Non-cooperative communication QC-LDPC code parameter estimation and check matrix reconstruction method, system and device

Abstract

The present application relates to the technical field of communication signal processing and channel coding identification, in particular to a non-cooperative communication QC-LDPC code parameter estimation and check matrix reconstruction method, system and device. The present application firstly estimates the base structure parameters jointly, generates candidate expansion factors based on code length positive divisors, calculates the optimal expansion factor and initial support matrix through cyclic diagonal fast matching calculation, zero / non-zero block judgment and three-section type segmented comprehensive scoring to compress the search space; then adopts a phased mixed optimization architecture, globally optimizes the support matrix through a genetic algorithm, and finally splices to generate a quasi-cyclic sparse check matrix after fine optimization of the shift value by simulated annealing. The present application can significantly reduce the reconstruction complexity, improve the reconstruction accuracy in full density scenarios, and the reconstructed matrix can be directly used for blind decoding, which is suitable for engineering scenarios such as communication reconnaissance and spectrum monitoring.

Description

Technical Field

[0001] This invention relates to the fields of communication signal processing and channel coding identification technology, specifically a joint coarse estimation method for QC-LDPC code base structure parameters under non-cooperative communication, a reconstruction method for QC-LDPC code sparse parity-check matrix under non-cooperative communication, a reconstruction system for QC-LDPC code sparse parity-check matrix under non-cooperative communication, and a non-cooperative communication signal processing device. Background Technology

[0002] In scenarios such as non-cooperative communication, communication reconnaissance, electronic warfare, and cognitive radio, the receiver cannot obtain the channel coding parameters of the transmitter in advance. It is necessary to restore the coding structure through blind identification technology in order to further realize decoding and information parsing. Therefore, blind identification of channel coding is a core supporting technology in the field of non-cooperative communication signal processing.

[0003] Quasi-cyclic low-density parity-check (QC-LDPC) codes, with their excellent error correction performance and low hardware implementation complexity, have been widely used in mainstream communication standards such as 5G NR and IEEE 802.16e, and are also one of the most common coding types in non-cooperative scenarios. Structurally, the parity-check matrix of a QC-LDPC code is composed of several z×z sub-blocks, where z is called the spread factor, and the matrix used to mark the zero / non-zero distribution of the sub-blocks is called the support matrix. The spread factor and the initial support matrix are the prerequisite parameters for sparse reconstruction of QC-LDPC codes, and their estimation accuracy and efficiency directly determine the engineering usability of the entire reconstruction scheme.

[0004] Existing blind identification methods for linear block codes typically employ Gaussian elimination to solve for the parity-check matrix (PCM). However, these methods result in a dense PCM matrix that is not sparse, with an average row weight reaching several hundred, which does not conform to the sparsity characteristics of LDPC codes. Therefore, it cannot be directly applied to standard decoding procedures and requires prior structural parameter estimation and sparsification reconstruction. For example, patent application CN202310555258.7 discloses an LDPC code parameter identification method that obtains a non-sparse PCM matrix by solving for the dual vectors through Gaussian column transformation, and then reconstructs the sparse PCM matrix using a general sparsification algorithm. This enables open-set blind identification of LDPC codes in non-cooperative scenarios. However, this type of scheme has obvious engineering limitations when dealing with QC-LDPC codes: First, it does not fully utilize the quasi-cyclic block structure characteristics of QC-LDPC codes and does not set up a pre-screening and coarse estimation step for the expansion factor; Second, under the condition of non-cooperative zero prior, the candidate values ​​of the expansion factor of QC-LDPC codes are all the positive divisors of the code length, which can be dozens of kinds. If a complete fine optimization and reconstruction is performed on each candidate, the computational load will increase linearly with the number of candidates, making it difficult to meet the real-time requirements of engineering scenarios such as communication reconnaissance and dynamic spectrum access.

[0005] It is evident that existing blind identification schemes for LDPC codes mostly follow the processing approach of general linear block codes, without designing a preliminary parameter coarse estimation stage for the quasi-cyclic block characteristics of QC-LDPC. This results in excessively high reconstruction computation and insufficient real-time performance in zero-prior scenarios, and the inability to efficiently output structured priors such as expansion factors and initial support matrices. This has become a bottleneck restricting the implementation of QC-LDPC code reconstruction projects.

[0006] After coarsely estimating the basic structural parameters, the process moves to the fine-grained reconstruction stage of the QC-LDPC code sparse parity-check matrix. For the reconstruction optimization of QC-LDPC codes, some existing technologies introduce intelligent optimization algorithms to improve global search capabilities. These algorithms encode the binary values ​​of the support matrix and the shift values ​​of each sub-block as the same set of optimization variables, and then use a single algorithm for synchronous iterative solution. For example, patent application CN202111264759.7 discloses a genetic optimization method applicable to block-structured LDPC codes. Although it optimizes the shift values ​​of block-structured LDPC codes, this scheme assumes that the base matrix support structure is completely known and only uses the shift values ​​as optimization variables. If extended to the reconstruction scenario, the support matrix variables need to be introduced simultaneously and optimized synchronously, which will directly lead to an explosion of the solution space dimensions. The common problems of this type of synchronous optimization mode are mainly reflected in three aspects: First, the 0-1 combination variables of the support matrix and the integer variables of the shift values ​​have significantly different properties. Coupled with the two in the same optimization process, a single algorithm cannot adapt to the optimization characteristics of the two variables at the same time, resulting in an inability to balance the breadth of global search and the accuracy of local optimization. Second, the dimension of the solution space will increase linearly with the size of the base matrix, which not only significantly reduces the convergence speed of the algorithm, but also makes it easy to get trapped in local optima, making it difficult to guarantee the reconstruction effect in high-density matrix scenarios. In addition, this type of scheme generally adopts a completely random initialization method, resulting in a large number of optimization iterations and large fluctuations in results, making it difficult to meet the requirements of engineering applications in terms of reconstruction stability.

[0007] It is evident that existing QC-LDPC code reconstruction schemes mostly adopt a synchronous processing mode that couples the support matrix and the shift value. This mode cannot adapt to the differentiated optimization characteristics of the two types of variables and generally suffers from problems such as slow convergence speed, easy getting trapped in local optima, and insufficient result stability. It is difficult to achieve high-precision and high-efficiency sparse parity matrix reconstruction in non-cooperative scenarios and cannot balance reconstruction accuracy and engineering real-time performance. Summary of the Invention

[0008] To address the technical problem in non-cooperative zero-prior scenarios where the lack of a low-complexity joint coarse estimation method for the QC-LDPC code expansion factor and initial support matrix leads to excessive computational load in subsequent sparse reconstruction, making it difficult to meet the real-time requirements of engineering, this invention provides a joint coarse estimation method for the basic structural parameters of QC-LDPC codes under non-cooperative communication.

[0009] Furthermore, in order to address the technical problem that existing QC-LDPC code sparse reconstruction adopts a synchronous optimization mode that couples the support matrix and shift value, resulting in poor algorithm adaptability, slow convergence, and easy getting trapped in local optima, making it impossible to balance reconstruction accuracy and efficiency, this invention provides a method for reconstructing the sparse parity check matrix of QC-LDPC codes under non-cooperative communication.

[0010] Based on the above two schemes, the present invention also provides a QC-LDPC code sparse parity-check matrix reconstruction system under non-cooperative communication and a non-cooperative communication signal processing device.

[0011] To achieve the above objectives, the present invention provides the following technical solution: A joint coarse estimation method for QC-LDPC code base structure parameters under non-cooperative communication is applied to channel coding blind identification scenarios. The method estimates parameters based on the dense parity-check matrix obtained by Gaussian elimination of the intercepted signal, and includes the following steps: Based on the code length of the dense parity check matrix, generate a set of all candidate expansion factors, where each candidate expansion factor is a positive divisor of the code length; Iterate through each candidate expansion factor and divide the dense parity check matrix into sub-block arrays of corresponding sizes according to the current expansion factor. For each sub-block, adopt a circular diagonal mapping method and take the difference between the column index and row index in the block modulo the current expansion factor. Iterate through all elements in the sub-block in one go and accumulate them according to the shift value to directly obtain the matching score of the sub-block under all shift values. Extract the maximum matching score of each sub-block. Based on the preset non-zero sub-block determination threshold, the zero / non-zero attributes of each sub-block are determined according to the maximum matching score of each sub-block, and the initial support matrix corresponding to the current candidate expansion factor is generated. A comprehensive scoring function is used to score the current candidate expansion factor and its corresponding initial support matrix; The candidate expansion factor with the highest comprehensive score is selected as the optimal expansion factor, and its initial support matrix is ​​output accordingly. This matrix serves as the structural prior for subsequent sparse reconstruction, supporting blind decoding processing under non-cooperative communication.

