Hierarchical LDPC decoding method based on syndrome feedback self-adaption

Abstract

The invention discloses a layered LDPC (Low Density Parity Check Code) decoding method based on syndrome feedback self-adaption, and aims to solve the problem of poor decoding performance caused by fixed correction factors and poor adaptability to different check matrixes and channel conditions in the existing LDPC decoding algorithm. The method comprises the following steps of: firstly, dividing rows of a check matrix into a plurality of independent layers by adopting a hierarchical scheduling architecture so as to accelerate propagation and convergence of decoding information; secondly, in each iteration process of hierarchical decoding, a self-adaptive mechanism based on syndrome feedback is used, and a core offset factor beta in an offset minimum sum algorithm is adjusted in real time; the mechanism intelligently increases or decreases the offset factor by diagnosing the current decoding state (i.e., the number of unsatisfied check equations) to help the decoding process jump out of local optimum and stably converge. A simulation result shows that compared with a traditional offset minimum sum algorithm, the decoding performance is effectively improved and the bit error rate is reduced on the premise that the complexity is not obviously increased, and particularly, the performance gain is obvious in a medium-high signal-to-noise ratio region, and the robustness is higher.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital communication, in particular to error correction code decoding technology in channel coding, and especially to a high-performance and high-adaptability decoding method for low-density parity-check (LDPC) codes. BACKGROUND

[0002] Low-density parity-check (LDPC) codes are linear block codes with performance approaching the Shannon limit, which have been widely applied in 5G communication, Wi-Fi, satellite communication and other standards. The decoding algorithm of LDPC codes is mainly based on message propagation on the Tanner graph, and the Sum-Product algorithm has the best performance but extremely high computational complexity. In order to facilitate hardware implementation, researchers have proposed a variety of simplified algorithms, such as the Min-Sum (MS) algorithm.

[0003] The MS algorithm significantly reduces the complexity by approximate calculation, but brings performance loss. To make up for this loss, the normalized Min-Sum (NMS) algorithm and the offset Min-Sum (OMS) algorithm are derived. The OMS algorithm corrects the overestimation problem of the MS algorithm by subtracting a fixed offset factor β from the minimum value of the check node update message. However, the β value of the existing OMS algorithm and some improved algorithms (such as DAOMS) is usually a fixed parameter or follows an open-loop rule. This fixed or open-loop setting has significant defects: when the check matrix H of the LDPC code or the channel condition changes, the original optimal β value is no longer applicable, resulting in algorithm performance degradation or instability, and lack of robustness. In addition, the traditional flooding scheduling strategy updates information slowly, which limits the convergence speed, and the existing hierarchical scheduling strategy is usually combined with a fixed parameter decoder, which fails to solve the problem of parameter adaptation. Therefore, how to design a decoding algorithm that can automatically adapt to the current decoding environment, has fast convergence speed and superior performance is a technical problem to be solved in the field. SUMMARY

[0004] The purpose of the present application is to overcome the shortcomings of the prior art and provide a hierarchical LDPC decoding method based on check sub-feedback adaptation, which aims to solve the problems of fixed correction factor, poor adaptability and slow convergence speed of the traditional OMS algorithm.

[0005] The core idea of the present application is to deeply integrate an efficient hierarchical scheduling architecture with a novel closed-loop feedback adaptive offset mechanism to produce synergistic gains.

[0006] Firstly, the present application adopts a hierarchical scheduling architecture, which updates information layer by layer in iterations by grouping the rows of the check matrix. This architecture enables information to be quickly propagated within a single iteration, thereby accelerating the overall convergence speed of the algorithm and reducing the average number of iterations required to achieve correct decoding.

[0007] Secondly, and also the key innovation point of the present application, the present application abandons the traditional practice of using fixed offset factors. Instead, an adaptive adjustment mechanism based on internal state feedback of the decoder is designed. After each complete iteration, the progress of decoding is accurately diagnosed by calculating the syndrome weight (i.e. the number of unsatisfied check equations). If the decoding progresses smoothly, the offset factor β is appropriately reduced for stability; if the decoding is stuck, the offset factor β is increased to provide "perturbation" to help it jump out of the local optimal solution.

