A method and apparatus for interleaved scheduling and hierarchical decoding of LDPC frames based on mutual information

Abstract

This invention discloses a layered decoding method and apparatus for LDPC frames based on mutual information, relating to the field of digital signal transmission in the physical layer of wireless communication. The method includes the following steps: 1) Calculating the LLR value of a given codeword to be decoded, and performing Q-ratio specific-point quantization to maximize mutual information on the entire frame's LLR value; 2) Calculating the PP value of the Q-ratio specific-point quantized LLR value according to Bayes' formula; 3) Layering the PP values ​​of the LLR values ​​of the data to be decoded according to the coefficients of the parity check matrix H; 4) Calculating the information within the information node; 5) Performing parity check node update calculations according to the PMS decoding algorithm to obtain the information value outside the parity check node; 6) Updating the information node using the information value outside the parity check node; 7) Symbol decoding; 8) Calculating the syndrome according to H. If the result of the syndrome is 0 or the maximum number of iterations is reached, outputting the three-frame codeword result; otherwise, returning to step 3) for the next decoding iteration. This invention can effectively improve the decoding processing speed of each frame.

Description

Technical Field

[0001] This invention relates to a layered decoding method and apparatus for LDPC frame interleaving scheduling based on mutual information in the field of digital signal transmission for physical layer communication. More specifically, it is a channel decoding method that employs frame interleaving scheduling, performs fixed-point quantization processing to maximize mutual information of information awaiting decoding, performs probability-based minimum summation of the second smallest value during check node updates in decoding to reduce decoding delay, improve decoding speed, and optimize space resource usage. Background Technology

[0002] In modern wireless communication systems, to bridge the gap between 5G and 6G and further increase communication speeds—aiming for another order-of-magnitude improvement—high-speed and ultra-high-speed communication are currently key research directions. In digital communication, to achieve accurate, fast, and reliable information transmission, it is necessary to restore digital signals that have passed through noisy channels, enabling the receiver to quickly and accurately receive information from the sender. Channel coding and decoding technology provides the technical support for ensuring reliable transmission.

[0003] In the past two decades of development in the field of wireless communication, channel coding and decoding technology has played a crucial role. Many excellent coding methods have emerged, such as LDPC coding, Turbo coding, and Polar coding, which have achieved error correction capabilities approaching the Shannon limit and all possess mature hardware implementations. Due to its simple and unique structure and the ease of implementation of its decoding process in practical hardware, LDPC has long been a key research target for researchers in the field of channel coding. Recent research has seen LDPC decoder designs achieve throughputs of tens of Gb / s. However, building upon current research, designing a low-power, low-resource-consumption LDPC decoding algorithm capable of reaching tens or hundreds of Gb / s remains a significant technical challenge, as it maintains good bit error rate performance and flexibility. Summary of the Invention

[0004] This invention is used to improve decoding throughput, reduce hardware resource consumption, and enhance LDPC error correction capabilities when using LDPC encoding in ultra-high-speed communication.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A hierarchical decoding method for LDPC frames based on mutual information, targeting a number of information nodes N. b The number of verification nodes is N c The LDPC code specifically includes the following steps:

[0007] (1) Calculate the LLR value for each frame of data to be decoded, and perform Q-ratio specific point quantization to maximize mutual information on the LLR value of the whole frame.

[0008] (2) Calculate the posterior probability of the LLR value after Q-ratio quantization at a specific point according to Bayes' formula;

[0009] (3) The posterior probabilities of the LLR values ​​of the multi-frame data to be decoded after fixed-point quantization are arranged in a hierarchical manner according to the coefficients of the parity check matrix H.

[0010] (4) Calculate N based on the posterior probability of the LLR value and the extrinsic information of the multi-frame data, combined with the index position after hierarchical arrangement. b The information within each information node is the posterior probability minus the extrinsic information of the verification node; where the extrinsic information is 0 in the initial iteration.

[0011] (5) Using N b The information within each information node is updated and calculated using the PMS decoding algorithm to obtain the information value outside the verification node.

[0012] (6) Calculate and update the information node information using the information node update formula in the minimum sum decoding algorithm based on the external information of the verification node;

[0013] (7) Decode the updated data using symbols and make a hard decision to obtain the decoded codeword result;

[0014] (8) Calculate the syndrome based on the parity check matrix H and the codeword result. If the result of the syndrome is 0 or the maximum number of decoding iterations is reached, the process ends. Otherwise, return to step (3) and proceed to the next decoding iteration.

