Large nuclear polarization code BP decoding method and system based on pruning permutation factor graph
By introducing the method of pruning permutation factor graphs in large-core polar code BP decoding, the problem of balancing decoding performance and complexity is solved, and the decoding performance is improved and the complexity is reduced. It is suitable for 5G and future 6G communication systems.
Patent Information
- Application Number
- CN202510826933.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-19
- Publication Date
- 2025-10-10
AI Technical Summary
Existing polar code decoding algorithms struggle to strike a balance between decoding performance and complexity. Especially for large-core polar codes, traditional BP decoding suffers from high decoding delay, high complexity, and is unsuitable for iterative decoding.
A large kernel polar code BP decoding algorithm based on pruned permutation factor graph is adopted. By randomly generating a permutation factor graph and pruning it, low-contribution nodes such as frozen variable nodes and Degree-1 check nodes and their connected variable nodes are deleted. The factor graph structure is optimized by combining iterative pruning and stopping conditions.
It achieves an effective balance between decoding performance and complexity, significantly improves decoding reliability, reduces decoding delay and computational complexity, and is suitable for coding schemes in 5G and future 6G communication systems.
Smart Images

Figure CN120768382A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to, but is not limited to, the field of polar code technology, and in particular relates to a large-kernel polar code BP decoding method and system based on a pruned permutation factor graph. Background Art
[0002] Polar codes have been adopted as the coding scheme for control channels in 5G enhanced mobile broadband (eMBB) scenarios and are also a key technology candidate for future 6G.
[0003] Arikan constructs polar codes using a 2×2 kernel matrix. Reference [2] shows that using a large kernel matrix G m (m ≥ 3) to replace the 2×2 kernel matrix, achieving higher polarization rates with more flexible code lengths. Fazeli et al. demonstrated that polar codes based on large kernels in a binary erasure channel (BEC) can not only approach the Shannon limit but also reach it with optimal scaling of their block length as a function of the capacity gap.
[0004] Professor Arikan also proposed the Successive Cancellation (SC) decoding algorithm. Tal et al. proposed the Successive Cancellation List (SCL) decoding algorithm. A major disadvantage of SC decoding and SCL decoding is the long decoding delay caused by their serial decoding characteristics, in which the information bits are decoded one at a time, reducing the overall throughput of the decoder. In addition, the SCL decoder does not provide soft input / soft output information and is not suitable for iterative decoding and detection. Gallager proposed the BP decoding algorithm. Arikan applied the BP decoding algorithm to polar code decoding. Compared with the decoding algorithm, the performance of the BP decoding algorithm is comparable to that of the SC decoding, and BP decoding uses parallel decoding. After the iteration is completed, the information bits are decoded simultaneously, which can achieve higher throughput than serial decoding.
[0005] For a polar code of length N, each layer of its decoding factor graph can be rearranged through permutation operations, generating different graph structures with the same decoding characteristics. Due to differences in the order of processing during decoding, the performance of a single permutation factor graph corresponding to each codeword and noise implementation varies. In other words, if decoding fails on a specific graph, other factor graphs can be tried until a preset stopping condition is met.
[0006] BP decoding using permutation factor graphs can improve accuracy, but it also introduces a certain amount of decoding delay. A paper proposes a method for finding a sparse parity check matrix suitable for iterative decoding for the corresponding LDPC decoder by pruning redundant variable nodes (VN) and check nodes (CN). Pruning removes nodes that have no contribution to decoding or some nodes with low contribution, reducing the number of operations required to perform a single permutation factor graph, thereby reducing complexity while minimizing the impact on bit error rate. Summary of the Invention
[0007] In view of the problems existing in the prior art, the present invention provides a large kernel polar code BP decoding algorithm based on a pruned permutation factor graph.
