LDPC decoding method and device based on MIPS architecture, medium and product

Through the parallel processing of the kernel binding technology and the MSA instruction set, the problem of low LDPC decoding performance on the MIPS architecture is solved, high throughput and low latency are achieved, and the performance needs of domestic communication systems are met.

CN119051670APending Publication Date: 2024-11-29COMMUNICATION UNIVERSITY OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411081258.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

On CPUs based on MIPS architecture, LDPC decoding performance is limited by the lack of effective SIMD vector extension instruction set, resulting in low throughput and difficult to meet the high performance needs of domestic communication systems.

Method used

The kernel binding technology is used to bind the LDPC decoding task with the CPU core of the MIPS architecture, and the data is segmented and parallelized by using the MSA instruction set, combining the normalized minimum sum or offset minimum sum algorithm, and through fixed-point processing and data Recombination improves decoding efficiency.

Benefits of technology

It significantly improves the throughput and system performance of LDPC decoding, reduces latency, and is suitable for high-demand wireless communication systems, achieving similar performance improvements to X86 and ARM architectures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119051670A_ABST
    Figure CN119051670A_ABST
Patent Text Reader

Abstract

The invention discloses an LDPC decoding method, device, medium and product based on an MIPS architecture, and relates to the field of communication, the method comprises the following steps: binding an LDPC decoding task with a CPU core outside an operating system core of a general CPU based on the MIPS architecture by using a core binding technology; carrying out fixed-point processing on log-likelihood ratio (LLR) data to be decoded; the processed data are grouped; performing data segmentation on the grouped data; simultaneously performing LDPC decoding iteration operation on each group of data after data segmentation by using an MSA instruction set of an MIPS architecture; and carrying out data recombination on decoding results of all the data after data segmentation to obtain a decoding result. According to the method, the LDPC decoding throughput rate can be improved through the parallel instructions on the basis of the SIMD technology on the CPU ecology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of communications, and in particular to an LDPC decoding method, device, medium and product based on a MIPS architecture. Background Art

[0002] Low Density Parity Check (LDPC) is a linear block error correction code proposed by Robort Gallagner in 1962. Its characteristic is that the check matrix is ​​extremely sparse. Due to the excellent performance of LDPC and the convenience of hardware implementation, it has been widely used as a channel coding scheme in various communication standards. Such as 5G-NR, DVB-S2, CMMB, IEEE802.16E, etc. At present, the bottleneck that limits the performance of LDPC is that LDPC decoding requires a lot of calculations. Optimizing the LDPC decoding scheme can bring higher throughput and better overall performance. LDPC decoding algorithms can be divided into two categories: hard decision algorithms and soft decision algorithms. In order to obtain better bit error rate performance, an improved minimum sum algorithm based on the belief propagation algorithm is usually used for LDPC decoding.

[0003] With the development and maturity of communication technology, wireless communication systems are evolving towards intelligence, flexible customization, and software deployment. Software radio platforms based on general-purpose CPUs are becoming a hot research topic. Among them, the biggest bottleneck limiting the rate of the entire communication system is the LDPC decoding part. Therefore, optimizing the LDPC decoding scheme based on the characteristics of general-purpose CPUs and improving throughput are important ways to improve the overall performance of the system.

[0004] General-purpose CPUs can be divided into two categories according to different processor architectures: general-purpose CPUs based on complex instruction sets and general-purpose CPUs based on simple instruction sets. The former is typically represented by the X86 architecture, while the latter are typically represented by the ARM architecture and the MIPS architecture. The Loongson 3A series independently designed and developed by the leading Chinese chip company Loongson uses the MIPS architecture, which has a very different technical route from the X86 architecture and needs to be optimized for the LDPC decoder based on the characteristics of the MIPS architecture.