[0015] Furthermore, step (1) specifically involves:

[0016] To process multi-frame data awaiting decoding using a frame-interleaved scheduling method, multiple channels are first prepared. The LLR (Limited Range) values ​​of the multi-frame data awaiting decoding are calculated using the log-likelihood ratio (LRR) formula, and then Q-bit fixed-point quantization is performed on the LLR values ​​according to the mutual information maximization rule, where the most significant bit of the Q-bit is the sign bit. Next, the mutual information between the LLR value of each frame of data awaiting decoding and the fixed-point quantized data sequence is calculated. The quantized value corresponding to the maximum mutual information is obtained as the quantization result. The formula for calculating the mutual information between two sequences X and Y is shown below:

[0017]

[0018] Further, step (3) specifically involves: arranging the posterior probabilities of the LLR values ​​of the multi-frame data to be decoded after fixed-point quantization in a hierarchical manner according to the parallelism Z of the check nodes in the check matrix H. Each frame of data is divided into Z layers, and the arrangement method is determined according to the index value of the check matrix H.

[0019] Furthermore, step (5) specifically involves: using the PMS decoding algorithm to calculate the N required for updating the verification node. b The maximum and second largest values ​​of information values ​​within each information node are determined, and then the information outside the check node is updated using the check node update formula of the minimum sum decoding algorithm.

[0020] Furthermore, the maximum and second largest values ​​are calculated as follows: N... b The information value within each information node is divided into k equal parts. The maximum value of each part is calculated using a binary search method in parallel, denoted as Max1, Max2, ..., Maxk. The maximum value Max of the entire information node is the maximum value among Max1, Max2, ..., Maxk. The second largest value among Max1, Max2, ..., Maxk is selected as N. b The second largest value Smax of the information value within each information node; where k is a set value.

[0021] Further, step (8) specifically involves: performing symbol decoding based on the information node update value, obtaining the decoding result codewords Frame1, Frame2, ..., FrameN through hard decision, multiplying each frame codeword with the parity check matrix H to obtain the result of the parity check syndrome. If the syndrome result is zero, the decoding process ends; otherwise, return to step (3) for the next iteration, incrementing the iteration number by 1; where N is the number of frames of data to be decoded.

[0022] Furthermore, an LDPC frame interleaving scheduling layered decoding device suitable for millimeter-wave ultra-high-speed communication uses the above-mentioned method for high-speed LDPC decoding.

[0023] The present invention has the following advantages due to the adoption of the above technical solutions:

[0024] 1. This invention adopts a frame interleaved scheduling operation mode, which processes multiple frames of data to be decoded in parallel through interleaved scheduling. Each frame of data to be decoded can be layered according to the characteristics of the parity check matrix, which can effectively improve the decoding speed of each frame. Moreover, the layering can maximize the parallel processing capability.

[0025] 2. In the decoding algorithm process adopted in this invention, fixed-point quantization operation based on maximizing mutual information is used for the information to be decoded. This not only improves decoding efficiency and reduces the number of decoding iterations, thereby reducing decoding latency, but also saves storage space due to the use of fixed-point quantization operation that maximizes mutual information, enabling LDPC to be implemented in a smaller hardware resource space. When updating the check node, the PMS algorithm is used to calculate the maximum and second maximum values, which can further improve the update speed of the check node information, thereby improving the decoder throughput. Attached Figure Description

[0026] Figure 1 This is a flowchart of the present invention.

[0027] Figure 2 This is a diagram of the layered decoding structure of the present invention.

[0028] Figure 3 This is a schematic diagram of frame interleaving scheduling when N=3 according to the present invention.

[0029] Figure 4 This is the coding gain diagram of the present invention. Detailed Implementation

[0030] To clearly illustrate the technical features of this solution, specific implementation methods are described below, along with their appendices. Figure 1-4 The present invention will be described in detail below. Specific examples are provided below to implement the present invention.

[0031] A layered decoding method for LDPC frames based on mutual information, such as... Figure 1 As shown, it includes the following steps:

[0032] (1) Calculate the LLR (Log-Likelihood Ratio) value for the codewords to be decoded in multiple frames of a given Gaussian channel to obtain the LLR value of each frame of data to be decoded. Without loss of generality, the number of frames is set to 3 in this example, and the Q ratio specific point quantization processing to maximize mutual information is performed on the LLR value of the whole frame.