[0012] As a further improvement to the above scheme, the comprehensive scoring function is specifically expressed as follows: ; ; In the formula, score is the overall score; α is the piecewise weight coefficient; exp(·) is the natural exponential function; ρ is the support matrix density, i.e., the proportion of non-zero elements in the support matrix under the current candidate expansion factor; d est The LDPC code matrix density is estimated empirically; σ is the standard deviation of the Gaussian function, used to control the smoothness of score decay; S i,jTo support the element in the i-th row and j-th column of the matrix, the value is 0 or 1, where 1 represents that the element is in a non-zero sub-block and 0 represents a zero sub-block; match i,j (s) is the sum of the corresponding elements of the (i,j)th sub-block and the standard cyclic permutation matrix with shift value s; The (i,j)th sub-block represents the highest matching score achievable with the cyclic permutation matrix among all possible shift values ​​s; support is the set of sub-blocks in the support matrix that are determined to be non-zero; |·| is the absolute value operation.

[0013] As a further improvement to the above scheme: the matching score ranges from 0 to the value of the current candidate expansion factor. The higher the score, the higher the structural overlap between the sub-block and the cyclic permutation matrix. When the maximum matching score of a sub-block is less than 0.5 times the current candidate expansion factor, the sub-block is directly determined as a zero sub-block and is not included in the subsequent zero / non-zero sub-block determination process. Conversely, when the maximum matching score of a sub-block is greater than or equal to 0.5 times the current candidate expansion factor, the sub-block enters the subsequent determination stage, based on the non-zero sub-block determination threshold coefficient. Further determine its zero / non-zero properties.

[0014] As a further improvement to the above scheme: when calculating the matching score of each sub-block, the optimal shift value corresponding to the maximum matching score of each sub-block is recorded simultaneously; while outputting the initial support matrix, the optimal shift values ​​corresponding to all non-zero sub-blocks are output simultaneously and organized into an initial shift vector according to row and column indices; this initial shift vector serves as the structured hot start point for local fine optimization of shift values ​​in subsequent sparse reconstruction, replacing the random initialization method.

[0015] A method for reconstructing a sparse parity-check matrix of a QC-LDPC code under non-cooperative communication involves taking the encoded signal intercepted in a non-cooperative communication scenario, demodulating it, and performing Gaussian elimination to obtain a dense parity-check matrix, followed by the following steps: By performing a joint coarse estimation method for the base structure parameters of QC-LDPC codes under non-cooperative communication, the optimal spread factor and the corresponding initial support matrix are obtained by performing a joint coarse estimation of the base structure parameters of the dense parity check matrix. A semi-deterministic initial population is constructed based on the initial support matrix. A genetic algorithm adapted to the binary support structure is used to globally optimize the distribution of non-zero sub-blocks in the support matrix and output the optimal support matrix. The positions of the non-zero sub-blocks in the fixed optimal support matrix are used, and the cyclic shift values ​​of each non-zero sub-block are locally fine-tuned using the simulated annealing algorithm with adaptive integer shift fine-tuning, and the optimal shift value matrix is ​​output. Based on the optimal expansion factor, optimal support matrix, and optimal shift value matrix, a sparse parity check matrix conforming to a quasi-cyclic structure is generated by concatenating sub-blocks. The sparse parity check matrix is ​​directly input into the LDPC decoder for blind decoding and information recovery of non-cooperative communication signals.

[0016] As a further improvement to the above scheme, global optimization includes the following steps: Fitness is calculated for each candidate support matrix individual in the population. Fitness is used to measure the degree of fit between the candidate support matrix and the structural characteristics of the QC-LDPC code. An elite retention strategy is adopted, in which the two individuals with the highest fitness in the current population are directly retained to the next generation. A tournament selection mechanism was used to select parent individuals, and a new generation population was generated through row and column crossover and site mutation operations. Repeated iterative evolution continues until no more individuals with better fitness can be generated or the preset maximum number of generations is reached. The individual with the highest fitness in the last generation population is selected as the optimal support matrix.

[0017] As a further improvement to the above scheme, local fine-tuning specifically includes the following steps: Set the initial temperature, annealing rate and maximum number of iterations for simulated annealing, and use the initial shift vector as the starting point for optimization. At the current temperature, a non-zero sub-block is randomly selected and its shift value is adjusted to generate a candidate shift vector. The energy change between the candidate shift vector and the current shift vector is calculated. After completing a preset number of iterations at the current temperature, a cooling operation is performed. The iterations are repeated until the shift value can no longer be optimized or the maximum number of iterations is reached, and the optimal shift value matrix is ​​output.

[0018] As a further improvement to the above scheme: after calculating the energy change, the Metropolis criterion is followed to determine whether to accept the candidate shift vector. If the energy value corresponding to the candidate shift vector is less than the energy value of the current shift vector, then the candidate shift vector is accepted directly. If the energy value corresponding to the candidate shift vector is greater than or equal to the energy value of the current shift vector, then the candidate shift vector is accepted with probability exp(-Δf / T), where Δf is the energy change and T is the current temperature.

[0019] A sparse parity-check matrix reconstruction system for QC-LDPC codes under non-cooperative communication is provided, comprising a joint coarse estimation method for the basis structure parameters of QC-LDPC codes under non-cooperative communication, and a reconstruction method for the sparse parity-check matrix of QC-LDPC codes under non-cooperative communication, including: The basic structure parameter coarse estimation module is used to receive the dense parity check matrix obtained by Gaussian elimination, generate a set of all candidate expansion factors based on the code length of the dense parity check matrix, traverse each candidate expansion factor to complete sub-block partitioning, matching score calculation, zero / non-zero sub-block determination and comprehensive scoring, and output the optimal expansion factor and the corresponding initial support matrix. The global optimization module for the support matrix is ​​used to construct a semi-deterministic initial population based on the initial support matrix. It employs a genetic algorithm adapted to the binary support structure to globally optimize the distribution of non-zero sub-blocks in the support matrix and outputs the optimal support matrix. The shift value local optimization module is used to fix the positions of non-zero sub-blocks in the optimal support matrix. It uses a simulated annealing algorithm with adaptive integer shift fine-tuning to perform local fine optimization of the cyclic shift values ​​of each non-zero sub-block and outputs the optimal shift value matrix. The parity check matrix splicing output module is used to generate a sparse parity check matrix that conforms to a quasi-cyclic structure by splicing sub-blocks according to the optimal expansion factor, optimal support matrix and optimal shift value matrix, and output it to the LDPC decoding unit to support blind decoding processing of non-cooperative communication signals.

[0020] A non-cooperative communication signal processing device includes a QC-LDPC code sparse parity-check matrix reconstruction system under non-cooperative communication, and also includes a signal receiving and demodulation unit, a memory, and a processor. The signal receiving and demodulation unit is used to intercept radio frequency coded signals in non-cooperative communication scenarios. The signals are then demodulated, hard-decision, and Gaussian elimination dual space solution is performed to obtain a dense parity check matrix, which is then transmitted to the reconstruction system. Memory is used to store intermediate data during computer programs and signal processing; The processor loads and runs the QC-LDPC code sparse parity-check matrix reconstruction system, which is connected to the signal receiving and demodulation unit and the memory, respectively, and is used to schedule each module to execute the corresponding parameter estimation and matrix reconstruction process.

[0021] Compared with the prior art, the beneficial effects of the present invention are: 1. In the joint coarse estimation method, based on the inherent structural constraints of QC-LDPC codes, low-complexity joint coarse estimation of parameters is achieved through hierarchical dimensionality reduction and two-dimensional evaluation: First, utilizing the structural rule that "the expansion factor must be a positive divisor of the code length," the unbounded search of the expansion factor is converged to a finite candidate set, thus limiting the search range from the source; Second, the dense parity check matrix is ​​divided into sub-blocks according to the candidate expansion factors. By matching the sub-blocks with the standard cyclic permutation matrix, the quasi-cyclic structure identification is transformed into a fast-calculating matching score, retaining only the maximum matching score of the sub-block as the judgment criterion to control computational overhead; In particular, the fast matching method of cyclic diagonal mapping is adopted, which transforms the matching calculation of a single sub-block from bit-by-bit shifting... The complex traversal is reduced to a single element-wise traversal, decreasing the computational load by an order of magnitude and further compressing the computation time of the coarse estimation stage. Subsequently, based on a preset non-zero sub-block determination threshold, the zero / non-zero attribute of the sub-blocks is determined, generating an initial support matrix. This compresses the search space for subsequent optimization from the full matrix element level to the base matrix sub-block level, achieving an order-of-magnitude reduction in search size. Then, by fusing a comprehensive scoring function of density compliance and structural matching, and relying on the inherent properties of sparsity and quasi-cyclicity of QC-LDPC codes, the quantization and sorting of all candidates can be completed without complex iterations. Finally, the optimal expansion factor and initial support matrix are output, providing a high-quality structural prior for subsequent sparse reconstruction and avoiding redundant iterations caused by random initialization. In particular, the optimal shift value of the non-zero sub-blocks output by this method constitutes the structured hot-start initial solution for subsequent shift value optimization, transforming the originally blind and discrete shift search into a deterministic local fine-tuning problem. This significantly reduces the effective search radius of the simulated annealing algorithm and ensures the consistency of multiple batch processing results, providing reliable deterministic input for engineering deployment. The entire coarse estimation process consists of only three low-complexity operations: block operation, threshold determination, and weighted summation. It can complete the initial screening of parameters and spatial dimensionality reduction under zero prior conditions, effectively reducing the computational load of subsequent sparse reconstruction and supporting the real-time processing requirements in engineering scenarios.