[0008] In the present application, layered scheduling and adaptive offset form an organic combination: layered scheduling accelerates the propagation of correct (or incorrect) information, while the adaptive offset mechanism ensures that the propagated information is within the correct range. The combination of the two avoids the rapid performance degradation problem that layered scheduling may cause under incorrect parameters, and achieves the unification of speed and accuracy.

[0009] The present application has the beneficial effects of:

[0010] 1. Performance improvement: By optimizing the offset factor in real time, the present method can find more effective decoding paths under various conditions, thereby achieving a lower bit error rate than the traditional OMS algorithm using fixed factors.

[0011] 2. Convergence acceleration: Combined with the layered scheduling architecture, the information propagation efficiency is higher, and the average number of iterations required to reach the final decoding result is less.

[0012] 3. Strong robustness: The algorithm no longer relies on pre-computed optimal parameters for specific H matrices, greatly enhancing its versatility. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 Basic block diagram of the communication system to which the present application is applied.

[0014] Figure 2 Overall flowchart of the L-SA-OMS decoding method proposed by the present application, showing the complete logical closed loop from initialization to iteration, layered scheduling, diagnosis, adaptive adjustment, and termination judgment.

[0015] Figure 3 Schematic diagram of the check node feedback adaptive control logic, which details how to select different paths to increase or decrease the offset factor based on the comparison results of the current and historical check node weights. DETAILED DESCRIPTION

[0016] To make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. The logical flow of the method of the present application can be understood as a system comprising a fast execution path and a policy control path. The fast execution path is responsible for efficient message computation and propagation, while the policy control path is responsible for intelligently regulating the behavior of the execution path.

[0017] I. Initialization phase

[0018] Before the decoding starts, the system performs preparations. The externally input channel Log-Likelihood Ratio (LLR) sequence is loaded as the initial a posteriori probability information of each variable node, denoted as L_v_app. Meanwhile, the matrix Uji used to store the Check-to-Variable (C2V) messages is initialized to all zeros. According to the given check matrix H, an efficient layering scheme layer_scheme is pre-computed by analyzing the column index conflicts among the rows of its base matrix. Finally, the initial value of the offset factor β (e.g. 0.5) and its dynamic adjustment control parameters, such as maximum / minimum value, increment / decrement factor, etc., are set.

[0019] II. Iteration and layer-wise update phase

[0020] This phase is the core of message passing. Referring to Figure 2 , the system enters an outer iteration loop, the number of which is limited by iterMax. In each outer iteration, the system sequentially traverses all the layers in layer_scheme and performs an inner layer-wise loop.

[0021] For each layer in the inner loop, the check nodes contained therein perform a complete message update. Specifically, for any variable node i connected to a check node j, the information update of i follows the following precise steps:

[0022] 1. Extrinsic information computation: First, the variable node i needs to compute its extrinsic information (Extrinsic Information) passed to the check node j, which represents the confidence provided by all other sources to i except j. The computation is L_v->c(i,j) = L_v_app(i) - Uji(j,i), i.e. subtracting the old information contributed by j to i in the last iteration from the current a posteriori probability of i.

[0023] 2. C2V message update: Check node j collects all incoming extrinsic information L_v->c from all variable nodes connected to it, and computes new C2V message for each outgoing direction (e.g. to i). The computation follows the offset min-sum rule: U_new(j,i) = (Π sign) * max(min(|L_v->c|) - β, 0). Where β is the current global, adaptively adjusted offset factor.

[0024] 3. A posteriori probability instant update: This is the essence of hierarchical scheduling. The newly computed C2V message U_new(j,i) is immediately used to update the a posteriori probability of variable node i: L_v_app(i)_new = L_v_app(i)_old - Uji_old(j,i) + U_new(j,i). This updated L_v_app(i)_new can be immediately used in the computation of the next layer in the current iteration, thus greatly accelerating the convergence of correct information.