[0005] Due to process limitations, after the CPU frequency reaches above 3GHz, the performance gain brought by continuing to increase the clock frequency becomes smaller and smaller, and power wall and energy consumption wall problems will appear. Therefore, the method of processing multiple sets of data simultaneously within one CPU clock is usually adopted for parallel processing to increase throughput. For example: Single Instruction Multiple Data (SIMD), Single Program Multiple Data (SPMD), Single Instruction Multiple Thread (SIMT), etc. In recent years, the vector extension instruction set based on SIMD architecture has developed rapidly, and its supported bit width has developed from the early 64bit bit width to a higher degree of parallelism. Intel launched the AVX512 instruction set in 2016, which supports 512-bit wide vector extension instruction set for the first time, and has plans to launch a 1024-bit wide instruction set. The Loongson series processors support the MSA instruction set for SIMD operations with a bit width of 128. When using SIMD technology for LDPC decoding parallel processing, it is necessary to optimize the processing according to the characteristics of the instruction set.

[0006] Since the decoding algorithm of LDPC is relatively complex but has a good parallel structure, parallel design of LDPC decoder can obtain higher throughput. On the general CPU platform based on X86 architecture, many papers have made work related to LDPC decoding parallel computing. For example, in the literature (Gal BL, Jego C. High-Throughput Multi-Core LDPC Decoders Based on x86 Processor [J]. IEEE Transactions on Parallel & Distributed Systems, 2016, 27 (5): 1373-1386.), Intel's high vector extension instruction set (SSE4.2) and advanced vector extension instruction set 2 (AVX2) are used to implement data frame-level parallel offset minimum sum (OMS) and normalized minimum sum (NMS) LDPC decoder design on a multi-core GPP device based on traditional SIMD architecture. In the literature (Y.Xu, W.Wang, Z.Xu and X.Gao, "AVX-512Based Software Decoding for 5GLDPCCodes," 2019IEEE International Workshop on Signal Processing Systems (SiPS), 2019, pp.54-59.), the QC-LDPC of the 5GNR standard is implemented on a multi-core GPP device through Intel's AVX512 high vector extension instruction set. For example, the patent (a software implementation QC-LDPC decoding method based on single instruction multiple data stream) uses a vector extension instruction set based on the traditional SIMD architecture to greatly improve the decoder performance, and optimizes memory reading and writing and preprocessing based on the characteristics of the instruction set in mask processing. The patent (a multi-rate and multi-code length LDPC code decoding method based on the SIMD instruction set) combines the characteristics of the quasi-cyclic structure LDPC and uses a fixed-point layered decoding scheme to implement the design of an LDPC decoder based on the vector extension instruction set of the traditional SIMD architecture.

[0007] Since the CPU ecosystem based on the MIPS architecture is relatively weak, there are very few solutions for LDPC decoding in the CPU ecosystem using the SIMD vector extension instruction set MSA of the MIPS architecture. In order to implement parallel acceleration and performance improvement of LDPC decoders on domestic CPUs based on the MIPS architecture, it is necessary to design a general high-throughput LDPC decoding method and system based on the MIPS architecture to meet the requirements for LDPC decoders in different application scenarios in domestic communication systems, and to improve the LDPC decoding throughput through parallel instructions based on SIMD technology, and to have stable acceleration performance when inputting data with different code lengths and different code rates under different standards. Summary of the invention

[0008] The purpose of this application is to provide an LDPC decoding method, device, medium and product based on the MIPS architecture, which can improve the LDPC decoding throughput rate through parallel instructions based on SIMD technology in the CPU ecosystem.

[0009] To achieve the above objectives, this application provides the following solutions:

[0010] In a first aspect, the present application provides an LDPC decoding method based on a MIPS architecture, and the LDPC decoding method based on a MIPS architecture includes:

[0011] The core binding technology is used to bind the LDPC decoding task to the CPU core other than the operating system core of the general-purpose CPU based on the MIPS architecture;

[0012] The LLR data to be decoded is processed into fixed points; and the processed data is grouped;

[0013] Segment the grouped data;

[0014] The MSA instruction set of the MIPS architecture is used to perform LDPC decoding iteration operations on each group of data after data segmentation;

[0015] The decoding results of all the data after data segmentation are reorganized to obtain the decoding results.

