a Z×Z all-zero matrix (b mn =-1) or take the value b of the non-negative element in the basis matrix mn The mapped unit matrix is ​​cyclically shifted (b mn≥0), 1 ≤ m ≤ M b , 1 ≤ n ≤ N b . The matrix size of the final LDPC code can be defined as M × N, where M = M b × Z, N = N b × Z.

[0060] First, read the base matrix B = [bmn] of the initial block-structured LDPC code to be optimized, obtain its number of rows Mb and number of columns Nb, and its expansion factor group Z = {Z1, Z2, …, ZS} involved in optimization. Here, the element bmn in the base matrix corresponds to the maximum expansion factor ZS involved in optimization, Z1 < Z2 <, …, < ZS; and obtain the region to be optimized in the base matrix m ∈ [M1, M2], n ∈ [N1, N2], where 1 ≤ M1 < M2 ≤ Mb, 1 ≤ N1 < N2 ≤ Nb.

[0061] In one embodiment, assume the read base matrix is as shown in Equation (1). Through analysis, obtain its basic parameters M b = 4, N b = 6. Assume its expansion factor group Z = {4, 8, 16} involved in optimization. The base matrix element b mn in Equation (1) corresponds to the maximum expansion factor Z3 = 16 involved in optimization; assume the region to be optimized is the entire matrix, i.e., m ∈ [1, 4], n ∈ [1, 6];

[0062]

[0063] S02: Set test information

[0064] For the expansion factor group Z = {4, 8, 16} applicable to the above base matrix structure, set the signal-to-noise ratio Γ0 = {γ1, γ2, γ3} for evaluating performance and the frame error rate Θ0 = {θ1, θ2, θ3} of the initial codeword at the corresponding signal-to-noise ratios;

[0065] S03: Row initialization

[0066] Initialize the row to be optimized m = 1;

[0067] S04: Analysis of the initial vector of the row to be optimized

[0068] Analyze the row weight ρ of the row to be optimized within the evaluation region; according to the properties of the rows in the block-structured base matrix, form the position vector p and the parameter vector e from the positions of its non-zero elements and their corresponding shift parameters, and then form the description vector d = {p, e} of this row as the initial vector for optimizing this row;

[0069] For the row to be optimized, m=1 is taken as an example, and the row weight ρ=4 of the row in the evaluation area is counted; according to the properties of the row of the block structure basis matrix, the initial position vector p={1,3,5,6} and parameter vector e={8,4,7,2} of the row can be obtained, and then the initial row description d={p,e} is formed as the initial vector for the optimization of the row;

[0070] S05: Algorithm initialization

[0071] Initialize the number of cyclic generations k = 0; set the maximum number of cyclic generations K = 30; set the number of individuals in the population I = 20;

[0072] S06: Initialize the population

[0073] According to the row weight ρ and the maximum expansion factor Z3=16 in the expansion factor group Z, for each individual, randomly generate ρ different positive integers in the range [1,6] to form the position vector p 0i , randomly generate ρ non-negative integers in the range [0,15] to form the parameter vector e 0i , and then form the i-th description vector d of the 0th generation of this row 0i ={p 0i ,e 0i}, in this way, 20 different description vectors are randomly generated to form the initial 0th generation population P0 = {d 01 ,d 02 ,…,d 0I};

[0074] For example, for the first row of this example, according to the row weight ρ = 4, we can randomly generate individual d 0i ={p 0i ,e 0i}, where p 0i ={1,2,4,5},e 0i ={7,0,2,5}, as the i-th individual in the initial population. The initial vector d={p,e} optimized in this row may also become an individual in the initial population, assuming it is the j-th individual, that is, d 0j ={p 0j ,e 0j}, where p 0j ={1,3,5,6},e 0j ={8,4,7,2}.

