A decoding scheduling method for 5G LDPC code

Abstract

The present application belongs to the technical field of channel coding and decoding, and relates to a decoding scheduling method for 5G LDPC code. The LPSRB scheduling scheme designed by the present application is based on static scheduling LPHD and dynamic scheduling RB-LBP. Since the complexity of RB-LBP is high, the advantages of static scheduling and dynamic scheduling are combined. The present application thus proposes the LPSRB scheduling scheme, the main idea of which is to firstly use the number of non-zero element values in the two columns before puncturing in the NR LDPC base graph to make preliminary sorting, and secondly use the improvement of RB-LBP to make sorting again for the rows with the same priority, and combine static scheduling and static scheduling to obtain the final scheduling sequence, so as to reduce the complexity. When sorting the rows with the same priority, the LPSRB scheduling scheme proposed by the present application is different from RB-LBP which uses the maximum value of each row residual to sort, but sums each row residual and sorts according to the sum. The simulation results show that the average iteration number and decoding bit error rate performance of the LPSRB proposed by the present application are better than those of LPHD and RB-LBP.

Description

Technical Field

[0001] This invention belongs to the field of channel coding and decoding technology, and relates to a decoding and scheduling method for 5G LDPC codes. Background Technology

[0002] Low-density parity-check codes, also known as LDPC codes, have been selected by 3GPP as the channel coding scheme for data channels in 5G NR enhanced mobile broadband (eMBB). LDPC codes are widely used due to their performance approaching the Shannon limit, and many communication standards adopt them as channel coding. The choice of different decoding algorithms significantly impacts their performance. A popular decoding algorithm is the belief propagation (BP) algorithm, a flooding scheduling method that updates all parity check nodes in a single iteration. Flooding scheduling has strong parallelism, but the latest information obtained in the current iteration cannot be used until the next iteration, leading to slower convergence. To address this issue, sequential scheduling, which allows for the rapid use of the latest information, has been proposed. A typical sequential scheduling method is hierarchical BP (LBP), which allows lower layers to use the latest information obtained from upper layers. With appropriate scheduling schemes, both convergence speed and error rate performance can be improved simultaneously. Scheduling schemes are mainly divided into two categories: static scheduling and dynamic scheduling.

[0003] In static scheduling, there are schemes such as LPHD (least-punctured and highest-degree) proposed by Bingbing Wang et al. and Novel BGSS (Novel base graph based static scheduling) proposed by Kuangda Tian et al. Both of these static scheduling schemes take into account the unique properties of NR LDPC, and are based on the number of non-zero punctured nodes in the base matrix and the node degree sorting. LPHD sorting is based on two rules: ascending order of non-zero punctured nodes and descending order of check node degree. The Novel BGSS scheme is based on three rules: ascending order of non-zero punctured nodes, ascending order of check node degree, and descending order of the sum of update counts of variable nodes adjacent to the check node.

[0004] In terms of dynamic scheduling, there are several schemes, including the RB-LBP (residual-based LBP) proposed by Bingbing Wang et al., the RBP (residual belief propagation) scheme proposed by G. Elidan et al., and the NW-RBP (node-wise RBP) scheme proposed by AIV Casado et al. RBP and NW-RBP have faster convergence speeds, but both require significant computation. Compared to these two schemes, the RB-LBP scheme achieves an excellent trade-off between additional complexity and performance. All three schemes are based on residuals, i.e., the difference between the information generated in the current iteration and the information generated in the previous iteration. The principle is that the larger this difference is, the further that part of the node information is from convergence. Summary of the Invention

[0005] This invention designs a scheduling method that combines static and dynamic scheduling to improve both convergence speed and error rate performance while reducing additional complexity. Based on the LPHD and RB-LBP scheduling schemes, the LPSRB designed in this invention utilizes the unique property of 5G LDPC—a non-zero number of punctures—and further determines the complete sequence scheduling by incorporating residual-based ideas.

[0006] The technical solution of the present invention is as follows: Figure 1 As shown, an LBP decoding and scheduling method for 5G LDPC codes is defined, comprising the following steps:

[0007] S1. Based on the number of non-zero shift values ​​in the first two columns of the 5G LDPC base map, sort them according to the following row priority order: rows with 0 non-zero shift values, rows with 1 non-zero shift value, and rows with 2 non-zero shift values; for ease of description, this is defined as the LP rule.

[0008] S2. Determine whether there are multiple rows with the same priority in the priority order obtained by S1. If yes, proceed to S3; otherwise, proceed to S4.

[0009] S3. Calculate the residual information passed from each row's check node to the variable node, and sort the information residuals in each row according to their priority to obtain the final priority ranking; this is defined as the SRB rule.

[0010] S4. Decoding and scheduling are performed according to the obtained priority order.

[0011] The key to obtaining a complete sequential schedule lies in sorting the two or three priorities obtained after sorting by the LP rules, and then further sorting the rows with the same priority (same number of non-zero punches) using the SRB rules. The specific method is as follows:

[0012] Assume that after the initial sorting by the LP rule, we obtain three priority row indices for the BG matrix, corresponding to 0, 1, and 2 non-zero elements punched in the BG matrix, respectively. Also, assume that there are k rows in the BG matrix with one non-zero element punched. Then there are k*z rows (where z is the expansion factor) with the same priority, indices {P1, P2, ..., P...}. k*z}

[0013] Calculate the residual information passed from the check nodes to the variable nodes in rows k*z. in Obtained from the following formula:

[0014]

[0015] in For verifying node c i Passed to variable node v j Information, i = P1, P2, ..., P k*z . Pre-computation for the next update is only needed when determining the order of the layers and is never propagated. Therefore, pre-computation does not need to be very precise, and is calculated using the following formula:

[0016]

[0017] in It is the variable node v b Passed to the verification node c i Information, where N(c i ) / v j To verify node c i Remove variable node v from adjacent variable nodes j A set of.