[0008] The present invention is implemented as follows: a large kernel polar code BP decoding algorithm based on a pruned permutation factor graph, the algorithm comprising:
[0009] S1: Select a 3×3 kernel matrix The codeword is constructed using the Gaussian approximation method; the encoding code length is 3 5 =243, code rate is 0.5; the codeword is modulated by binary phase shift keying and then passes through the additive white Gaussian noise channel;
[0010] S2: Using the factor graph generation algorithm, pseudo code for factor graph generation is given; N is the code length of the codeword, n = log3N is the number of layers of the factor graph; it is a function that generates a random permutation array of length n based on the input integer n;
[0011] S3: Iteratively prune the factor graph using the factor graph pruning algorithm until the size of the factor graph no longer changes; pruning the factor graph is an effective way to reduce latency. The literature defines different types of VNs: Channel variable node (VNCH): Figure 2 Represented by green nodes; Hidden variable node (Hiddenvariablenode.VNH): Figure 2 Represented by black nodes; frozen variable node (Frozen variable node, VNF); Figure 2 Represented by red nodes. Pruning can be performed on the following nodes, where Degree represents the number of connections between the node and other nodes:
[0012] Among them, H orig is the original factor graph; H pruncedFor the pruning factor graph, de_VNF means deleting VNF; de_d1CN means deleting degree1CN; con_d1VNCH means merging Degree-2CN and VNH into VNCH; de_d1VNH means deleting Degree-1VNH and the CN connected to it; con_d2VNH means merging the two CNs connected to Degree-2VNH into one; con_d1CN means merging the two VNHs connected to Degreee-2CN into one;
[0013] S4: The following stopping condition is adopted: the maximum number of factor graph selections is reached. The present invention adopts a maximum number of factor graph selections of 5; the practical judgment condition is: in, To receive the codeword, is the decoding result codeword; prior judgment conditions: Among them, u is the original codeword.
[0014] Furthermore, the factor graph generation algorithm specifically includes:
[0015] Input: N,n
[0016] Output: H prunced
[0017]
[0018] Furthermore, the factor graph pruning algorithm specifically includes:
[0019] Input: H orig
[0020] Output: H prunced
[0021]
[0022]
[0023] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:
[0024] First, the present invention proposes a large-kernel polar code BP decoding algorithm based on pruned permutation factor graphs. This algorithm effectively balances decoding performance and complexity by introducing the concepts of permutation factor graphs and pruned factor graphs. The algorithm first randomly generates a permutation factor graph and rearranges the layers of the decoding factor graph to generate different graph structures to improve decoding performance. Next, the generated permutation factor graph is pruned, removing nodes with low or no contribution to decoding, such as frozen variable nodes (VNFs), degree-1 check nodes (CNs), and their connected variable nodes (VNs), thereby reducing the computational effort. During the iterative decoding process, if a stopping condition is not met, the algorithm replaces the factor graph and continues to perform pruned permutation factor graph BP decoding until a stopping condition is met, such as reaching the maximum number of factor graph selections or the decoding result meeting a preset accuracy requirement. Simulation results show that using multiple permutation factor graphs for decoding can significantly improve decoding performance, but also increase decoding complexity. Pruning the factor graph effectively reduces decoding complexity while minimizing the impact on decoding performance. The 3×3 kernel matrix also allows for a wider range of code lengths. The pruned permutation factor graph BP decoding proposed in this paper can achieve a balance between bit error rate and complexity.
[0025] Second, the present invention effectively fills the technical gap in the field of BP decoding algorithms for large-core polar codes in China through an innovative permutation factor graph and node pruning strategy. This algorithm addresses the problem of difficulty in balancing performance and complexity in traditional BP decoding, and proposes to randomly generate and screen permutation factor graphs to avoid the short loop effect, thereby improving decoding reliability. By deleting low-contribution nodes (such as VNF, Degree-1 CN, etc.), the problem of increased complexity caused by the permutation factor graph is effectively improved, which is significantly better than the unpruned permutation factor graph scheme. This algorithm effectively fills the technical gap in the field of BP decoding algorithms for large-core polar codes in China through an innovative permutation factor graph and node pruning strategy. This algorithm addresses the problem of difficulty in balancing performance and complexity in traditional BP decoding, and proposes to randomly generate and screen permutation factor graphs to avoid the short loop effect, thereby improving decoding reliability. By deleting low-contribution nodes (such as VNF, Degree-1 CN, etc.), the problem of increased complexity caused by the permutation factor graph is effectively improved, which is significantly better than the unpruned permutation factor graph scheme, achieving a balance between low complexity and high performance under the framework of large-core polar codes. In addition, the algorithm supports a rich range of code length configurations, providing a coding scheme that is both reliable and flexible for 5G enhanced mobile broadband (eMBB) and future 6G communication systems, promoting the evolution of domestic communication technology towards higher efficiency and lower latency. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 This is a flow chart of a large kernel polar code BP decoding algorithm based on a pruned permutation factor graph provided by an embodiment of the present invention;
[0027] Figure 2 is a BP decoding factor graph of the P(9,4) polar code provided by an embodiment of the present invention;
[0028] Figure 3 The original factor graph (left) and permutation factor graph (right) of the P(9,4) polar code provided by an embodiment of the present invention are shown.