[0016] Optionally, the algorithm of the LDPC decoding iterative operation includes: a normalized minimum sum decoding algorithm or a shifted minimum sum algorithm.

[0017] Optionally, the MSA instruction set of the MIPS architecture is used to simultaneously perform LDPC decoding iteration operations on each group of data after the data is segmented, specifically including:

[0018] The subtraction instruction in the MSA instruction set is used to determine the value transmitted from the message node to the check node in the current iteration number, and the maximum value instruction and the minimum value instruction are used to perform the limiting operation;

[0019] The XOR instruction, absolute value instruction, subtraction instruction and data mixing instruction in the MSA instruction set are used to determine the value passed from the check node to the message node in the current iteration number;

[0020] Determine the absolute value of the value passed from the message node to the check node and the absolute value of the value passed from the check node to the message node using the absolute value instruction in the MSA instruction set;

[0021] Use the maximum value instruction and minimum value instruction in the MSA instruction set to limit the absolute value;

[0022] The absolute value of the value passed from the message node to the check node and the absolute value of the value passed from the check node to the message node are used to determine the value of the message node in the current iteration number by using the bitwise AND instruction in the MSA instruction set; and the maximum value instruction and the minimum value instruction are used to perform amplitude control according to the minimum sum algorithm principle;

[0023] Determine whether the current number of iterations is less than the maximum number of iterations; if so, add 1 to the number of decoding iterations and iterate again; if it is the maximum number of iterations, use the comparison instructions in the MSA instruction set to perform a hard decision operation and output the decoding result of the data after data segmentation.

[0024] Optionally, judging whether the current number of iterations is less than the maximum number of iterations, the process also includes:

[0025] An early stopping mechanism is adopted; the early stopping mechanism performs hard decision decoding according to the value of the message node updated according to the current iteration number.

[0026] Optionally, the decoding results of all the data after the data segmentation are reorganized to obtain the decoding results, which specifically includes:

[0027] Use the data movement instructions and data mixing instructions in the MSA instruction set of the MIPS architecture to reorganize data;

[0028] The data is stored using the data load instruction in the MSA instruction set of the MIPS architecture.

[0029] In a second aspect, the present application further provides an LDPC decoding device based on a MIPS architecture, wherein the LDPC decoding device based on a MIPS architecture comprises:

[0030] A core binding module, used to bind the LDPC decoding task to a CPU core other than the operating system core of a general-purpose CPU based on the MIPS architecture by using the core binding technology;

[0031] The data preprocessing module is used to perform fixed-point processing on the log-likelihood ratio (LLR) data to be decoded, and to group the processed data;

[0032] A data grouping module, used for segmenting the grouped data;

[0033] LDPC decoding module, used to use the MSA instruction set of MIPS architecture to perform LDPC decoding iteration operation on each group of data after data segmentation;

[0034] The data reorganization module is used to reorganize the decoding results of all the data after data segmentation to obtain the decoding results.

[0035] In a third aspect, the present application also provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the LDPC decoding method based on the MIPS architecture.

[0036] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the LDPC decoding method based on the MIPS architecture.

[0037] In a fifth aspect, the present application also provides a computer program product, including a computer program, which implements the LDPC decoding method based on the MIPS architecture when executed by a processor.