[0018] Information residuals in each row Summing, we get Where N(c) i ) for verification node c i The set of adjacent variable nodes is used to sort the priority order Q1 of the k*z rows.

[0019] Using the same method, Q0, Q1, and Q2 can be obtained separately. Combining Q0, Q1, and Q2 yields a complete sequential schedule.

[0020] The purpose of the above scheme is to reduce the complexity of dynamic scheduling by using static scheduling rules, while taking advantage of the faster convergence speed and better error rate performance of residual-based schemes, and at the same time, to take advantage of the unique properties of NRLDPC in dynamic scheduling.

[0021] The beneficial effect of this invention is that the LPSRB scheme utilizes the residual-based idea of ​​RB-LBP, but unlike RB-LBP which sorts by the maximum residual value of each row, it sorts by the sum of the residuals of each row. This allows for a comprehensive consideration of the residuals of each node in each row, i.e., the degree to which the information of each node is close to convergence, ensuring that each node in each updated row can contribute significantly to the current iteration. This enables the LPSRB scheme to reduce complexity while improving convergence speed and decoding performance compared to LPHD and RB-LBP. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the LPSRB scheduling scheme of the present invention;

[0023] Figure 2 Simulation diagrams comparing the bit error rates of traditional LBP decoding, RB-LBP scheduling scheme decoding, LPHD scheduling scheme decoding, and the proposed LPSRB scheduling scheme decoding.

[0024] Figure 3 Simulation results show the average number of iterations (ANI) required for traditional LBP decoding, RB-LBP scheduling scheme decoding, LPHD scheduling scheme, and the proposed LPSRB scheduling scheme at the same signal-to-noise ratio. Detailed Implementation

[0025] The technical solution of the present invention has been described in detail in the Summary of the Invention section. The effects of the technical solution of the present invention will be described below with reference to the accompanying drawings:

[0026] (1) Comparison of bit error rates between traditional LBP decoding, RB-LBP scheduling scheme decoding, LPHD scheduling scheme decoding, and LPSRB scheduling scheme:

[0027] The bit error rate (BER) results for a 5G LDPC code with a TBS size of 248, a code length of 564, a code rate of 22 / 47, a base matrix of BG1, and a shift factor z of 12, after 30 iterations of traditional LBP decoding, RB-LBP scheduling scheme decoding, LPHD scheduling scheme decoding, and LPSRB scheduling scheme decoding are as follows: Figure 2 As shown in the figure, the LPSRB scheduling scheme has the best decoding error rate performance, achieving a gain of approximately 0.05 dB compared to RB-LBP at 3 dB.

[0028] (2) Comparison of average iteration counts for traditional LBP decoding, RB-LBP scheduling scheme decoding, LPHD scheduling scheme decoding, and LPSRB scheduling scheme:

[0029] The TBS is a 5G LDPC code with a size of 248, a code length of 564, a code rate of 22 / 47, a base matrix of BG1, and a shift factor z of 12. It employs traditional LBP decoding, RB-LBP scheduling scheme decoding, LPHD scheduling scheme decoding, and LPSRB scheduling scheme decoding, with the termination condition being Hc. T =0, the ANI result obtained with a maximum number of iterations of 30 is as follows Figure 3 As shown in the figure. The comparison shows that the proposed LPSRB scheme requires less ANI than the RB-LBP scheme in the 1-2.5dB range.

[0030] (3) Analysis of the one-time iteration complexity of traditional LBP decoding, RB-LBP scheduling scheme decoding, LPHD scheduling scheme decoding, and LPSRB scheduling scheme:

[0031] Let E be the number of 1s in the parity check matrix H, and M be the number of parity check nodes. Since the proportions of rows containing 0, 1, and 2 non-zero elements with punctures differ between the two different basis matrices BG1 and BG2, the decoding complexity under each basis matrix is ​​analyzed separately. The results are shown in Tables 1 and 2.

[0032] Table 1 shows the number of computations required for each step of one iteration when the basis matrix is ​​BG1.

[0033]

[0034] Table 2 shows the number of calculations required for each step of one iteration when the base matrix is ​​BG2.

[0035]

Claims

1. A decoding and scheduling method for 5G LDPC codes, characterized in that, Includes the following steps: S1. Based on the number of non-zero shift values ​​in the first two columns of the NR LDPC base map, sort them according to the following row priority order: rows with 0 non-zero shift values, rows with 1 non-zero shift value, and rows with 2 non-zero shift values. S2. Determine whether there are multiple rows with the same priority in the priority order obtained by S1. If yes, proceed to S3; otherwise, proceed to S4. S3. Calculate the residual information passed from each row's validation node to the variable node, and sort the rows according to their priority based on the residual information in each row to obtain the final priority ranking. Specifically, define k rows with multiple rows of equal priority; then there are k*z rows with the same priority. z is the expansion factor; Calculate the residual information passed from the check nodes to the variable nodes in rows k*z. : ， in For verification nodes Passed to variable node Information, , Pre-calculation for the next update: ， in It is a variable node Passed to the verification node Information, To verify the node Remove variable node from adjacent variable node A set; Information residuals in each row Summing, we get ,in To verify the node The priority order of rows k*z is obtained by sorting adjacent variable node sets in descending order; S4. Decoding and scheduling are performed according to the obtained priority order.

Images