[0022] 2. In the reconstruction method, this invention adopts an architecture of "pre-evaluation coarse estimation dimensionality reduction + staged decoupling optimization": the 0-1 binary distribution of the support matrix is ​​a discontinuous, high-dimensional combinatorial optimization problem with a large number of local extrema in its solution space, requiring algorithms with strong global exploration capabilities; while the cyclic shift value, after fixing the support structure, degenerates into a low-dimensional, smooth integer continuous fine-tuning problem, relying more on high-precision local mining. To address this conflict of "heterogeneous variables," this invention forcibly decouples two types of variables: In the first stage, the support matrix is ​​modeled as an independent optimization objective, and a genetic algorithm adapted to binary encoding is used to perform a global space search to accurately locate the optimal support distribution. In the second stage, with the supporting structure completely locked, the problem is reduced to a smooth space containing only integer shift values, and the simulated annealing algorithm is used for fine local optimization.

[0023] This forced decoupling and phased adaptation strategy ensures that each of the two optimization objectives is matched with the optimal solver. From an architectural design perspective, it fundamentally avoids the dual shortcomings of traditional coupled optimization models: insufficient fine-tuning of genetic algorithms and inefficient global exploration via simulated annealing. This achieves Pareto optimality in both the breadth of global exploration and the depth of local development. Furthermore, because the solution space dimension of each stage is reduced by an order of magnitude compared to coupled optimization models, the overall convergence speed is significantly improved, effectively addressing the technical pain points of poor adaptability, slow convergence, and the difficulty in balancing accuracy and efficiency in coupled optimization models. First, by using a coarse estimation method combined with pre-defined base structure parameters, the optimal expansion factor and high-quality initial support matrix are quickly output under zero prior conditions, replacing the traditional random initialization method and avoiding the redundant overhead caused by blind iteration, thus providing a high-quality starting point for subsequent optimization. Second, the originally coupled support structure optimization and shift value optimization are decomposed into two independent stages. For the binary discrete combination characteristics of the support matrix, a genetic algorithm adapted to the binary structure is used to perform global optimization, fully leveraging the global exploration capability of the genetic algorithm in complex non-convex search spaces to accurately locate the optimal support distribution and avoid the shortcomings of insufficient global search capability in coupled optimization. After determining the optimal support matrix, the positions of non-zero sub-blocks are fixed, and only for the integer fine-tuning problem of cyclic shift values, a simulated annealing algorithm is used to perform local fine optimization. Relying on the probabilistic escape mechanism of the Metropolis acceptance criterion, this avoids getting trapped in local optima. At the same time, the search dimension of the shift optimization is significantly reduced after decoupling, significantly improving the convergence speed. Finally, a standard sparse parity-check matrix conforming to a quasi-cyclic structure is generated by splicing sub-blocks. This phased decoupling architecture allows each of the two types of optimization problems to be matched with the most suitable algorithm, which not only ensures the reconstruction accuracy in full-density scenarios, but also significantly shortens the overall processing time, effectively balancing reconstruction accuracy and engineering efficiency. It effectively solves the technical pain points of poor adaptability, slow convergence, and difficulty in balancing accuracy and efficiency under coupled optimization mode. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall process of the QC-LDPC code sparse parity-check matrix reconstruction method under non-cooperative communication.

[0025] Figure 2 This is a schematic diagram of the joint coarse estimation method for QC-LDPC code base structure parameters under non-cooperative communication.

[0026] Figure 3 A schematic diagram illustrating the global optimization process of the matrix genetic algorithm.

[0027] Figure 4 A schematic diagram of the local fine-tuning process for cyclic shift values ​​simulated annealing.

[0028] Figure 5 A schematic diagram of the initial support matrix distribution obtained by joint coarse estimation of the basic structural parameters.

[0029] Figure 6 This is a schematic diagram of the optimal support matrix distribution obtained by global optimization using a genetic algorithm.

[0030] Figure 7 A schematic diagram of the reconstruction result of the sparse parity check matrix of QC-LDPC code.

[0031] Figure 8 This is a schematic diagram of the row redistribution of the dense parity-check matrix obtained by Gaussian elimination.

[0032] Figure 9 This is a schematic diagram of the row redistribution of the reconstructed sparse parity-check matrix.

[0033] Figure 10 This diagram illustrates the comparison of reconstruction rates between two reconstruction methods under different matrix densities.

[0034] Figure 11 This diagram illustrates the comparison of reconstruction factors between two reconstruction methods under different matrix densities. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] like Figures 1-4 As shown, a typical application scenario of this embodiment is non-cooperative communication reconnaissance and spectrum monitoring: the receiver has no prior coding parameters, intercepts the QC-LDPC encoded radio frequency signal in space through the antenna, and obtains the received bit sequence through radio frequency demodulation and hard decision processing in sequence. Then, the dual space is solved by Gaussian elimination to obtain the dense parity check matrix. On this basis, the joint coarse estimation of the base structure parameters and the sparse parity check matrix reconstruction process of this embodiment are executed, and finally the standard sparse parity check matrix that conforms to the quasi-cyclic structure is output, which can be directly input into the LDPC decoder to complete blind decoding and information recovery.

[0037] I. Joint Coarse Estimation Method for QC-LDPC Code Base Structure Parameters under Non-Cooperative Communication

[0038] like Figure 2As shown, this joint coarse estimation method takes the dense parity-check matrix obtained by Gaussian elimination of the intercepted signal as input, and outputs the optimal expansion factor and initial support matrix to compress the search space for subsequent reconstruction. This joint coarse estimation method can be executed independently or as a pre-processing step for sparse reconstruction. The specific implementation steps are as follows: (a) Generating a set of candidate expansion factors Obtain the number of columns (i.e., the code length) N of the dense parity-check matrix. Based on the structural characteristics of QC-LDPC codes, where the parity-check matrix is ​​composed of z×z sub-blocks, the expansion factor z must be a positive divisor of the code length N. Perform positive divisor decomposition on N, i.e., find all the positive divisors of N, obtaining all positive integers that satisfy the divisibility condition, forming a candidate expansion factor set Z = {z1, z2, ..., z...}. k}, z1, z2, z k These are the 1st, 2nd, and kth expansion factors, respectively.

[0039] Taking IEEE 802.16e as an example, its positive divisors are the expansion factors, including 24, 28, 32, 36, 40, etc. All of the above values ​​are included in the candidate expansion factor set Z and evaluated one by one.

[0040] (II) Sub-block partitioning and matching score calculation

[0041] For each candidate value z in the candidate expansion factor set Z, perform the following processing: The dense parity-check matrix is ​​divided into rows and columns with a step size of z, resulting in a matrix containing M... b ×N b A sub-block array of M sub-blocks, where M b N is the quotient of the number of rows in the dense parity-check matrix and z. b The value is the quotient of code length N and z, and the size of each sub-block is z×z.

[0042] For the (i,j)th sub-block H in the array ij The matching score for all shift values ​​is quickly calculated using a circular diagonal mapping method: For a standard cyclic permutation matrix of size z×z, all "1" elements in the matrix with shift value s satisfy the distribution rule of "column index - row index = s(mod z)". Therefore, the result of taking the difference between the column index and the row index of each element in the sub-block modulo z is mapped to the corresponding shift value. mod represents the modulo operation.

[0043] By iterating through all elements within a sub-block in a single pass and accumulating the element values ​​according to their shift values, the matching score for all shift values ​​s (s∈{0,1,…,z-1}) can be obtained at once. i,j (s).

[0044] The physical meaning of the matching score is the number of overlapping positions of "1" elements in the two matrices, with a value range of 0 to z. A higher score indicates that the sub-block better conforms to the quasi-cyclic structure characteristics. When the matching score is below 0.5z, it is considered a non-match and the sub-block is removed; otherwise, it is retained.

[0045] After iterating through all shift values ​​for each sub-block, record the maximum matching score for that sub-block. .

[0046] (III) Generating the initial support matrix

[0047] Preset non-zero sub-block determination threshold coefficient (Value range 0.6~0.8, preferred in this embodiment is 0.7), The product of the product with the current candidate expansion factor z is used as the threshold for determining non-zero sub-blocks.