[0038] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0039] The present application provides a LDPC decoding method, device, medium and product based on MIPS architecture, which uses core binding technology to bind LDPC decoding tasks to CPU cores other than the operating system core of a general-purpose CPU based on MIPS architecture, avoiding interruptions in the operating system process, frequent context switching, and reducing CPU processing capabilities. It is not limited to specific communication standards and can run on all MIPS architecture CPUs (such as Loongson A series general-purpose CPUs and Ingenic X series general-purpose CPUs), and has good applicability. The vector extension instruction set capability of the MSA instruction set of the MIPS architecture is utilized to process multiple groups of data simultaneously in parallel, which greatly improves the throughput performance. Further, it can effectively reduce the LDPC decoding delay, and is suitable for application in wireless communication systems with high requirements for real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A flowchart of an LDPC decoding method based on a MIPS architecture provided in one embodiment of the present application;

[0042] Figure 2 A schematic diagram of the principle of adding an early stopping mechanism to an LDPC decoding method based on a MIPS architecture provided in one embodiment of the present application;

[0043] Figure 3 This is a schematic diagram of data splitting;

[0044] Figure 4 This is a schematic diagram of the VN_to_CN information update process;

[0045] Figure 5 This is a schematic diagram of the CN_to_VN information update process;

[0046] Figure 6 This is a schematic diagram of data reorganization.

[0047] Figure 7 This is a schematic diagram of throughput improvement under the 5G-NR standard;

[0048] Figure 8 Schematic diagram of throughput improvement under different code lengths. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0051] In an exemplary embodiment, Figure 1 As shown, a LDPC decoding method based on the MIPS architecture is provided, which is executed by a computer device, specifically, it can be executed by a computer device such as a terminal or a server alone, or it can be executed by a terminal and a server together. In the embodiment of the present application, the method includes the following S101 to S105. Among them:

[0052] S101, using core binding technology to bind the LDPC decoding task to a CPU core other than the operating system core of a general-purpose CPU based on the MIPS architecture; core binding technology includes soft affinity and hard affinity; among them, soft affinity is to adjust the priority of the LDPC task to the highest in the Linux system settings, so that it can be processed on the same CPU core as persistently as possible, and frequent switching can be avoided as much as possible; hard affinity is to force the process to be executed on the selected CPU core no matter what, and completely prohibit the operating system from switching, which may cause necessary system processes to be blocked and cause the system to crash, and the Linux system kernel version (5.15) used in this application supports efficient scheduling, and soft affinity binding can meet the requirements, so the soft affinity solution is selected.

[0053] Specifically, the potential of multi-core CPUs is fully utilized and tasks are divided into n groups (n is the number of CPU cores), where core 0 is used to run the Linux operating system and other services other than LDPC decoding. The core binding technology is used to bind the LDPC decoding task to cores other than the operating system core to avoid interruptions to the operating system process, frequent context switching, and reduced CPU processing power.

[0054] Assume that the hardware platform used by the decoder has N CPU cores, numbered CPU0CPU1...CPU(N-1). This application will allocate cores for services other than LDPC decoding according to the amount of resources occupied by the operating system. Due to the characteristics of the Linux system, some important threads must be started from CPU0 to trigger the subsequent operation of the system. Therefore, CPU0 is used as the system core and the CPU is divided into two groups. The first group contains only one CPU, CPU0, which is used to run operating system-related processes and processes other than LDPC decoding. The second group contains N-1 cores from CPU1 to CPU(N-1), and the system priority is set to 90 (the highest except for system processes). In the LDPC decoding design, all LDPC decoding-related threads are bound to the second group of CPUs. The advantage of this is that interrupts are reduced to 10%-20% of those without core binding. After the interrupts are reduced, the delay caused by system context switching is greatly improved, which is conducive to improving the performance of the LDPC decoder. Experimental results show that the use of core binding technology can reduce the delay by about 60%, effectively enhancing the real-time performance of the system.