[0029] Figure 4 This is a comparison of bit error rates among original BP decoding, permutation factor graph BP decoding, and pruned permutation factor graph BP decoding provided by an embodiment of the present invention.
[0030] Figure 5 This is a comparison diagram of the Polar decoding BER performance under 60 iterations provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0031] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0032] The present invention uses P(N,K) to represent polar codes. The present invention does not discuss multi-core polar codes, so the code length N = m n (n,m∈N * ), K is the length of information bits.
[0033] Let W:X→Y be a binary discrete memoryless channel (BDMC), where X={0,1} and the transition probability is W(x|y). Represents a vector (a0, a1, ..., a N-1 ). Let W N :X N →Y N represents a channel consisting of N independent channels W, so for each and The conditional transition probability is:
[0034]
[0035] The polar code generation matrix is:
[0036]
[0037] Among them, B N is the bit-flip matrix, Do n Kronecker products for the kernel matrix, G m is an m×m kernel matrix.
[0038] The polar code encoding principle is:
[0039]
[0040] in, is the original codeword, is the encoded codeword.
[0041] The factor graph of the polar code BP decoder contains n=log m N layers, each layer contains two types of nodes, VN and CN. The factor graph contains N·(n+1) VN{v ij} and N·n CN{c ik}, where 1≤i≤N,1≤j≤n,1≤k≤n. Taking P(9,4) polar code as an example, its factor graph is as follows Figure 1 shown.
[0042] The present invention uses x v→c Denotes the log-likelihood ratio (LLR) from VNv to CNc, with y v→c represents the LLR from the check node c to the variable node v. In addition, the present invention defines c v is the set of variable nodes connected to the check node c, v c is the set of check nodes connected to the variable node v, such as Figure 2 shown.
[0043] The update of CN is calculated as follows:
[0044]
[0045] in,
[0046]
[0047] The update of VN is calculated as follows:
[0048]
[0049] in, is the initial value of the variable node v.
[0050] 1.2 Permutation Factor Graph
[0051] Polar code-specific generator matrix G N This results in a large number of short cycles in the decoding factor graph. The existence of short cycles greatly affects the reliability of message node updates, such as Figure 3 shown.
[0052] The factor graph obtained by permuting the factor graph hierarchy corresponding to the generator matrix is called permuted factor graph. It is shown in the literature that BP decoding can be performed on different factor graphs and the performance of BP decoding can be improved. Figure 3 The original factor graph of P(9,4) polar code and one of the permuted factor graphs are given. In fact, for a polar code with a kernel matrix G m , N = m n , there are n! valid permuted factor graphs.
[0053] It can be seen from Figure 3 that different permutation of factor graph makes the order of decoding process different, and the decoding performance of different factor graph is different. Using permuted factor graph can improve the bit error rate, and the reason for this performance improvement is that the order of different factor graph is different, the iteration loop structure is different, and the short loop is avoided to improve the decoding performance.
[0054] The number of random permuted factor graphs required by the BP decoding of the permuted factor graph of the polar code is often very large, thus causing a certain decoding delay.
[0055] 1.3 Pruning factor graph
[0056] Pruning factor graph is an effective method to reduce the delay. The literature defines different types of VN:
[0057] Channel variable node (VNCH): in Figure 2 represented by green nodes.
[0058] Hidden variable node (VNH): in Figure 2 represented by black nodes.
[0059] Frozen variable node (VNF): in Figure 2 represented by red nodes.