[0048] Each sub-block is compared with its maximum matching score and a decision threshold: if the maximum matching score is greater than or equal to the decision threshold, the sub-block is determined to be a non-zero sub-block, and the corresponding support matrix element S is determined. ij The value is 1; if the maximum matching score is less than the judgment threshold, the sub-block is judged as a zero sub-block, and the corresponding support matrix element Sij is set to 0. The first level, 0.5z, is a fast pre-screening threshold used to directly remove zero sub-blocks that clearly do not have a quasi-cyclic structure, reducing subsequent computation; the second level... To accurately determine the threshold and ensure the recognition accuracy of non-zero sub-blocks, the two-level threshold balances efficiency and accuracy.

[0049] After traversing all sub-blocks, generate the 0-1 initial support matrix S corresponding to the current candidate expansion factor z.

[0050] (iv) Comprehensive scoring of candidate structures

[0051] A comprehensive scoring function combining fusion density prior and matching degree is used to quantitatively evaluate the current candidate expansion factor and its corresponding initial support matrix. The formula for the comprehensive scoring function is as follows: ; ; Among them, based on the support matrix density ρ and the empirically predicted density d est degree of deviation Divide into multiple density intervals.

[0052] In the formula, score is the comprehensive score; exp(·) is the natural exponential function; support is the set of sub-blocks in the support matrix that are judged to be non-zero; α is the piecewise weight coefficient, with a value range of 0.35~0.65, and 0.5 is preferred in this embodiment. It is used to balance the weight ratio of density conformity and sub-block matching degree. The greater the deviation of density from the target value, the lower the weight of the density conformity term and the relatively higher the weight of the structural matching degree term, thereby avoiding misjudgment of candidate expansion factors caused by density prior bias.

[0053] when When the value is very small, it indicates that the support matrix density corresponding to the candidate expansion factor is very close to the density of the typical QC-LDPC basis matrix, and the reliability of the density prior is higher. Therefore, increasing the value of the weight coefficient α increases the score proportion of the density conformity term; when... A large α indicates that the density prior may be inaccurate due to code rate, standard differences, or noise interference. In this case, reducing α increases the weight of the structure matching term, relying more on the actual matching degree between the sub-block and the cyclic permutation matrix; d represents the density of the current support matrix, i.e., the ratio of the number of non-zero elements in the support matrix to the total number of elements, reflecting the sparsity of the base matrix. est The LDPC code matrix density is estimated based on experience, with a value ranging from 0.01 to 0.05. In this embodiment, a value of 0.025 (d) is preferred. est The results are derived from engineering statistics of QC-LDPC code parity-check matrix density in mainstream communication standards such as IEEE 802.16e, 5G NR, and DVB-S2. The parity-check matrix density of QC-LDPC codes in these standards is generally distributed in the range of 0.01 to 0.05. Therefore, the midpoint of the range, 0.025, is taken as the preferred value for general scenarios. If there is prior information about known coding standards in the application scenario, the target density value can be adjusted accordingly to improve matching accuracy. σ is a scale parameter used to control the penalty gradient for density deviation, with a value range of 0.05 to 0.2. In this embodiment, 0.1 is preferred.

[0054] Left side item The density compliance term is constructed using a Gaussian function: the closer the support matrix density is to the typical density of QC-LDPC codes in engineering (0.025), the higher the score of this term; the further the density deviates, the score decreases smoothly, achieving a soft constraint effect.

[0055] Right side item The average matching degree term is the average of the maximum matching scores of all non-zero sub-blocks, reflecting the degree of fit between the non-zero sub-blocks in the support matrix and the quasi-cyclic structure.

[0056] To quantitatively evaluate the degree of agreement between the current candidate expansion factors and their corresponding initial support matrices and the actual QC-LDPC code structure, this embodiment constructs a comprehensive scoring function based on the two inherent structural properties of QC-LDPC codes: sparsity and quasi-cyclicity. The specific construction process is as follows: 1. Basis for designing the scoring function The quasi-cyclic LDPC parity-check matrix has a hierarchical block cyclic structure, and its core structural constraints include two points: One is the sparsity constraint: the basis matrix (support matrix) of QC-LDPC code naturally has extremely low density. In engineering, the parity check matrix density of mainstream QC-LDPC codes is usually concentrated in the range of 0.01 to 0.05. Density feature is one of the core dimensions for distinguishing whether candidate expansion factors are reasonable. Secondly, there is the quasi-cyclic constraint: each non-zero sub-block must be a cyclic permutation matrix obtained by cyclically shifting the identity matrix. The degree of matching between the sub-block and the standard cyclic permutation matrix directly reflects the compliance of the structure under this block partitioning method.

[0057] Single-dimensional evaluation is susceptible to noise and matrix error, which can lead to misjudgment. Therefore, this embodiment quantifies the two types of constraints into independent evaluation indicators, and forms a comprehensive score through weighted fusion. The optimal value that best fits the real QC-LDPC structure is selected from all candidate expansion factors.

[0058] (2) Derivation of density adaptation index (first item)

[0059] To achieve the soft constraint effect of "higher scores for density closer to typical values ​​and smoother score decay for greater deviations," a Gaussian function is used to construct the density fit index score. ρ The formula is in the form of: ; In the formula, ρ is the density of the support matrix under the current candidate expansion factor, defined as the ratio of the total number of non-zero sub-blocks to the total number of sub-blocks in the support matrix, i.e., ρ = N. nz / (M b ×N b );N nz M represents the total number of non-zero sub-blocks. b N b These represent the number of rows and columns of the support matrix, respectively.

[0060] d target As a reference value for the target density, the typical density center value of 0.025 of the QC-LDPC code parity check matrix in the project is selected. If there is prior density information for a specific scenario, this value can be adjusted. σ is the standard deviation of the Gaussian function, used to control the smoothness of the score decay. In this embodiment, σ is fixed at 0.1, which allows the density to fluctuate slightly within a certain range while still maintaining a high score, thus improving the scenario adaptability of the solution.

[0061] When the support matrix density is exactly equal to the target density of 0.025, the density fit index reaches its maximum value of 1; the further the density deviates from the target value, the closer the index score is to 0.

[0062] (3) Derivation of the structural matching index (second item)

[0063] To quantify the degree of fit between non-zero sub-blocks and quasi-cyclic structures, a structural matching index is constructed based on the matching scores between sub-blocks and cyclic permutation matrices.

[0064] For each sub-block after partitioning, iterate through all shift values ​​s∈[0,z-1] and calculate the matching score match between the sub-block and the standard cyclic permutation matrix of the corresponding shift value. i,j (s), taking the maximum value under all shifts. This serves as the cyclic similarity of the sub-block.

[0065] To eliminate scoring bias caused by differences in the number of non-zero sub-blocks, only sub-blocks with non-zero support matrices are counted, and the average of the cycle similarity of all non-zero sub-blocks is taken to obtain the structure matching score. match : ; This index approaches 1 when the sub-block perfectly matches the cyclic permutation matrix; it approaches 0 when it does not match at all.

[0066] (4) Integration of comprehensive scoring functions

[0067] The density fit index and structure matching index are fused using a linear weighting method. Weighting coefficients α and 1-α are assigned to balance the importance of each index, ultimately yielding a comprehensive scoring function: ; target density d target Replace with empirically estimated LDPC code matrix density d est We can obtain: ; Substituting the typical parameter d in this embodiment est =0.025, σ=0.1, can be expanded as follows: ; The comprehensive score ranges from (0,1). A higher score indicates that the structure corresponding to the candidate expansion factor is more consistent with the dual structure characteristics of the QC-LDPC code.

[0068] Therefore, this comprehensive scoring function deeply integrates the two inherent structural properties of QC-LDPC codes—sparseness and quasi-cyclicity—and adopts a two-dimensional evaluation architecture with linear weighting of density-adaptive Gaussian terms and structural matching terms. Compared with single-dimensional evaluation methods, it has significant performance and engineering advantages. The dual-index cross-constraint significantly improves the scheme's robustness against interference. It avoids the susceptibility to noise and bit error interference leading to misjudgment of expansion factors when relying solely on sub-block matching, and also compensates for the shortcomings of relying solely on density features, which cannot distinguish structural details and is prone to candidate confusion. The two types of structural features mutually verify each other, stably selecting candidate expansion factors that best fit the real coding structure in zero-prior open-set scenarios. The density constraint uses a Gaussian function to achieve soft decision-making, which has a smooth decay characteristic compared to hard threshold judgment. This ensures that candidates with densities close to typical values ​​achieve high scores, while allowing for high scores even when densities fluctuate slightly within a reasonable range, thus improving adaptability to QC-LDPC codes of different standards and code rates. It is stronger, and the target density and standard deviation parameters can be flexibly adjusted according to the scenario prior. The structure matching degree term adopts the normalization design of the average matching score of non-zero sub-blocks, which eliminates the scoring bias caused by the difference in the number of non-zero sub-blocks under different candidate expansion factors, and ensures that the scores among candidates are fair and comparable. At the same time, the overall computational complexity of the function is low, which only includes exponential operation, summation and linear weighting operation. It does not require iterative optimization and can quickly complete the quantitative scoring of all candidate expansion factors. It is fully adapted to the low complexity and fast response of engineering positioning in the coarse estimation stage. It can significantly compress the subsequent reconstruction search space while ensuring the accuracy of parameter estimation, and provide a high-quality initial structure prior for subsequent sparse reconstruction.