[0055] S102, performing fixed-point processing on the log-likelihood ratio (LLR) data to be decoded; and grouping the processed data;

[0056] The log-likelihood ratio LLR data is a high-precision floating-point number that requires 32 bits for storage, calculation, and processing, which wastes resources. Therefore, the input LLR data is fixed-pointed to 8 bits through a data fixed-point solution. The fixed-point method is to process the input floating-point LLR data through the max and min instructions in the vector extension instruction set. The processing power is increased fourfold after fixed-pointing. Then the data is split and the input LLR data is divided into multiple groups, each with 128 bits. If the input data length is N, it is divided into Each group of data with a length of 128 is further divided into 16 groups, each group contains only one 8-bit fixed-point data, and Figure 3 shown.

[0057] S103, segmenting the grouped data;

[0058] S104, using the MSA instruction set of the MIPS architecture (a vector extension instruction set of MIPS with a 128-bit bit width based on the MIPS architecture processor, MIPS SIMDARCHITECUTRE) to simultaneously perform LDPC decoding iteration operations on each group of data after data segmentation; the algorithms of the LDPC decoding iteration operations include: normalized minimum sum decoding algorithm (NMS) or offset minimum sum algorithm (OMS).

[0059] S104 specifically includes:

[0060] The subtraction instruction in the MSA instruction set is used to determine the value transmitted from the message node to the check node in the current iteration number, and the maximum value instruction and the minimum value instruction are used to perform the limiting operation;

[0061] The XOR instruction, absolute value instruction, subtraction instruction and data mixing instruction in the MSA instruction set are used to determine the value passed from the check node to the message node in the current iteration number;

[0062] Determine the absolute value of the value passed from the message node to the check node and the absolute value of the value passed from the check node to the message node using the absolute value instruction in the MSA instruction set;

[0063] Use the maximum value instruction and minimum value instruction in the MSA instruction set to limit the absolute value;

[0064] The absolute value of the value passed from the message node to the check node and the absolute value of the value passed from the check node to the message node are used to determine the value of the message node in the current iteration number by using the bitwise AND instruction in the MSA instruction set; and the maximum value instruction and the minimum value instruction are used to perform amplitude control according to the minimum sum algorithm principle;

[0065] Determine whether the current number of iterations is less than the maximum number of iterations; if so, add 1 to the number of decoding iterations and iterate again; if it is the maximum number of iterations, use the comparison instructions in the MSA instruction set to perform a hard decision operation and output the decoding result of the data after data segmentation.

[0066] S105, reorganizing the decoding results of all the data after data segmentation to obtain a decoding result.

[0067] In a specific embodiment, LDPC decoding iteration operation is performed on 16 groups of data, and the specific scheme is as follows:

[0068] Set the row iteration count m to 1, and set the value of VN_to_CN (message node to check node) for all rows in the current row Calculate, where i represents the number of decoding iterations, n and m represent the nth information node VN and the mth check node CN respectively. The specific calculation formula is as follows:

[0069]

[0070] Calculate the value of VN_to_CN After that, you need to adjust the value of CN_to_VN The present application can use the normalized minimum sum or the offset minimum sum method to calculate the value of CN_to_VN. Make updates;

[0071] If the LDPC decoding algorithm is set to the OMS algorithm, the calculation formula is as follows:

[0072]

[0073] If the LDPC decoding algorithm is set to the NMS algorithm, the calculation formula is as follows:

[0074]

[0075] The parameter α is the normalization parameter of the NMS algorithm, the parameter β is the offset parameter of the OMS algorithm, and n' represents an integer between 1 and N.

[0076] After the calculation is completed, the absolute value of the data is taken and the amplitude is limited to ensure that the data value is greater than the minimum limit. Then the row subscript m is incremented by 1 to point to the next position, and the next row of related and Calculation.