[0060] The following nodes can be pruned, and Degree represents the number of connections between the node and other nodes:
[0061] (1) VNF: VNF has no contribution to the decoding process, and it can be deleted from the factor graph.
[0062] (2) Degree-1 CN and VN connected to it: Degree-1 CN is a CN connected to only one VN, which can be deleted from the graph.
[0063] (3) Degree-1 VNH: Degree-1 VNH and the CN connected to it can be deleted together.
[0064] (4) Degree-2 CN connected with Degree-1 VNCH: Degree-2 CN can be deleted, and VNH must be replaced by VNCH.
[0065] (5) Degree-2 VNH: for Degree-2 VNH connected with only two adjacent CNs, two CNs can be combined into one to delete Degree-2 VNH.
[0066] (6) Degreee-2 CN: CNs are not connected, that is, Degreee-2 CN is connected with only two VNHs. Degreee-2 CN can be deleted by combining two VNHs into one.
[0067] As shown in the prior art, Figure 1 the embodiment of the present application provides a BP decoding algorithm of big core polar code based on pruning substitution factor graph, which comprises the following steps:
[0068] S1: selecting a 3*3 core matrix Using the Gaussian approximation method to construct a code word; the code length is 3 5 =243, and the code rate is 0.5; after the code word is modulated by binary phase shift keying, the code word is transmitted through an additive white Gaussian noise channel.
[0069] S2: using a factor graph generation algorithm, a pseudo code for generating a factor graph is given; N is the code length of the code word, n=log3N is the number of layers of the factor graph; is a function for generating a random permutation array with a length of n according to an input integer n;
[0070] S3: using a factor graph pruning algorithm to iteratively prune the factor graph until the size of the factor graph no longer changes; wherein, H orig is an original factor graph; H prunced de_VNF represents deleting VNF; de_d1CN represents deleting degree 1 CN; con_d1VNCH represents combining Degree-2 CN and VNH into VNCH; de_d1VNH represents deleting Degree-1 VNH and the CN connected therewith; con_d2VNH represents combining two CNs connected with Degree-2 VNH into one; and con_d1CN represents combining two VNHs connected with Degreee-2 CN into one.
[0071] S4: the following stopping condition is adopted: the maximum factor graph selection times are reached. In the present application, the maximum factor graph selection times are 5; the practical judgment condition is: wherein, is a received code word, is a decoding result code word; and the prior judgment condition is: Among them, u is the original codeword.
[0072] Furthermore, the factor graph generation algorithm specifically includes:
[0073] Input: N,n
[0074] Output: H prunced
[0075]
[0076] Furthermore, the factor graph pruning algorithm specifically includes:
[0077] Input: H orig
[0078] Output: H prunced
[0079]
[0080]
[0081] The present invention provides simulation results to show the impact of the proposed large core polar code BP decoding algorithm with pruned permutation factor graph on decoding performance. Figure 4 、 Figure 5 As shown in the figure, at different iteration numbers, the bit error rate (BER) of decoding using the permutation factor graph is significantly reduced compared to traditional BP decoding. At a BER of 10-2, there is a gain of approximately 0.3 dB. BP decoding using the pruned factor graph has a smaller impact on the BER, with a maximum reduction of approximately 0.1 dB.
[0082] Decoding method Average activated CN Average spread information Average latency BP <![CDATA[1.6×10 4 ]]> <![CDATA[3.8×10 4 ]]> 60 Permutation Factor Graph <![CDATA[9.2×10 4 ]]> 14.0 x 10 4 ]] 280 Trim displacement <![CDATA[2.5×10 4 ]]> <![CDATA[4.8×10 4 ]]> 97
[0083] Table 1. Comparison of the complexity of BP decoding and pruned permutation factor graph BP decoding for P(243,122) polar code under BER=10-4, Nit,max=200.
[0084] Table 1 compares the complexity of the three decoding methods. In terms of average activated CN, average propagated information, and average delay, BP decoding has the lowest complexity. The complexity of the permutation factor graph is more than three times that of BP decoding. The complexity of the pruned permutation factor graph is between the two and less than twice the complexity of BP decoding.