[0069] (V) Optimal Parameter Selection and Output

[0070] After traversing all candidate expansion factors and calculating the comprehensive score, the candidate expansion factor with the highest comprehensive score is selected as the optimal expansion factor z. best Synchronously output its corresponding initial support matrix S init .

[0071] The output optimal expansion factor and initial support matrix are directly used as the structural prior for subsequent sparse reconstruction, which can compress the search space of subsequent optimization by more than 90% and significantly reduce the amount of reconstruction computation. At the same time, this set of parameters can also be used independently for other blind identification links such as coding type determination and coarse code rate estimation, supporting blind decoding processing under non-cooperative communication.

[0072] II. Reconstruction Method of Sparse Parity-Check Matrix of QC-LDPC Code under Non-Cooperative Communication

[0073] This reconstruction method takes a dense parity-check matrix as input, connects the above-mentioned coarse estimation process of basic structure parameters, and adopts a staged architecture that first supports global optimization and then shifts local optimization to finally output a decodeable sparse parity-check matrix, forming a complete reconstruction closed loop from signal to decoding. The specific implementation steps are as follows: (a) Joint coarse estimation of basic structural parameters The joint coarse estimation method for QC-LDPC code basis structure parameters described above is used to process the input dense parity-check matrix to obtain the optimal spread factor z. best The corresponding initial support matrix S init Simultaneously, it can output the optimal shift value corresponding to each non-zero sub-block, forming the initial shift vector p. init This provides a high-quality starting point for subsequent shift optimization. Simultaneously, the optimal shift values ​​corresponding to each non-zero sub-block are arranged according to their spatial indices in the support matrix to construct a structured initial shift vector p. init This vector constitutes deterministic prior knowledge in the shift optimization stage, representing a 'warm-start' initialization mechanism distinct from traditional random generation strategies. Through this mechanism, the initial energy value for subsequent simulated annealing optimization lies within the attraction domain of the global optimum. Compared to random initialization, this mechanism reduces the number of iterations required for algorithm convergence by approximately 20%–30%, and significantly reduces the variance of multiple independent runs, providing a high-quality, low-uncertainty starting point for subsequent fine-grained shift optimization.

[0074] (ii) Global optimization of the support matrix

[0075] like Figure 3 As shown, with the initial support matrix S init A semi-deterministic initial population is constructed based on this foundation. A genetic algorithm adapted to the binary support structure is used to perform a global search optimization on the distribution of non-zero sub-blocks in the support matrix. The specific process is as follows: Population initialization: Set the population size to 50, with 50% of the individuals directly using the initial support matrix S. init The remaining 50% of individuals, through S init The elements are randomly flipped within the range of 10% to 30%. In this embodiment, the flipping probability is set to 20% to ensure that the population retains both high-quality initial priors and sufficient diversity.

[0076] Fitness Calculation: For each candidate support matrix individual in the population, its fitness value is calculated. The core indicator of fitness is the "satisfaction rate of the check equation between the corresponding structure of the support matrix and the original dense matrix." A higher fitness indicates that the support structure is closer to the true coding structure. This satisfaction rate measures the degree of agreement between the quasi-cyclic check structure corresponding to the candidate support matrix and the original dense check matrix. The calculation logic is as follows: For the quasi-cyclic check structure constructed by the candidate support matrix and the current expansion factor, the orthogonality of its row vectors with the corresponding row vectors of the original dense check matrix is ​​checked row by row, i.e., whether the check equation is valid. The proportion of rows that satisfy the orthogonality relationship is counted out of the total number of rows, which is the satisfaction rate. A higher satisfaction rate indicates that the structure of the candidate support matrix is ​​closer to the true coding structure, which is the core basis for judging the quality of individuals in the genetic algorithm.

[0077] Iterative evolution: In each generation of evolution, an elite retention strategy is adopted, and the two individuals with the highest fitness in the current population are directly copied to the next generation population; the remaining individuals are selected for the parent generation through a tournament selection mechanism, and three individuals are randomly selected from each group, with the one with the highest fitness retained as the parent generation; the parent individuals generate offspring individuals through row and column crossover and site mutation operations, with the mutation probability set to 0.05.

[0078] Convergence Output: Iterative evolution continues until no individual with better fitness emerges for 10 consecutive generations, or until the preset maximum number of generations is reached (50 generations in this example). The iteration then stops, and the individual with the highest fitness in the last generation is selected as the optimal support matrix S. best After obtaining the optimal support matrix S~best~ through the above genetic iteration, the first stage of global support matrix optimization is completed. At this point, the QC-LDPC code reconstruction problem is precisely reduced from a large space of 0-1 binary combinations to an 'integer continuous fine-tuning' space. Based on this, the second stage begins, where local fine-tuning is performed on the shift values ​​of the locked support positions: (III) Local fine-tuning of shift values like Figure 4 As shown, the optimal support matrix S is fixed. best For the non-zero sub-block positions, only the cyclic shift values ​​of each non-zero sub-block are optimized. A simulated annealing algorithm with adaptive integer shift fine-tuning is used to achieve local fine-tuning. The specific process is as follows: Parameter initialization: Set the initial simulated annealing temperature T0=100, annealing rate β=0.95, iterations at each temperature to 20, and the maximum total number of iterations. The initial shift vector p output by the coarse estimation stage. init As a starting point for optimization.

[0079] Neighborhood search: At the current temperature, randomly select a non-zero sub-block and shift its value between 0 and z.best Randomly increment or decrement by 1 within the range of -1 (mod z) best (operation), generating candidate shift vector p candidate ; Calculate the candidate shift vector p candidate The energy difference Δf is defined as the sum of the matching errors between the shifted sub-block structure and the original dense sub-block. A lower energy value indicates a higher structural fit. The energy value itself measures the degree of structural matching error between the sub-block structure corresponding to the current shift vector and the sub-block corresponding to the original dense parity-check matrix. The calculation logic is as follows: For each non-zero sub-block, a standard cyclic permutation sub-matrix is ​​constructed according to the current shift value. This sub-matrix is ​​then compared element-by-element with the corresponding sub-block in the original dense parity-check matrix. The number of positions where the element values ​​are inconsistent is counted, which is the matching error of that sub-block. The matching errors of all non-zero sub-blocks are summed to obtain the total energy value corresponding to the current shift vector. The energy difference Δf is the difference between the total energy value corresponding to the candidate shift vector and the total energy value corresponding to the current shift vector. A negative energy difference indicates a higher structural fit of the candidate shift; a positive energy difference indicates a decreased fit. The simulated annealing algorithm uses this energy difference in conjunction with the Metropolis criterion to determine whether to accept the new shift value.

[0080] Acceptance Decision: The Metropolis criterion is followed to determine whether to accept the candidate shift vector: if Δf < 0, meaning the candidate solution has better energy, then the candidate shift vector p is directly accepted. candidate If Δf≥0, then with probability Accept candidate shift vectors (T is the current temperature, e is the natural exponential function), and avoid the algorithm getting trapped in local optima by probabilistically accepting inferior solutions.

[0081] Cooling convergence: After completing all iterations at the current temperature, perform a cooling operation T = β·T, which is the current temperature multiplied by the annealing rate to obtain the temperature for the next round; repeat the iteration until the temperature drops below 0.01, or the maximum total number of iterations is reached, and output the optimal shift value matrix P. best .

[0082] (iv) Sparse parity check matrix splicing and output

[0083] Based on the optimal expansion factor z best Optimal support matrix S best With the optimal shift value matrix P best The final sparse parity-check matrix is ​​generated by concatenating sub-blocks: Iterate through each position of the support matrix, if the optimal support matrix S best The (i,j)th element S in best If (i,j)=0, then the corresponding position has a construction size of z. best ×z bestA matrix of all zeros; if S best If (i,j)=1, then the corresponding position is constructed with a shift value of P. best (i,j) (Optimal shift matrix P) best The cyclic permutation matrix of the (i,j)th element in the matrix is ​​obtained. All submatrices are concatenated in row and column order to obtain the QC-LDPC sparse parity-check matrix that conforms to the quasi-cyclic structure.

[0084] The generated sparse parity-check matrix is ​​directly input into the LDPC decoder to perform blind decoding on the intercepted non-cooperative communication signals, recover the original information bits, and complete the complete engineering link from signal interception to information parsing.