[0075] S07: Calculate the fitness of the kth generation population, k = 0, 1, ..., 30

[0076] For each individual in the population, according to the population individual d ki , i = 1, 2, ..., 20, restore the row according to the block structure and replace the original basis matrix row to form the test basis matrix Bki For example, according to d 0i ={p 0i ,e 0i}, where p 0i ={1,2,4,5},e 0i ={7,0,2,5} to get the test basis matrix B 0i As shown in formula (2)

[0077]

[0078] On this basis, by optimizing each expansion factor Z s ,s=1,2,3, and get the corresponding test code word H ki (s), for different expansion factors Z s ,s=1,2,3, its basis matrix elements can be obtained by the basis matrix elements b corresponding to Z3=16 mn Modulo different expansion factors mod(b mn ,Z s )get.

[0079] Through decoding simulation test in γ s The frame error rate θ under ki (s), and then calculate the comprehensive test parameters:

[0080]

[0081] Among them, θ ki (s) is the i-th individual of the k-th generation, corresponding to the expansion factor Z s The matrix of the signal-to-noise ratio γ s The frame error rate under θ s The expansion factor before optimization is Z s The original matrix in the corresponding evaluation signal-to-noise ratio γ s The following test obtains the frame error rate, s = 1, 2, 3;

[0082] Then calculate its fitness f ki =1 / σ ki ,i=1,2,…,20;

[0083] S08: Select Operation

[0084] For individuals in the population, the selection probability corresponding to each individual is formed according to the fitness of the population Then construct the roulette probability RP of each individual k [i], set RP k [0] = 0, using the iterative formula RP k [i] = RP k [i-1]+SP k[i], i = 1, 2, ..., 20, forming the roulette probability RP k [1],RP k [2],…,RP k

[20] , where RP k

[20] =1.

[0085] Generate a random number w within 1, 0≤w<1, and determine the roulette area where w is located. If RP k [i-1]≤w <RP k [i], then select the i-th individual and describe its vector d ki ={p ki ,e ki}, including the position vector p ki and parameter vector e ki Information saved.

[0086] S09: Mutation Operation

[0087] For each individual in the population, whether to mutate is determined according to the mutation probability MP = 0.1, and a mutation point α∈[1,ρ] is randomly generated. ki ={p ki ,e ki} position vector and parameter vector The position and parameters at the mutation point α are mutated to positions different from other position coordinates in the range [1,6] and parameters in the range [0,15] Get individual

[0088] If the position vector p ki There are positions α other than the mutation point p The value at If the original position vector is the same as α, then p Swap the values ​​at .

[0089] Assume that the individual d in the population of generation 0 0i ={p 0i ,e 0i}, p 0i ={1,2,4,5},e 0i ={7,0,2,5} After being determined to be subject to mutation, the randomly generated mutation point α=2, its position vector p 0i The value here varies to 3, and the parameter vector e 0i The value here is mutated to 10, and the mutated individual is obtained in, If during the mutation process the position vector p0i The value at the mutation point α = 2 is mutated to 5, and at the position vector p 0i At α p = 4, if the values are the same, then p 0i At α = 2 and α p = 4, the values are exchanged. At this time, the position vector in the new individual obtained is

[0090] S10: Crossover operation

[0091] The population is evenly divided into two parts, and the individuals in the population are paired in pairs, that is, the (2v - 1)-th and the 2v-th individuals are paired, where v = 1, 2,..., 10. For each pair of individuals, it is judged whether to perform crossover according to the crossover probability CP = 0.1. A random crossover point β ∈ (1, ρ) is generated, and the position vectors p k(2v-1) and p k(2v) of the two individuals to be crossed d k(2v-1) and p k(2v) as well as the parameter vectors e k(2v-1) and e k(2v) at β, β + 1,..., ρ are exchanged respectively to obtain two crossed individuals and

[0092] If after crossover, a certain position value exists more than once in the position vectors of the two individuals or then skip the crossover operation for d k(2v-1) and d k(2v) ;

[0093] Suppose in the 0-th generation population, the individuals d 0i and d 0j are paired and it is judged that crossover is required. If a random crossover point β = 2 is generated, the position vectors and parameter vectors of the two individuals are crossed after the random crossover point β = 2, and two new individuals and can be obtained. Among them If the random crossover point is β = 3, then position 5 exists more than once in the position vector so the crossover operation for d 0i and d 0j needs to be skipped;

[0094] S11: Update of the number of generations: k = k + 1

[0095] S12: If k < K, then enter step S07; otherwise enter step S13;

[0096] S13: Temporarily store the optimal result

[0097] The saved individual position vector and parameter vector information is used to restore the row in a block structure and replace the original basis matrix row to form the basis matrix B m to save.