[0085] The BER of pruned permutation factor graph BP decoding is close to that of permutation factor graph BP decoding, which is significantly improved compared with BP decoding. At the same time, the complexity can be controlled within twice that of BP decoding, which has better decoding performance.
[0086] Figure 4Polar decoding BER performance comparison under 30 iterations Figure 5 Polar decoding BER performance comparison under 60 iterations
[0087] Figure 4 The BER variation trend of three different algorithms, original BP decoding (red square), permutation factor graph BP decoding (blue square) and pruning permutation factor graph BP decoding (blue diamond), with the increase of Eb / N0 is shown when the decoding iteration is fixed at 30 times. Overall, the blue curve and the purple curve are always below the red curve, indicating that the message passing path obtained by factor graph permutation and subsequent pruning is more efficient, which can significantly reduce the bit error rate within the same iteration budget.
[0088] From 1dB to 4dB, the maximum performance gain of the pruning permutation factor graph scheme compared with the original BP appears at the high SNR end (3.5-4dB), and the BER decreases by nearly half an order of magnitude; in the low Eb / N0 segment (1-2.5dB), it still maintains a stable advantage, but the benefit is slightly smaller. This shows that the pruned sparse graph not only reduces redundant cycles, but also retains enough information flow channels, making the iteration converge faster.
[0089] Figure 5 When the number of iterations is increased to 60, the overall bit error rate curve moves further down. With the number of iterations doubled, the performance gap between the three curves still exists and slightly expands: the BER advantage of permutation factor graph and pruning permutation factor graph is particularly obvious at 2.5-3dB. It is worth noting that the slope of the blue diamond curve is slightly larger than that of the blue square curve at high iteration number, indicating that the pruning strategy has more gain in the late stage of convergence.
[0090] The permutation factor graph BP decoding can significantly improve the parallel message passing efficiency of the polar code, and the addition of topology pruning on its basis can further compress the short loop and reduce the invalid node update, achieving the best compromise between performance and complexity. Therefore, for communication scenarios with high real-time requirements or limited hardware resources, pruning permutation factor graph BP decoding provides a feasible solution to obtain lower BER within a limited number of iterations.
[0091] Embodiment one: 5G eMBB downlink control channel receiving end
[0092] In the 5G NR smartphone baseband, the base station adopts 3x3 kernel polar code for the downlink control channel, and the code length is set to N=3 6=729, with a code rate of 0.5, corresponding to a 364-bit information payload. The receiving terminal first constructs a frozen bit set using Gaussian approximation according to step S1, then completes AWGN channel reception of BPSK modulated frames. S2 invokes the permutation factor graph generation algorithm to obtain a 6-layer randomly permuted initial factor graph. S3 then iteratively executes six pruning rules, including "deleting VNFs / merging CNs," resulting in an average of only three rounds of pruning to obtain a sparse graph. During decoding, the maximum number of factor graph selections is set to 5. The practical judgment conditions are: The iteration is terminated early with the a priori CRC check. Actual chip tests show that at 3dB Eb / N0, the pruned permutation factor graph BP reduces at least 35% of multiplication and addition operations compared to the original BP, shortens the decoding delay by 28ns, and improves the BER by about 0.3dB.
[0093] Example 2: Low-orbit satellite downlink data transmission link
[0094] In the satellite-to-ground narrowband scenario, the system uses N=3 7 =2187 is used to adapt to telemetry packets of more than 1000 bits. GA construction and BPSK modulation are performed on the satellite before transmission; the ground station receiver (based on Kintex-7 FPGA) randomly generates a 7-layer permutation factor graph in S2 and stores it in Block RAM; then it is pruned layer by layer according to the S3 pipeline mode. The number of nodes in the pruned factor graph is reduced from 13068 to 6798, and the storage occupancy is halved. The decoding adopts a BP maximum 30 cycles + a maximum 5-time factor graph selection strategy, combined with CRC-10 prior termination (S4). Under the condition of 2.5dB Eb / N0, this scheme reduces the decoding power consumption by 42% compared with the traditional BP, saves 37% of FPGA on-chip resources, and reduces the BER from 1.8*10 -2 Reduced to 9.4*10 -3 , meeting the dual requirements of satellite link budget and ground station real-time processing.