[0085] III. Simulation Verification and Effect Analysis

[0086] To verify the effectiveness and engineering performance of the joint coarse estimation method for QC-LDPC code basis structure parameters under non-cooperative communication and the reconstruction method for the sparse parity-check matrix of QC-LDPC codes under non-cooperative communication, this embodiment conducts comparative experiments based on the Matlab simulation platform. The test object adopts the QC-LDPC code defined in the IEEE 802.16e standard, with specific parameters as follows: code length 648, information bit length 324, and native parity-check matrix density 0.0098. The comparison algorithm selects the mainstream baseline scheme in the field of LDPC sparse reconstruction: Gradual Row Transformation algorithm.

[0087] (I) Verification of the effect of coarse estimation of basic structural parameters

[0088] This phase verifies the ability of the joint coarse estimation method for basic structural parameters to screen expansion factors and the effect of generating the initial support matrix. For the dense parity-check matrix obtained by Gaussian elimination of the test code, the reconstruction method of this invention (hereinafter referred to as "this method") first performs positive divisor decomposition on the code length to obtain all candidate expansion factors. After sub-block partitioning, matching score calculation, zero / non-zero sub-block determination and comprehensive scoring, the optimal expansion factor of 27 can be accurately locked, corresponding to the generation of an initial support matrix of size 12×24.

[0089] Figure 5 This diagram illustrates the initial support matrix distribution output during the coarse estimation stage, with discrete points representing the locations of non-zero sub-blocks. It can be seen that this method can quickly locate potential non-zero sub-block distribution regions without fine-tuning, initially determining the size and approximate spatial location of the support matrix. This compresses the search space for subsequent optimization from the full matrix element level to the base matrix sub-block level, providing high-quality initial priors for subsequent sparsity optimization and significantly improving the computational efficiency of the overall reconstruction process.

[0090] (II) Verification of the global optimization effect of the support matrix

[0091] After obtaining the initial support matrix through rough estimation, a genetic algorithm adapted to the binary support structure is used to globally optimize the distribution of non-zero sub-blocks in the support matrix. Figure 6 This is a schematic diagram of the optimal support matrix distribution obtained after iterative optimization using a genetic algorithm. Figure 5 Compared to the initial support matrix, the optimized distribution of non-zero sub-blocks is sparser and more concentrated, effectively eliminating a large number of misjudged non-zero candidate points. The number of non-zero sub-blocks converges from the initial full distribution to 78, corresponding to a non-zero sub-block ratio of 27.08% in the support matrix. Experimental results show that the genetic algorithm can achieve global optimization in a complex non-convex binary combination search space, effectively selecting the optimal support matrix combination. It can accurately recover the number of non-zero sub-blocks and precisely locate their spatial positions, thus improving the reliability and stability of the support structure.

[0092] (III) Final Reconstruction Results of the Sparse Parity-Check Matrix

[0093] After global optimization of the support matrix and local fine-tuning of the shift values, the final QC-LDPC sparse check matrix is ​​generated by splicing sub-blocks. The reconstruction effect is verified from two dimensions: matrix structure and row redistribution.

[0094] Figure 7 This is a schematic diagram of the sparse parity-check matrix structure obtained by this method; the matrix size is 324×648. From... Figure 7 As can be clearly seen, the recovered matrix exhibits significant sparse structure characteristics. The distribution of non-zero elements closely matches the original coding structure, demonstrating a high degree of element accuracy. The matrix rank is fully recovered to 324, which is highly close to the density characteristics of the original code. The results show that this method can accurately recover the position distribution of the support matrix and its corresponding cyclic shift values, completely reconstructing the target parity check matrix that conforms to a quasi-cyclic structure.

[0095] Figure 8 The row weight distribution of the dense parity-check matrix obtained directly from Gaussian elimination has an average row weight of 308.06 and a standard deviation of 12.98. The row weight values ​​are high and the distribution is discrete, which does not conform to the structural characteristics of low row weight of LDPC codes and cannot be directly input into a standard LDPC decoder.

[0096] Figure 9 The row redistribution of the parity-check matrix after sparse reconstruction using this method reduces the average row weight to 6.50 and the standard deviation to 1.98. The low row weight and compact overall distribution fully meet the core characteristics of low row weight and uniform distribution of LDPC code parity-check matrices. It can be directly input into a standard LDPC decoder for decoding, effectively reducing the computational complexity of subsequent decoding and solving the engineering problem that dense parity-check matrices cannot be directly decoded.

[0097] (iv) Performance comparison with existing technologies

[0098] To further verify the performance advantages of this method, it was compared with existing asymptotic transformation algorithms under different matrix density scenarios. The reconstruction performance was evaluated from two core indicators: matrix reconstruction rate and reconstruction factor (the reconstruction factor is the ratio of the number of non-zero element positions matched between the reconstructed matrix and the original matrix to the total number of original non-zero elements).

[0099] Figure 10 The figures show a comparison of the reconstruction rates of the two methods under different matrix densities. It can be seen that the proposed method maintains a high reconstruction rate for both low-density and high-density matrices, generally remaining above 0.8, and its overall performance is superior to the progressive transform algorithm. The progressive transform algorithm, on the other hand, only maintains a high reconstruction rate in low-density matrix scenarios; as the matrix density increases, the reconstruction rate decreases significantly. Experimental results demonstrate that the proposed method has good adaptability to QC-LDPC codes of different densities and has a wider range of applications.

[0100] Figure 11 The figures show a comparison of the reconstruction factors of the two methods under different matrix densities. It can be seen that the reconstruction factor of the proposed method increases steadily in an approximately proportional manner with increasing matrix density, maintaining a high level across the entire density range. In contrast, the reconstruction factor of the asymptotic transformation algorithm decreases slightly with increasing density in the low-density region, but decreases significantly in the medium-to-high-density region. Experimental results demonstrate that the phased architecture of this invention, which first supports global optimization and then shifts to perform local optimization, exhibits a more prominent reconstruction advantage in medium-to-high-density matrix scenarios, effectively addressing the shortcomings of existing greedy algorithms that easily get trapped in local optima and experience a sharp drop in reconstruction performance in high-density scenarios.

[0101] (v) Verification Conclusion

[0102] In summary, the simulation results demonstrate that the joint coarse estimation method for QC-LDPC code basis structure parameters under non-cooperative communication conditions proposed in this invention can quickly lock the optimal expansion factor and initial support structure, significantly compressing the search space for subsequent optimization. The sparse parity-check matrix reconstruction method, through a phased hybrid architecture of "global optimization using genetic algorithm + local fine-tuning using simulated annealing," effectively balances global search capability and local optimization accuracy, significantly improving reconstruction accuracy and stability under different density scenarios. The reconstructed sparse parity-check matrix fully conforms to the quasi-cyclic structure characteristics of QC-LDPC codes and can be directly used for blind decoding processing under non-cooperative communication conditions, possessing significant engineering application value.

[0103] IV. QC-LDPC Code Sparse Parity-Check Matrix Reconstruction System

[0104] This system is designed to implement the functional modules of the two methods mentioned above. It can run on a communication signal processing hardware platform and includes a basic structure parameter coarse estimation module, a support matrix global optimization module, a shift value local optimization module, and a check matrix splicing and output module. Each module is connected in sequence to complete the complete reconstruction process from a dense matrix to a sparse matrix.

[0105] (a) Coarse estimation module for basic structural parameters

[0106] The basic structural parameter coarse estimation module is the system's preprocessing unit, internally integrating a candidate factor generation unit, a sub-block matching calculation unit, a support determination unit, a comprehensive scoring unit, and an optimal parameter output unit. The module receives the dense parity check matrix input from the front end, completes candidate expansion factor traversal, sub-block matching calculation, zero / non-zero determination, and comprehensive scoring, ultimately outputting the optimal expansion factor and initial support matrix, which are then passed to the subsequent optimization module.

[0107] (ii) Global Optimization Module for Support Matrix

[0108] The global support matrix optimization module receives the initial support matrix output by the coarse estimation module and integrates a population initialization unit, a genetic iteration unit, and an optimal support output unit. The module constructs a genetic algorithm population based on a semi-deterministic initialization strategy. Through multiple generations of selection, crossover, and mutation evolution, it completes the global optimization of the non-zero distribution of the support matrix and outputs the optimal support matrix to the shift value optimization module.

[0109] (III) Local Optimization Module for Shift Values

[0110] The shift value local optimization module receives the optimal support matrix and the initial shift value, and internally integrates an annealing parameter configuration unit, a neighborhood search unit, a Metropolis decision unit, and an optimal shift output unit. With the support structure fixed, the module performs a refined local search on the shift value using a simulated annealing algorithm, and outputs the optimal shift value matrix to the matrix concatenation module.

[0111] (iv) Verification matrix splicing and output module

[0112] The parity-check matrix concatenation output module receives the optimal expansion factor, optimal support matrix, and optimal shift matrix, and internally integrates sub-block construction units, matrix concatenation units, and decoding interface units. The module concatenates the matrix according to quasi-cyclic rules to generate a complete sparse parity-check matrix, and outputs it to an external LDPC decoding unit through a standard interface to support subsequent blind decoding processing.