[0098] S14: Optimize row number update: m=m+1

[0099] S15: If m<4, go to step S04; otherwise go to step S16;

[0100] S16: Output the best result

[0101] Output the optimal basis matrix B4 after optimizing the last row.

[0102] S17: Terminate the algorithm.

[0103] The present invention is further described below with reference to a specific embodiment.

[0104] The codewords in the LDPC code BG1 specified in the 5G standard are optimized, with a code rate of R = 1 / 2 and a row number of M. b =24, number of columns N b =46, expansion factor Z = 48. For this matrix, {6, 12, 24, 48} in the expansion factor group with index 2 in Table 1 is selected as the optimized expansion factor group, and the code length is comprehensively optimized. The optimization range only considers Figure 1 In the D section of the optimization test, the initial frame error rate is 10 -5 Around, you can Figure 3 The parameters of the D block represented by Figure 4 result. Figure 5 Comparing the frame error rate curves of each codeword after optimization with those before optimization, it can be seen that the scheme of the present invention has achieved a certain degree of optimization on several codewords in the tested expansion factor group. -4 The following all show good optimization effects, and the optimization effect on some code lengths is obvious.

[0105] Table 1: Expansion factors Z supported by a collection

[0106]

[0107]

[0108] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A genetic optimization method for block-structured LDPC codes. Features: The following steps are involved: Step S01, statistical basis matrix information: It includes a base matrix B of an initial block structure LDPC code to be optimized, its number of rows Mb and number of columns Nb, and an expansion factor group Z involved in the optimization; Step S02, setting test information: Includes the signal-to-noise ratio Γ to evaluate performance 0 and the frame error rate Θ of the initial codeword under the corresponding signal-to-noise ratio 0 ; Step S03, row initialization: Initialize the row to be optimized m=M 1 ; Step S04, initial vector analysis of the row to be optimized: Analyze the row weight ρ of the row to be optimized in the evaluation area; according to the properties of the row of the block structure basis matrix, its non-zero element position and its corresponding shift parameter constitute the position vector p and the parameter vector e, and then form the description vector d = {p, e} of the row as the initial vector for the optimization of the row; Step S05, genetic evolution initialization: Initialize the cyclic algebra k=0, specify the maximum cyclic algebra K, and specify the number of individuals in the population I, where I is a positive even number; Step S06, initializing the population; Step S07, calculating the fitness of the k-th generation population, k = 0, 1, ..., K; Step S08, selecting an operation; Step S09, mutation operation; Step S10, crossover operation; Step S11, loop algebraic update: k=k+1; Step S12, if k<K, proceed to step S07, otherwise proceed to step S13; Step S13, temporarily storing the optimal result: The row is restored according to the block structure by using the saved individual position vector and parameter vector information, and the original basis matrix row is replaced to form the basis matrix Bm, which is saved; Step S14, optimizing the number of rows to update: m=m+1; Step S15, if m<M 2 , then go to step S04, otherwise go to step S16; Step S16, output the optimal result: The optimal basis matrix result after optimizing the last row Output; Step S17, terminating the algorithm; In the step S01, for a block structure LDPC code, its check matrix is ​​represented by a base matrix B with Mb rows and Nb columns and an expansion factor Z. To obtain the final LDPC code, each element bmn of the base matrix B is replaced by a Z×Z all-zero matrix (bmn=-1) or a unit matrix cyclic shift (bmn≥0) mapped by the non-negative element value bmn in the base matrix, 1≤m≤Mb, 1≤n≤Nb; the matrix size of the final LDPC code is defined as M×N, where M=Mb×Z, N=Nb×Z; First, read the base matrix B = [bmn] of the initial block structure LDPC code to be optimized, obtain its number of rows Mb and columns Nb and its expansion factor group Z = {Z 1 ,Z 2 ,…,Z S }, where the element bmn in the basis matrix is ​​the maximum value of the expansion factor Z involved in the optimization S Correspondingly, Z 1 <Z 2 <,…, <Z S ; and obtain the region m∈[M 1 ,M 2 ],n∈[N 1 ,N 2 ], where 1≤M 1 <M 2 ≤Mb, 1≤N 1 <N 2 ≤Nb; In step S02, for each LDPC code check matrix to be optimized, that is, the extension factor group to be optimized applicable to the same base matrix structure Z={Z 1 ,Z 2 ,…,Z S }, set the signal-to-noise ratio Γ for performance evaluation 0 ={γ 1 ,γ 2 ,…,γ S } and the frame error rate Θ of the initial codeword under the corresponding signal-to-noise ratio 0 ={θ 1 ,θ 2 ,…,θ S }.