[0077] After the relevant calculation of the check node CN of each row is completed, the value of the message node VN is updated. The subscript of the message node VN is n. Read the value of VN_to_CN required each time and the value of CN_to_VN Determine the value E of the message node VN based on the sum of the two n ;

[0078] like Figure 2 As shown in the figure, in order to reduce the number of iterations, save time, and reduce the amount of calculation, an early stopping mechanism (ET) is used; the early stopping mechanism performs hard decision decoding according to the value of the message node updated according to the current number of iterations. However, if the early stopping mechanism is used, any thread that meets the early stopping condition and stops decoding needs to perform data splitting work, which will increase the processing time and may lead to a decrease in decoding efficiency. Therefore, it is necessary to decide whether to use the early stopping mechanism based on the actual working conditions of the decoder.

[0079] As another specific embodiment, the specific steps of parallel processing based on the MSA instruction set are as follows:

[0080] 1. Initialize the number of decoding iterations i and set it to 0.

[0081] 2. Initialize the parameters of LDPC decoding, including setting the maximum number of iterations (max_iteration), setting the degree distribution Colomn_Degrees, and the number of threads that can be opened max_threads. And initialize the number of row iterations m and column iterations n, and set them both to 0.

[0082] 3. The value E of the message node VN of the corresponding row n The E in the cache is read by the load instruction. n data.

[0083] 4. Complete VN_to_CN through subtraction instruction The value is calculated and the data is limited by the maximum and minimum value instructions.

[0084] LLR (log likelihood ratio) data of the check node CN, the data format is a floating point number. LLR data of the information node VN. VN_to_CN refers to the information update process from the information node VN to the check node CN during the LDPC decoding iteration process;

[0085] Taking 6 verification nodes CN and 4 information nodes VN as an example, the VN_to_CN information update process is as follows: Figure 4 shown.

[0086] Taking 6 verification nodes CN and 4 information nodes VN as an example, the CN_to_VN information update process is as follows: Figure 5 shown.

[0087] Calculated After the value is obtained, the sign bit instruction (sign instruction) is used to obtain each The sign of the value is checked, and all (positive and negative) sign values ​​except n=m (where n is the number of row iterations and m is the number of column iterations) are multiplied by an XOR instruction;

[0088] 5. Use the min command to record during the calculation process The minimum value minmun and the second smallest value second_smallest_value.

[0089] 6. Calculated by combining XOR instruction, absolute value instruction, subtraction instruction and data mixing instruction The value of .

[0090] 7. Convert VN_to_CN Value and CN_to_VN The absolute value instruction is used to obtain the absolute value of the data.

[0091] 8. Convert VN_to_CN Value and CN_to_VN The absolute value instruction is used to obtain the absolute value of the data, and then the maximum value instruction and the minimum value instruction are used to limit the data. Finally, the bitwise AND instruction is used to sum the transmitted information data to obtain the new E n value, and use the maximum and minimum value instructions to control the amplitude according to the minimum and algorithm principle.

[0092] 9. Add 1 to the number of decoding iterations, and if it does not exceed the maximum number of iterations max_iteration, repeat step 2, otherwise proceed to the next step to start the hard decision operation.

[0093] After the data is grouped according to the 128-bit register of the CPU, the above decoding result is 16 groups of 8-bit data, and the data needs to be reorganized, that is, the data is reordered; Figure 6 As shown, the reordering operation is:

[0094] Use the data movement instructions and data mixing instructions in the MSA instruction set of the MIPS architecture to reorganize data;

[0095] The data is stored using the data load instruction in the MSA instruction set of the MIPS architecture.

[0096] This application fully considers the characteristics of MIPS architecture general-purpose CPU, which is specifically manifested in: using the support of MIPS architecture general-purpose CPU for SIMD technology, the decoding process is accelerated in parallel. Since the MIPS architecture CPU has 32 128-bit registers for SIMD operations, making full use of these high-speed registers for data processing can improve efficiency. This application uses a variety of MSA instructions to process 16 groups of data in parallel at the same time, which will bring up to 16 times the throughput improvement compared with the traditional serial method. It can process multiple code blocks at the same time and improve the decoding throughput. This application can achieve performance improvements similar to the SSE (a 128-bit bit-width vector extension instruction set based on the X86 processor of Intel) instruction set under the X86 architecture, or the NEON (a 128-bit bit-width vector extension instruction set based on the ARM processor) instruction set used in the ARM architecture. The LDPC decoding throughput reaches more than 100Mbps, and the delay is controlled below 3ms, thereby greatly improving the key indicators such as the rate of the overall communication system.