[0095] It should be noted that embodiments of the present application can be realized by hardware, software, or a combination of software and hardware. The hardware portion can be realized by a special logic; the software portion can be stored in a memory and executed by a proper instruction execution system, such as a microprocessor or a specially designed hardware. A person of ordinary skill in the art can understand that the above-mentioned apparatus and method can be realized by computer executable instructions and / or included in processor control codes, such as a carrier medium, such as a magnetic disk, CD or DVD-ROM, a programmable memory, such as a read-only memory (firmware), or a data carrier, such as an optical or electronic signal carrier. The apparatus of the present application and its modules can be realized by a hardware circuit, such as a very large scale integrated circuit or a gate array, a semiconductor, such as a logic chip, a transistor, or a programmable hardware device, such as a field programmable gate array, a programmable logic device, or the like, by software executed by various types of processors, or by a combination of the above-mentioned hardware circuit and software, such as firmware.
[0096] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any modification, equivalent replacement, and improvement within the technical range disclosed by the present application, and within the spirit and principle of the present application, should be covered within the protection scope of the present application.
Claims
1. A large kernel polar code BP decoding method based on pruned permutation factor graph, characterized in that: The method comprises: a. Select a three-by-three kernel matrix and use Gaussian approximation to construct the polar code, with a code length of N, a code rate of 0.5, and binary phase-shift keying modulation, and transmit it over an additive white Gaussian noise channel; b. Construct a permutation factor graph based on the factor graph generation algorithm, where N is the code length, n is the base-3 logarithm of N, and the random permutation function generates a permutation array of length n; c. Perform iterative pruning on the permutation factor graph, including deleting VNFs, deleting CNs with degree 1, merging CNs with degree 2 and VNHs into VNCHs, deleting VNHs with degree 1 and their connected CNs, merging two CNs connected to a VNH with degree 2, and merging two VNHs connected to a CN with degree 2, until the factor graph size remains unchanged; d. Perform belief propagation decoding on the pruned factor graph, and the process ends when any of the following stopping conditions is met: the factor graph selection number reaches five; the Hamming distance between the received codeword and the decoded codeword is zero; the Hamming distance between the decoded codeword and the original codeword is zero; e. Output the decoding result.
2. The method according to claim 1, characterized in that The factor graph generation algorithm in step b completes the connection between variable nodes and check nodes in sequence through nested triple loops, ensuring that the connection relationship at each layer satisfies the permutation rule.
3. The method according to claim 1, characterized in that The pruning in step c calculates the total number of factor graph nodes at the end of each iteration and proceeds to the next round of pruning only if the total number of nodes changes.
4. The method according to claim 1, wherein The belief propagation decoding in step d adopts the log-likelihood ratio message passing form and updates the message values of all variable nodes and check nodes after each iteration.
5. A polar code decoding device, characterized in that: The device includes a processor, a memory and a decoding module, wherein the decoding module is configured to execute the large kernel polar code BP decoding method based on the pruned permutation factor graph according to any one of claims 1 to 4.
6. The device according to claim 5, characterized in that The decoding module includes a factor graph construction subunit, a factor graph pruning subunit and a belief propagation subunit, which communicate in sequence through an internal bus.
7. A communication receiving system, characterized in that: The polar code decoding device comprises a channel receiver and the polar code decoding device according to claim 5, wherein the channel receiver provides the decoding device with a received codeword modulated by binary phase shift keying and transmitted through an additive white Gaussian noise channel.
8. A data processing device, characterized in that: The device implements the method according to any one of claims 1 to 4 in hardware logic, and improves decoding throughput through a parallel message passing architecture.
9. A computer-readable storage medium having instructions stored thereon, wherein when the instructions are executed by a processor, the processor is caused to execute the method according to any one of claims 1 to 4.
10. An integrated circuit chip, characterized in that: A permutation factor graph construction unit, a factor graph pruning unit, and a belief propagation decoding unit are integrated on a single chip to implement the decoding method described in any one of claims 1 to 4 in a hardware manner.
Citation Information
Cited By
Decoding method and device, electronic equipment and non-transient computer readable storage medium
CN121567285A
Decoding method, device, electronic equipment and non-transitory computer-readable storage medium
CN121567285B