[0113] V. Non-cooperative communication signal processing equipment

[0114] This device is the physical hardware unit that carries the above-mentioned reconstruction system. It is used in field scenarios such as non-cooperative communication reconnaissance and spectrum monitoring. Specifically, it includes a signal receiving and demodulation unit, a memory, a processor, and the above-mentioned QC-LDPC code sparse parity-check matrix reconstruction system.

[0115] (a) Signal receiving and demodulation unit

[0116] The signal receiving and demodulation unit is the physical signal input unit of the device, comprising a receiving antenna, RF front-end circuitry, a demodulator, and a hard decision module. This unit intercepts the RF coded signal in space through the antenna, sequentially performs down-conversion, demodulation, and hard decision processing to obtain the received bit sequence, and then solves the dual space using Gaussian elimination to generate a dense parity-check matrix, which is output to the reconstruction system. In this invention, when solving the dual space using Gaussian elimination, column pivoting is employed, and the column exchange mapping relationship is recorded simultaneously. Before sub-block partitioning, the original column order is restored according to the column mapping relationship to ensure the spatial consistency of the quasi-cyclic block structure.

[0117] (ii) Memory

[0118] The memory includes read-only memory (ROM) and random access memory (RAM); ROM is used to store the computer program of the reconstruction algorithm and preset parameter configurations (such as threshold coefficients, weight coefficients, number of algorithm iterations, etc.); RAM is used to cache intermediate matrix data, population individual data and temporary calculation results during signal processing.

[0119] (iii) Processor

[0120] The processor adopts an FPGA+ARM heterogeneous architecture, communicating with the signal receiving and demodulation unit, memory, and reconstruction system respectively. The ARM core is responsible for overall process scheduling and parameter configuration, while the FPGA core is used to execute computationally intensive operations such as sub-block matching, genetic iteration, and annealing search in parallel, scheduling the various modules to collaboratively complete the entire process of parameter estimation and matrix reconstruction.

[0121] (iv) Overall working process of the equipment

[0122] The signal receiving and demodulation unit intercepts and processes the radio frequency signal to generate a dense parity check matrix; the processor schedules and reconstructs the system to sequentially perform coarse estimation of the base structure, global optimization of the support matrix, local optimization of the shift value, and matrix splicing; the finally generated sparse parity check matrix is ​​transmitted to the decoding unit through the output interface to complete the blind decoding and information recovery of the non-cooperative signal.

[0123] VI. Typical Application Scenarios and Engineering Implementation

[0124] The joint coarse estimation method for basic structural parameters and the sparse parity-check matrix reconstruction method of this invention are both implemented on a physical non-cooperative communication signal processing device. The device integrates a signal receiving and demodulation unit, a heterogeneous processor, and a storage unit. It executes the complete processing flow through a built-in reconstruction system. It can be adapted to various product forms such as portable reconnaissance equipment, electronic countermeasure jammers, fixed spectrum monitoring stations, and protocol testing platforms according to different application scenarios. The specific engineering implementation is as follows: (I) Communication reconnaissance and intelligence analysis scenario: portable communication reconnaissance equipment In this scenario, the deployment device is a handheld / vehicle-mounted portable communication and reconnaissance equipment. The main body of the equipment includes a high-sensitivity receiving antenna, a multi-band radio frequency front-end, an FPGA+ARM heterogeneous processor, a display and control terminal, and a storage unit.

[0125] When the device is working, it first intercepts unknown radio frequency coded signals in space through the receiving antenna. After downconversion and demodulation by the radio frequency front-end and demodulation by the demodulator, the received bit sequence is generated by the hard decision module. Then, the dense parity check matrix is ​​obtained by the Gaussian elimination operation scheduled by the ARM core. Subsequently, the built-in reconstruction system is called. First, the basic structure parameter coarse estimation module traverses the candidate expansion factors. The optimal expansion factor and the initial support matrix are quickly locked by the density-matching degree comprehensive score, which compresses the search space for subsequent optimization by more than 90%. Then, the FPGA core executes the global optimization of the genetic algorithm and the local optimization of simulated annealing in parallel to complete the high-precision reconstruction of the sparse parity check matrix.

[0126] The average row weight of the reconstructed sparse parity-check matrix can be reduced from over 300 to around 6.5, and the matrix rank is fully recovered. It can be directly input into the device's built-in LDPC decoder for blind decoding, and finally the recovered plaintext information is output to the display and control terminal for intelligence analysts to analyze. Compared with traditional row-by-row sparsification schemes, this technology can reduce the reconstruction time of medium-to-long code length signals by more than 40%, and still maintain a structural matching degree of about 98% for high-density coded signals, adapting to the diverse coding systems and real-time requirements of field reconnaissance scenarios.

[0127] (II) Electronic Countermeasures and Jamming Suppression Scenarios: Vehicle-mounted Electronic Countermeasures and Jamming Equipment

[0128] In this scenario, the deployment platform is a vehicle-mounted / airborne electronic countermeasures jamming device, which includes a broadband signal receiving channel, a jamming signal transmitting channel, a baseband processing board, and a power amplification unit.

[0129] In response to sudden non-cooperative communication signals, the receiving channel of the device first intercepts and down-converts the signal. The ARM core of the baseband processing board quickly calls the joint coarse estimation method of the basic structure parameters of this invention. The expansion factor screening and initial judgment of the coding system can be completed in milliseconds through sub-block matching and comprehensive scoring, thus seizing the time window for electronic countermeasure response. If it is necessary to generate a precise jamming signal, the FPGA core is further scheduled to execute a complete reconstruction process to restore the complete sparse parity check matrix and shift parameters. Based on the reconstructed coding structure, a matching coherent jamming waveform is generated, which is then amplified and radiated through the transmitting antenna to improve jamming effectiveness.

[0130] If used in anti-jamming reception scenarios, the reconstructed sparse parity-check matrix is ​​input into the iterative decoding module to optimize the decoding iteration strategy, improve the signal reception sensitivity and demodulation accuracy under strong interference environments, and adapt to the two-way combat requirements in complex electromagnetic countermeasures environments.

[0131] (III) Spectrum Monitoring and Cognitive Radio Scenarios: Fixed-Station Spectrum Monitoring Equipment

[0132] In this scenario, the deployment devices are fixed-site spectrum monitoring equipment and cognitive radio base stations. The equipment includes a wideband monitoring antenna array, a multi-channel signal acquisition card, a server-level processor, and a spectrum management database.

[0133] During operation, the device performs batch processing on the massive amount of unknown signals intercepted across the entire frequency band: for the dense parity check matrix generated for each signal, the joint coarse estimation method of this invention is called to complete the identification of QC-LDPC encoded signals and the extraction of core parameters, and outputs feature parameters such as spread factor and support density into the spectrum management database for statistical analysis of the coding system distribution within the frequency band and fine identification of spectrum occupancy status.

[0134] If dynamic spectrum access and cooperative decoding are required, a complete reconstruction process is performed on the target signal to generate a standard sparse parity check matrix and synchronize it to the decoding unit of the cognitive base station, so as to realize the cooperative forwarding and spectrum sharing of unknown coded signals, and provide underlying technical support for dynamic spectrum resource scheduling and refined management of civilian spectrum.

[0135] (iv) Communication Protocol Reverse Analysis Scenario: Dedicated Communication Protocol Reverse Testing Platform

[0136] In this scenario, the implementation platform is an industrial / IoT communication protocol reverse testing platform, which includes a high-speed signal acquisition card, a host computer processor, protocol reverse analysis software, and testing interfaces.

[0137] For industrial control and IoT-specific communication systems without pre-defined protocol documents, the test platform intercepts physical layer communication bit data through a high-speed acquisition card. After preprocessing to generate a dense parity check matrix, the complete reconstruction method of this invention is called to restore all core coding structures of the QC-LDPC code: including complete parameters such as the base matrix size, spatial distribution of non-zero sub-blocks, and cyclic shift values ​​of each sub-block. The reconstruction result has a structural matching degree of over 98% with the original code.

[0138] The recovered coding structure parameters can be directly imported into protocol reverse analysis software, which can assist technicians in restoring the channel coding architecture, parsing frame format rules, conducting compatibility tests and security vulnerability analysis, and providing reliable experimental basis for technical verification and risk assessment of dedicated communication systems.

[0139] Overall, this invention addresses the engineering pain points of large computational load, easy getting trapped in local optima, and sharp performance drop in high-density scenarios for QC-LDPC code reconstruction in non-cooperative scenarios by adopting a technical approach of pre-coarse estimation dimensionality reduction and phased hybrid optimization. It can be directly adapted to various communication signal processing hardware devices and meet the diverse blind recognition engineering needs in complex electromagnetic environments.