[0097] The throughput index of this application under the 5G-NR standard and the improvement of throughput compared with the traditional serial method are as follows: Figure 7 As shown in Table 1, Figure 7 It can be seen that under different signal-to-noise ratio conditions, the present application has a 12-16 times throughput improvement. Experiments have shown that when the throughput is less than 2dB, the early stopping mechanism should not be used. When the bit error rate is greater than 2dB, the use of the early stopping scheme can reduce the number of iterations while ensuring the bit error rate requirements, thereby improving the throughput.

[0098] Table 1

[0099]

[0100] The decoding throughput of this application under different code lengths and the improvement of the throughput compared with the traditional serial method are shown in Table 2 and Figure 8 shown.

[0101] Table 2

[0102]

[0103] The LDPC decoding rates under different communication standards and the improvement in throughput compared with the traditional serial method are shown in Table 3.

[0104] Table 3

[0105]

[0106]

[0107] This application uses core binding technology to significantly reduce operating system interrupts, and the LDPC decoding delay is controlled within 2ms, which is at an excellent level compared with similar designs.

[0108] This application uses SIMD technology to perform parallel processing on the input codewords. Compared with the serial method, it can obtain 16 times the parallelism and increase the throughput by 9-16 times, effectively improving the working performance of the LDPC decoder. It also has relatively stable parallel acceleration performance for data to be decoded of different standards, different code types, different code lengths, and different code rates, and has a wide range of application scenarios.

[0109] For scenarios with different bit error rates, this application flexibly adopts an early stop scheme to reduce the number of iterations when the signal-to-noise ratio is high and the channel conditions are good, thereby further improving the throughput. At the same time, in order to maintain the bit error rate performance when the signal-to-noise ratio is low, the early stop mechanism is not used.

[0110] This application adopts core binding technology to reduce the adverse effects of interrupts generated by the operating system on decoding, significantly reduce latency, and control the latency to within 2ms, meeting the real-time requirements of the communication system.

[0111] In terms of hardware, this application uses the "Loongson 3A4000" model CPU. Loongson is the only domestic CPU company that builds an independent "Intel + Microsoft" system based on an independent instruction set system, which is completely safe, independent and controllable. In terms of software, the domestic Tongxin Kylin V10 system is used. The basic software and hardware are all localized, and the MSA instruction set under the MIPS architecture is used for parallel acceleration at the instruction set level, filling the gap in this direction.

[0112] Based on the same inventive concept, the embodiment of the present application also provides an LDPC decoding device based on the MIPS architecture for implementing the LDPC decoding method based on the MIPS architecture involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more LDPC decoding device embodiments based on the MIPS architecture provided below can refer to the limitations of the LDPC decoding method based on the MIPS architecture above, and will not be repeated here.

[0113] In an exemplary embodiment, a LDPC decoding device based on a MIPS architecture is provided, comprising:

[0114] A core binding module, used to bind the LDPC decoding task to a CPU core other than the operating system core of a general-purpose CPU based on the MIPS architecture by using the core binding technology;

[0115] The data preprocessing module is used to perform fixed-point processing on the log-likelihood ratio (LLR) data to be decoded, and to group the processed data;

[0116] A data grouping module, used for segmenting the grouped data;

[0117] LDPC decoding module, used to use the MSA instruction set of MIPS architecture to perform LDPC decoding iteration operation on each group of data after data segmentation;

[0118] The data reorganization module is used to reorganize the decoding results of all the data after data segmentation to obtain the decoding results.

[0119] In an exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (I / O for short) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a LDPC decoding method based on the MIPS architecture is implemented.