[0033] This example uses 860_1032LDPC encoding under the 5G-NR standard as an example to illustrate the present invention. Only one frame of data to be decoded is shown as an example here. The actual interleaved scheduling and parallel processing of three frames of data will be carried out according to the appendix. Figure 3 As shown, prepare multiple pipelines; the number of pipelines does not necessarily equal the number of data frames. Use a frame interleaving scheduling method to process multiple frames of data waiting to be decoded. Obtain a frame of information to be decoded with a length of 1032 from the Gaussian noise channel. Calculate the LLR (Log-Likelihood Ratio) value according to the formula below:

[0034]

[0035] Wherein, p(r|b i =0) indicates that when the sequence of information to be decoded obtained from the Gaussian noise channel is r, the i-th bit b i The probability of LLR is 1. After obtaining the LLR value, Q-ratio fixed-point quantization is performed on the LLR value according to the criterion of maximizing mutual information. The mutual information between the LLR value of each frame of data to be decoded and the data sequence after fixed-point quantization is calculated, and the quantization corresponding value when the mutual information is maximized is obtained as the quantization result; the mutual information of two sequences X and Y is calculated as follows:

[0036]

[0037] In this example, Q is 5, and the highest bit is the sign bit.

[0038] (2) Calculate the posterior probability PP of the LLR value after Q-ratio quantization at a specific point obtained in step (1) according to the Bayes formula;

[0039] The Bayes formula for calculating the posterior probability is shown below:

[0040] P(Y|X)=P(X|Y)*P(Y) / P(X)

[0041] (3) Based on the parallelism Z of the parity nodes of the parity-check matrix H encoded by LDPC860_1020, the posterior probabilities PP of the three frames of data awaiting decoding obtained in step (2) are arranged in layers. Each frame of data is divided into Z layers, and the arrangement is determined according to the index value of the parity-check matrix H. For the specific layered parallel structure diagram of the three frames of data, please refer to the appendix of this book. Figure 3 .

[0042] (4) Calculate N based on the posterior probability of the LLR value and the extrinsic information of the three frames of data (initialized to 0 in the first iteration; the maximum number of iterations of the decoding algorithm described in this invention is set to 15), and in conjunction with the index positions after the three frames of data are arranged in layers in step (3). b The information within each information node is the posterior probability PP minus the extrinsic information of the verification node. Since the extrinsic information of the verification node is set to 0 in the first iteration, N is... b The information within each information node is the posterior probability PP of the LLR value.

[0043] (5) Using N b The information within each information node is updated using the PMS (Probability Min Sum) decoding algorithm to obtain the information value outside the verification node.

[0044] The PMS algorithm is an extension of the minimum sum decoding algorithm. The probability of obtaining N b The second largest information value within each information node. Now let N... b The information value within each information node is divided into k equal parts. If data cannot be divided into k equal parts in practice, it can be padded with zeros, as this invention calculates the maximum and second-largest values. The maximum value of each part is calculated in parallel using a binary search method, denoted as Max1, Max2, ..., Maxk. The maximum value Max of the entire information node is then the maximum value among Max1, Max2, ..., Maxk, i.e., Max = max(Max1, Max2, ..., Maxk), where max is a maximum value function. The second-largest value among Max1, Max2, ..., Maxk is selected as N. b The second largest value Smax of the information value within each information node can be achieved in FPGA implementation using multiple 4-to-2 selectors. The external information value calculated in this step is stored for two purposes: firstly, as external information for multiple frames (Frame1, Frame2, ..., FrameN) in the next iteration; and secondly, to calculate the updated information node value used for decision-making, i.e., calculating the information within the information node minus the updated external information. The specific data flow in this process can be found in the appendix of this invention. Figure 2 .

[0045] When the PMS decoding algorithm of this invention is implemented in hardware, the hardware memory resource occupation during the verification node update calculation process can be reduced, and the calculation speed of the verification node process can be accelerated. This is because the data traversal length is reduced by nearly half, almost completely eliminating the time for traversing to find the second smallest value, thereby greatly reducing the clock delay of the entire decoding process.

[0046] (6) Update the information nodes using the maximum value Max and the second maximum value Smax obtained by the PMS algorithm in step (5), and store them;

[0047] Based on the external information of the verification node obtained in step (5), the information node information is calculated and updated using the information node update formula in the minimum sum decoding algorithm. The update formula is as follows:

[0048]

[0049] in, Let M(i) represent the update result of the i-th information node in the k-th iteration, and M(i) represent the set of verification nodes connected to the i-th information node. This represents the external information of the verification node connected to the information node during the (k-1)th iteration, which is the result obtained in step (5); This represents the initial value of the i-th information node during the first iteration.

[0050] (7) The updated data is symbolically decoded and hard-determined to obtain the three-frame codeword result.

[0051] The present invention adopts a symbol decoding algorithm, performs hard decision on the information node information of the 3 frames of codewords obtained in step (6), and takes the highest bit in the Q-bit result as the final decoding result, denoted as y.