[0140] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A joint coarse estimation method for QC-LDPC code base structure parameters under non-cooperative communication, applied to channel coding blind identification scenarios, performs parameter estimation based on the dense parity-check matrix obtained by Gaussian elimination of the intercepted signal, characterized in that... Includes the following steps: Based on the code length of the dense parity check matrix, generate a set of all candidate expansion factors, where each candidate expansion factor is a positive divisor of the code length; Iterate through each candidate expansion factor and divide the dense parity check matrix into sub-block arrays of corresponding sizes according to the current expansion factor. For each sub-block, adopt a circular diagonal mapping method and take the difference between the column index and row index in the block modulo the current expansion factor. Iterate through all elements in the sub-block in one go and accumulate them according to the shift value to directly obtain the matching score of the sub-block under all shift values. Extract the maximum matching score of each sub-block. Based on the preset non-zero sub-block determination threshold, the zero / non-zero attributes of each sub-block are determined according to the maximum matching score of each sub-block, and the initial support matrix corresponding to the current candidate expansion factor is generated. A comprehensive scoring function is used to score the current candidate expansion factor and its corresponding initial support matrix; The candidate expansion factor with the highest comprehensive score is selected as the optimal expansion factor, and its initial support matrix is ​​output accordingly. This matrix serves as the structural prior for subsequent sparse reconstruction, supporting blind decoding processing under non-cooperative communication.

2. The joint coarse estimation method for QC-LDPC code base structure parameters under non-cooperative communication according to claim 1, characterized in that, The comprehensive scoring function is expressed as follows: ； ； In the formula, score is the overall score; α is the piecewise weight coefficient; exp(·) is the natural exponential function; ρ is the support matrix density, i.e., the proportion of non-zero elements in the support matrix under the current candidate expansion factor; d est The LDPC code matrix density is estimated empirically; σ is the standard deviation of the Gaussian function, used to control the smoothness of score decay; S i,j The element in the i-th row and j-th column of the support matrix takes the value 0 or 1, where 1 represents that the element is in a non-zero sub-block and 0 represents a zero sub-block. match i,j (s) is the sum of the corresponding elements of the (i,j)th sub-block and the standard cyclic permutation matrix with shift value s; The (i,j)th sub-block represents the highest matching score achievable with the cyclic permutation matrix among all possible shift values ​​s; support is the set of sub-blocks in the support matrix that are determined to be non-zero; |·| is the absolute value operation.

3. The joint coarse estimation method for QC-LDPC code base structure parameters under non-cooperative communication as described in claim 1, characterized in that, The matching score ranges from 0 to the value of the current candidate expansion factor. The higher the score, the higher the structural overlap between the sub-block and the cyclic permutation matrix. When the maximum matching score of the sub-block is less than 0.5 times the current candidate expansion factor, the sub-block is directly determined to be a zero sub-block and is not included in the subsequent zero / non-zero sub-block determination process. Conversely, when the maximum matching score of a sub-block is greater than or equal to 0.5 times the current candidate expansion factor, the sub-block enters the subsequent judgment stage, based on the non-zero sub-block judgment threshold coefficient. Further determine its zero / non-zero properties.

4. The joint coarse estimation method for QC-LDPC code base structure parameters under non-cooperative communication as described in claim 1, characterized in that, When calculating the matching score of each sub-block, the optimal shift value corresponding to the maximum matching score of each sub-block is recorded simultaneously. While outputting the initial support matrix, the optimal shift values ​​corresponding to all non-zero sub-blocks are also output simultaneously and organized into an initial shift vector according to row and column indices. This initial shift vector serves as the structured hot start point for local fine optimization of shift values ​​in subsequent sparse reconstruction, replacing the random initialization method.

5. A method for reconstructing a sparse parity-check matrix of a QC-LDPC code under non-cooperative communication, characterized in that, For the coded signal intercepted in a non-cooperative communication scenario, after demodulation and Gaussian elimination to obtain a dense parity-check matrix, the following steps are performed: By performing joint coarse estimation of the base structure parameters of the QC-LDPC code under non-cooperative communication according to any one of claims 1-4, the optimal spread factor and the corresponding initial support matrix are obtained. A semi-deterministic initial population is constructed based on the initial support matrix. A genetic algorithm adapted to the binary support structure is used to globally optimize the distribution of non-zero sub-blocks in the support matrix and output the optimal support matrix. The positions of the non-zero sub-blocks in the fixed optimal support matrix are used, and the cyclic shift values ​​of each non-zero sub-block are locally fine-tuned using the simulated annealing algorithm with adaptive integer shift fine-tuning, and the optimal shift value matrix is ​​output. Based on the optimal expansion factor, optimal support matrix, and optimal shift value matrix, a sparse parity matrix conforming to a quasi-cyclic structure is generated by splicing sub-blocks. The sparse parity-check matrix is ​​directly input into the LDPC decoder for blind decoding and information recovery of non-cooperative communication signals.

6. The method for reconstructing a sparse parity-check matrix of a QC-LDPC code under non-cooperative communication as described in claim 5, characterized in that, Global optimization includes the following steps: Fitness is calculated for each candidate support matrix individual in the population. Fitness is used to measure the degree of fit between the candidate support matrix and the structural characteristics of the QC-LDPC code. An elite retention strategy is adopted, in which the two individuals with the highest fitness in the current population are directly retained to the next generation. A tournament selection mechanism was used to select parent individuals, and a new generation population was generated through row and column crossover and site mutation operations. Repeated iterative evolution continues until no more individuals with better fitness can be generated or the preset maximum number of generations is reached. The individual with the highest fitness in the last generation population is selected as the optimal support matrix.

7. The method for reconstructing a sparse parity-check matrix of a QC-LDPC code under non-cooperative communication as described in claim 5, characterized in that, Local fine-tuning specifically includes the following steps: Set the initial temperature, annealing rate and maximum number of iterations for simulated annealing, and use the initial shift vector as the starting point for optimization. At the current temperature, a non-zero sub-block is randomly selected and its shift value is adjusted to generate a candidate shift vector. The energy change between the candidate shift vector and the current shift vector is calculated. After completing a preset number of iterations at the current temperature, a cooling operation is performed. The iterations are repeated until the shift value can no longer be optimized or the maximum number of iterations is reached, and the optimal shift value matrix is ​​output.

8. The method for reconstructing a sparse parity-check matrix of a QC-LDPC code under non-cooperative communication as described in claim 5, characterized in that, After calculating the energy change, the Metropolis criterion is followed to determine whether to accept the candidate shift vector: If the energy value corresponding to the candidate shift vector is less than the energy value of the current shift vector, then the candidate shift vector is accepted directly. If the energy value corresponding to the candidate shift vector is greater than or equal to the energy value of the current shift vector, then the candidate shift vector is accepted with probability exp(-Δf / T), where Δf is the energy change and T is the current temperature.

9. A QC-LDPC code sparse parity-check matrix reconstruction system under non-cooperative communication, characterized in that, The method for performing the joint coarse estimation of QC-LDPC code basis structure parameters under non-cooperative communication as described in any one of claims 1-4, and the method for reconstructing the sparse parity-check matrix of QC-LDPC codes under non-cooperative communication as described in any one of claims 5-8, comprises: The basic structure parameter coarse estimation module is used to receive the dense parity check matrix obtained by Gaussian elimination, generate a set of all candidate expansion factors based on the code length of the dense parity check matrix, traverse each candidate expansion factor to complete sub-block partitioning, matching score calculation, zero / non-zero sub-block determination and comprehensive scoring, and output the optimal expansion factor and the corresponding initial support matrix. The global optimization module for the support matrix is ​​used to construct a semi-deterministic initial population based on the initial support matrix. It employs a genetic algorithm adapted to the binary support structure to globally optimize the distribution of non-zero sub-blocks in the support matrix and outputs the optimal support matrix. The shift value local optimization module is used to fix the positions of non-zero sub-blocks in the optimal support matrix. It uses a simulated annealing algorithm with adaptive integer shift fine-tuning to perform local fine optimization of the cyclic shift values ​​of each non-zero sub-block and outputs the optimal shift value matrix. The parity check matrix splicing output module is used to generate a sparse parity check matrix that conforms to a quasi-cyclic structure by splicing sub-blocks according to the optimal expansion factor, optimal support matrix and optimal shift value matrix, and output it to the LDPC decoding unit to support blind decoding processing of non-cooperative communication signals.

10. A non-cooperative communication signal processing device, characterized in that, The system for reconstructing a QC-LDPC code sparse parity-check matrix under non-cooperative communication as described in claim 9 further includes a signal receiving and demodulation unit, a memory, and a processor. The signal receiving and demodulation unit is used to intercept radio frequency coded signals in non-cooperative communication scenarios. The signals are then demodulated, hard-decision, and Gaussian elimination dual space solution is performed to obtain a dense parity check matrix, which is then transmitted to the reconstruction system. Memory is used to store intermediate data during computer programs and signal processing; The processor loads and runs the QC-LDPC code sparse parity-check matrix reconstruction system, which is connected to the signal receiving and demodulation unit and the memory, respectively, and is used to schedule each module to execute the corresponding parameter estimation and matrix reconstruction process.

Images