[0120] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0121] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0122] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0123] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0124] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0125] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A LDPC decoding method based on MIPS architecture, characterized in that: The LDPC decoding method based on the MIPS architecture includes: Utilize core binding technology to bind the LDPC decoding task to the CPU core other than the operating system core of a general-purpose CPU based on the MIPS architecture; Performing fixed-point processing on the log-likelihood ratio (LLR) data to be decoded; and grouping the processed data; Segment the grouped data; The MSA instruction set of the MIPS architecture is used to perform LDPC decoding iteration operations on each group of data after data segmentation. The decoding results of the data after all data segmentation are reorganized to obtain the decoding results.

2. The LDPC decoding method based on the MIPS architecture according to claim 1, characterized in that: The algorithms for the iterative operation of LDPC decoding include: a normalized minimum sum decoding algorithm or a shifted minimum sum algorithm.

3. The LDPC decoding method based on the MIPS architecture according to claim 1, characterized in that: The MSA instruction set of the MIPS architecture is used to simultaneously perform LDPC decoding iteration operations on each group of data after data segmentation, specifically including: The subtraction instruction in the MSA instruction set is used to determine the value passed from the message node to the check node in the current iteration, and the maximum value instruction and the minimum value instruction are used to perform the limiting operation; The XOR instruction, absolute value instruction, subtraction instruction and data mixing instruction in the MSA instruction set are used to determine the value passed from the check node to the message node in the current iteration; Determine the absolute value of the value passed from the message node to the check node and the absolute value of the value passed from the check node to the message node using the absolute value instruction in the MSA instruction set; Use the maximum value instruction and minimum value instruction in the MSA instruction set to limit the absolute value; The absolute value of the value passed from the message node to the check node and the absolute value of the value passed from the check node to the message node are used to determine the value of the message node in the current iteration number by using the bitwise AND instruction in the MSA instruction set; and the maximum value instruction and the minimum value instruction are used to perform amplitude control according to the minimum sum algorithm principle; Determine whether the current number of iterations is less than the maximum number of iterations; if so, add 1 to the number of decoding iterations and iterate again; if it is the maximum number of iterations, use the comparison instructions in the MSA instruction set to perform a hard decision operation and output the decoding result of the data after data segmentation.

4. The LDPC decoding method based on the MIPS architecture according to claim 3, characterized in that: Determine whether the current number of iterations is less than the maximum number of iterations, which also includes: An early stopping mechanism is adopted; the early stopping mechanism performs hard decision decoding according to the value of the message node updated according to the current iteration number.

5. The LDPC decoding method based on the MIPS architecture according to claim 1, characterized in that: The decoding results of all data after data segmentation are reorganized to obtain the decoding results, which specifically include: Use the data movement instructions and data mixing instructions in the MSA instruction set of the MIPS architecture to reorganize data; The data is stored using the data load instruction in the MSA instruction set of the MIPS architecture.

6. An LDPC decoding device based on MIPS architecture, characterized in that: The LDPC decoding device based on the MIPS architecture includes: A core binding module is used to bind the LDPC decoding task to a CPU core other than the operating system core of a general-purpose CPU based on the MIPS architecture using core binding technology; The data preprocessing module is used to perform fixed-point processing on the log-likelihood ratio (LLR) data to be decoded and group the processed data; A data grouping module is used to segment the grouped data; LDPC decoding module, used to use the MSA instruction set of the MIPS architecture to perform LDPC decoding iteration operations on each group of data after data segmentation; The data reassembly module is used to reassemble the decoding results of all data after data segmentation to obtain the decoding results.

7. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the LDPC decoding method based on the MIPS architecture according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the LDPC decoding method based on the MIPS architecture according to any one of claims 1 to 5 is implemented.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the LDPC decoding method based on the MIPS architecture according to any one of claims 1 to 5 is implemented.