[0052] (8) Calculate the syndrome based on the parity-check matrix H and the codeword results. The syndrome is calculated as follows:

[0053] y*H

[0054] If the result of the synaptic expression is 0 or the maximum number of iterations for decoding is reached (the maximum number of iterations in this invention is set to 15), then output three frames of codeword results; otherwise, return to step (3) and perform the next iteration, incrementing the iteration count by 1.

[0055] An LDPC frame interleaving scheduling layered decoding device suitable for millimeter-wave ultra-high-speed communication uses the above-mentioned method for high-speed LDPC decoding.

[0056] The decoding method of this invention achieves excellent results in terms of decoding speed and hardware resource consumption during the decoding process, without causing excessive loss in coding gain. Figure 4 This is the coding gain map obtained using the present invention.

Claims

1. A hierarchical decoding method for LDPC frames based on mutual information, characterized in that, For an information node number of N b The number of verification nodes is N c The LDPC code specifically includes the following steps: (1) Calculate the LLR value for each frame of data to be decoded, and perform Q-ratio specific point quantization to maximize mutual information on the LLR value of the whole frame. (2) Calculate the posterior probability of the LLR value after Q-ratio quantization at a specific point according to Bayes' formula; (3) The posterior probabilities of the LLR values ​​of the multi-frame data to be decoded after fixed-point quantization are arranged in a hierarchical manner according to the coefficients of the parity check matrix H; (4) Calculate N based on the posterior probability of the LLR value and the extrinsic information of the multi-frame data, combined with the index position after hierarchical arrangement. b The information within each information node is the posterior probability minus the extrinsic information of the verification node; where the extrinsic information is 0 in the initial iteration. (5) Using N b The information within each information node is updated and calculated using the PMS decoding algorithm to obtain the information value outside the verification node. (6) Calculate and update the information node information using the information node update formula in the minimum sum decoding algorithm based on the external information of the verification node; (7) Decode the updated data symbolically and make a hard decision to obtain the decoded codeword result; (8) Calculate the syndrome based on the parity check matrix H and the codeword result. If the result of the syndrome is 0 or the maximum number of decoding iterations is reached, the process ends; otherwise, return to step (3) and proceed to the next decoding iteration. Specifically, step (5) involves using the PMS decoding algorithm to calculate the N required for updating the verification node. b The maximum and second largest values ​​of information values ​​within each information node are determined, and then the information outside the check node is updated using the check node update formula of the minimum sum decoding algorithm. The maximum and second largest values ​​are calculated as follows: N b Information values ​​within each information node are divided into... k For each part, the maximum value is calculated in parallel using the binary search method, denoted as Max1, Max2, ..., Max... k Then the maximum value of the information value within the entire information node is Max1, Max2, ..., Max2. k The maximum value among Max1, Max2, ..., Max2 is selected. k The second largest value in is N b The second largest value Smax of the information value within each information node; where k This is the set value.

2. The LDPC frame interleaving scheduling layered decoding method based on mutual information according to claim 1, characterized in that, Step (1) is as follows: To process multi-frame data awaiting decoding using a frame-interleaved scheduling method, multiple channels are first prepared. The LLR (Limited Range) values ​​of the multi-frame data awaiting decoding are calculated using the log-likelihood ratio (LRR) formula, and then Q-bit fixed-point quantization is performed on the LLR values ​​according to the mutual information maximization rule, where the most significant bit of the Q-bit is the sign bit. Next, the mutual information between the LLR value of each frame of data awaiting decoding and the fixed-point quantized data sequence is calculated. The quantized value corresponding to the maximum mutual information is obtained as the quantization result. The formula for calculating the mutual information between two sequences X and Y is shown below: 。 3. The LDPC frame interleaving scheduling layered decoding method based on mutual information according to claim 1, characterized in that, Step (3) specifically involves: arranging the posterior probabilities of the LLR values ​​of the multi-frame data to be decoded after fixed-point quantization in layers according to the parallelism Z of the check nodes in the check matrix H. Each frame of data is divided into Z layers, and the arrangement method is determined according to the index value of the check matrix H.

4. The LDPC frame interleaving scheduling layered decoding method based on mutual information according to claim 1, characterized in that, Step (8) is as follows: perform symbol decoding based on the information node update value, and obtain the decoding result codewords Frame1 codeword, Frame2 codeword, ..., FrameN codeword by hard decision. Multiply each frame codeword with the parity check matrix H to obtain the result of the parity check syndrome. If the result of the syndrome is zero or the maximum number of decoding iterations is reached, the decoding process ends. Otherwise, return to step (3) and perform the next iteration, incrementing the iteration count by 1. Where N is the number of frames of data to be decoded.

Images