2. The genetic optimization method for block-structured LDPC codes according to claim 1, Features: In the step S07, for each individual in the population, according to the population individual dki, where i=1, 2, ..., I, the row is restored according to the block structure and replaces the original basis matrix row to form a test basis matrix Bki; on this basis, by expanding each optimized expansion factor Zs, s=1, 2, ..., S, the corresponding test codeword Hki(s) is obtained. For different expansion factors Zs, s=1, 2, ..., S-1, the basis matrix elements are obtained through Z S The basis matrix element bmn corresponding to the factor is obtained by taking modulo mod(bmn,Zs) for different expansion factors; Through decoding simulation test, the frame error rate θki(s) under γs is calculated, and then the comprehensive test parameters are calculated: Wherein, θki(s) is the frame error rate of the matrix of the i-th individual of the k-th generation corresponding to the expansion factor Zs under the signal-to-noise ratio γs, θs is the frame error rate of the original matrix with the expansion factor Zs before optimization tested under the corresponding evaluation signal-to-noise ratio γs, s = 1, 2, …, S; Then calculate its fitness fki=1 / σki,i=1,2,…,I.

3. The genetic optimization method for block-structured LDPC codes according to claim 1, Features: In step S08, for the individuals in the population, the selection probability corresponding to each individual is formed according to the population fitness. Then construct the roulette probability RPk[i] of each individual, set RPk[0]=0, use the iterative formula RPk[i]=RPk[i-1]+SPk[i], i=1,2,…,I, form the roulette probability RPk[1],RPk[2],…,RPk[I], where RPk[I]=1; Generate a random number w within 1, where 0 ≤ w < 1, and determine the roulette area where w is located. If RPk[i - 1] ≤ w < RPk[i], then select the i-th individual and save its description vector dki = {pki, eki}, including the position vector pki and the parameter vector eki information.

4. The genetic optimization method applicable to block-structured LDPC codes according to claim 1, characterized in that: In step S09, for each individual in the population, whether to mutate is determined according to the mutation probability MP, a mutation point α∈[1,ρ] is randomly generated, and the position vector of the individual to be mutated dki={pki,eki} is set to and parameter vector The position and parameters at the mutation point α are respectively mutated to [N 1 ,N 2 ] is different from other positions in the range and in [0,Z S -1] range of parameters Get individual in If there is a value at a position αp other than the variation point in the position vector pki that is different from If they are the same, swap the values ​​of the original position vector at α and αp.

5. The genetic optimization method applicable to block-structured LDPC codes according to claim 1, characterized in that: In step S10, the population is divided into two parts, and the individuals of the population are paired, that is, the 2v-1th and 2vth individuals are paired, v = 1, 2, ..., I / 2; for each pair of individuals, whether to crossover is determined according to the crossover probability CP, and a crossover point β∈[1,ρ] is randomly generated, and the two individuals to be crossed dk (2 v- 1) and dk (2 v ) The position vector and and the parameter vector and The values ​​at β, β+1, …, ρ are exchanged respectively to obtain two individuals after crossover. and in, If after the crossover, a position value is in the position vectors of two individuals or If it exists more than once, skip this pair of dk (2 v- 1) and dk (2 v ) Crossover operation.