[0025] III. Diagnosis and adaptive feedback phase

[0026] When the inner hierarchical loop is finished, it represents the end of a complete outer iteration. At this time, the system enters the strategy control path, referring to the logic of Figure 3 :

[0027] 1. Hard decision and diagnosis: The system makes a hard decision (LLR < 0 is 1, otherwise 0) according to the final a posteriori probability L_v_app of all variable nodes, forming the decoding codeword x of the current iteration. Then the syndrome weight W_s = sum(mod(H*x', 2)) is calculated, which accurately reflects the error degree of the current decoding result.

[0028] 2. Termination judgment: Check whether W_s is zero. If it is zero, the decoding is successful, and the flow ends. If it is not zero, check whether the iteration number reaches the upper limit. If it reaches, the flow also ends.

[0029] 3. Adaptive feedback: If it is not terminated, it enters the core adaptive adjustment link. Compare the current W_s and the W_s_prev of the last iteration. If W_s < W_s_prev, it means that the decoding has made effective progress, and the system judges that the current β value strategy is correct. In order to be stable, it is attenuated: β = β * decrease_factor. If W_s >= W_s_prev, it means that the decoding stagnates or worsens, and the system judges that the current β value may be too small or not suitable, so it is enhanced: β = β * increase_factor.

[0030] The adjusted β will be used in the next complete outer iteration.

[0031] Through the precise cooperation of the above initialization, layered updating and adaptive feedback, the method can realize efficient, high-precision and high-robustness decoding of the LDPC code.

Claims

1. A hierarchical LDPC decoding method based on checksum feedback adaptive decoding, characterized in that, Includes the following steps: Step 1, Initialization Phase: Obtain the channel log-likelihood ratio (LLR) as the initial variable node posterior information; initialize the check node to variable node (C2V) message matrix to all zeros; set the maximum number of decoding iterations; Based on the LDPC parity check matrix H to be decoded, determine its layering scheme; Set an initial offset factor β. Step 2, Iterative Decoding Phase: Within the maximum number of iterations, execute the outer iterative loop; Step 3, Layered Update Phase: In each outer layer iteration, according to the layered scheme, the inner layer layered loop is executed to process each layer in turn; when processing the current layer, according to the posterior information of the variable node connected to the verification node of this layer, combined with the offset factor β, the message from the verification node to the variable node is updated, and the posterior information of the variable node is updated in real time. Step 4, State Diagnosis Phase: After completing one full outer layer iteration (i.e., all layers have been updated), hard decision is made on the posterior information of all variable nodes to obtain the current decoded codeword, and the checksum obtained by multiplying the codeword with the check matrix H is calculated. The number of non-zero elements in the checksum is counted, which is the current checksum weight. Step 5, Adaptive Adjustment Phase: Compare the current parser weight with the parser weight of the previous iteration; if the current parser weight is less than the previous one, it is determined that the decoding has made progress, and the value of the offset factor β is reduced according to the preset decay rule; if the current parser weight is greater than or equal to the previous one, it is determined that the decoding has stalled, and the value of the offset factor β is increased according to the preset growth rule. Step 6, Termination Judgment Phase: Determine if the current checker weight is zero. If it is zero, the decoding is successful, the iteration ends and the decoded codeword is output. Otherwise, determine whether the maximum number of iterations has been reached. If it has, end the iteration and output the current decoded codeword; otherwise, return to step two to continue the next outer iteration.

2. The method according to claim 1, characterized in that, The calculation of the update check node to variable node message in step three follows the offset min-sum rule, that is, the sign product of all incoming messages except the target variable node is multiplied by the difference between the minimum absolute value of these messages and the offset factor β.

3. The method according to claim 1, characterized in that, The attenuation rule in step five is to multiply the current offset factor β by an attenuation factor less than 1; the growth rule is to multiply the current offset factor β by an growth factor greater than 1; and the value of the offset factor β is restricted to a preset maximum and minimum value.

4. The method according to claim 1, characterized in that, The hierarchical scheme described in step one is obtained by analyzing the base matrix of the parity matrix H. The partitioning principle is that within the same layer, the column indices of the non-zero elements contained in any two rows are different.

5. The method according to any one of claims 1-4, characterized in that, The LDPC code is a quasi-cyclic LDPC (QC-LDPC) code, and this method can be applied to digital communication systems such as 5G / 6G mobile communication, satellite communication, or optical communication.

Images