Reed-Muller decoder based on fast iterative projection aggregation decoding algorithm
Through fast iterative projection aggregation decoding algorithm and hardware optimization, the problem of delay and complexity of iterative projection aggregation decoding algorithm in uRLLC scenarios is solved, and the Reed-Muller translation with low latency, high throughput and high reliability is achieved Codec, meets the needs of 5G and 6G communications.
Patent Information
- Application Number
- CN202411196905.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-29
AI Technical Summary
The existing iterative projection aggregation decoding algorithm cannot meet the requirements of low latency and high reliability in uRLLC scenarios, and the hardware implementation is complex, making it difficult to meet the needs of 5G and 6G communications.
Using Reed-Muller decoder based on fast iterative projection aggregation decoding algorithm, the decoding process is simplified, the number of iterations is reduced and the hardware structure is optimized to achieve lower latency and High throughput.
It achieves lower latency, higher throughput and higher area efficiency, meets the low latency and high reliability requirements of uRLLC, simplifies the decoding process and reduces hardware complexity.
Smart Images

Figure CN118713684B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of communication technology, and in particular to a Reed-Muller decoder based on a fast iterative projection aggregation decoding algorithm. Background Art
[0002] The main application scenarios of 5G include: massive machine type communication (mMTC), enhanced mobile broadband (eMBB), and ultra-reliable and low latency communication (uRLLC). uRLLC is a key concept in 5G and even 6G and IoT technologies, mainly targeting application scenarios with extremely high latency requirements and high data transmission reliability requirements, such as industrial automation, self-driving cars, smart grids, smart transportation / freight, and augmented / virtual reality, etc. By providing reliable and low-latency communication services, uRLLC can support key applications such as real-time data transmission, remote control and collaborative operations, and promote the development of the IoT and the expansion of application scenarios.
[0003] RM (Reed-Muller) code is a binary linear block code with excellent decoding performance. Its encoding and decoding algorithms have low complexity and are easy to implement through hardware circuits. In addition, by changing the order r and the code length parameter m, RM codes can form subclasses with rich structures to adapt to different channels. Since RM codes have the advantages of high reliability, low complexity and suitability for high parallelism, they show great application prospects in uRLLC scenarios.
[0004] The recursive projection aggregation (RPA) decoding algorithm achieves excellent performance on low-code rate (such as second-order and third-order) RM codes. Compared with the polar code (Polar code) using the successive cancellation list (SCL) decoding method, the RM code based on RPA decoding has better decoding performance. In addition, the polar code based on the SCL algorithm has the disadvantages of not being highly parallel and complex hardware implementation. The RPA algorithm is not conducive to hardware implementation due to its unique recursive structure. In order to solve this problem, the prior art proposes an iterative projection aggregation (IPA) decoding algorithm. Since the recursive structure of the RPA algorithm is removed, the algorithm is easier to implement in hardware. However, the IPA algorithm does not reduce the components required in the projection stage compared to the RPA algorithm, resulting in the overall decoding process still taking too long and failing to meet the low latency requirements proposed by uRLLC. Summary of the invention
[0005] In order to solve the above technical problems, the purpose of the present invention is to provide a Reed-Muller decoder based on a fast iterative projection aggregation decoding algorithm, which can achieve lower delay, higher throughput and higher area efficiency, and meet the requirements of low delay and high reliability proposed by uRLLC.
[0006] The technical solution adopted by the present invention is: a Reed-Muller decoder based on a fast iterative projection aggregation decoding algorithm, comprising the following steps:
[0007] Performing fast projection on the received codeword to obtain the first-order projection component of the received codeword;
[0008] Decoding the first-order projection component of the received codeword based on the fast Hadamard transform to obtain a first-order decoding result;
[0009] Performing a hard decision on a first-order projection component of the received codeword, and comparing the hard decision result with the first-order decoding result based on the first-order projection component to obtain a comparison result;
[0010] Aggregating the received codeword based on the comparison result and the first-order decoding result to obtain a decoded codeword;
[0011] The decoded codeword is used as a new round of received codeword for iterative decoding until a condition for terminating iterative decoding is met to obtain a final decoded codeword.
[0012] Further, the step of performing fast projection on the received codeword to obtain a first-order projection component of the received codeword specifically includes:
[0013] Preliminarily pruning the projection components of the received codeword to determine the number of retained first-order projection components;
[0014] Performing index confirmation based on the number of the first-order projection components and storing the index in a first index variable according to the order;
[0015] Calculating an index pair based on the first index variable, and storing the index values in increasing order in a second index variable based on the index pair;
[0016] Rearranging the high-order second index variables based on the low-order second index variables to obtain the associated projection arrangement of each order of the projection components;
[0017] Based on the associated projection arrangement of each order of the projection components, the adjacent log-likelihood ratios (LLR) of each order of the received codeword are projected into a single value to obtain the first-order projection component of the received codeword.
[0018] Further, the step of pre-pruning the projection components of the received codeword to determine the number of retained first-order projection components specifically includes:
[0019] A fixed number of projection components are selected based on the code length of the received code word as each order projection of the received code word;
[0020] After projecting the received codewords step by step based on a fixed number of projection components, the number of retained first-order projection components is obtained.
[0021] Further, the step of decoding the first-order projection component of the received codeword based on the fast Hadamard transform to obtain a first-order decoding result specifically includes:
[0022] Iteratively calculate the first-order projection component of the received codeword based on Hadamard matrix multiplication to obtain a decoding vector;
[0023] Find the position index of the maximum absolute value of the decoded vector, and save the position index with the rightmost bit as the most significant bit;
[0024] Calculate a binary index of a first-order projection component of the received codeword based on the position index;
[0025] The decoding codeword is calculated based on the binary index and the first-order generation matrix of the received codeword to obtain a first-order decoding result.
[0026] Further, the step of aggregating the received codeword based on the comparison result and the first-order decoding result to obtain the decoded codeword specifically includes:
[0027] When the number of iterative decoding is less than the preset minimum number of iterations, the received codewords are flipped according to the first-order decoding results, and the average of the flipped received codewords is taken to obtain the decoded codeword through hard decision;
[0028] When the number of iterative decoding is greater than the preset minimum number of iterations, if the hard decision result is the same as the first-order decoding result, the aggregation step of the current round is skipped, and the iterative decoding is terminated, and the hard decision result of the current round is used as the final decoded codeword;
[0029] When the number of iterative decoding is greater than the preset minimum number of iterations, if the hard decision result is different from the first-order decoding result, the received codeword is flipped according to the first-order decoding result, and the average of the flipped received codeword is taken to obtain the decoded codeword through hard decision.
[0030] Furthermore, the conditions for terminating iterative decoding include that the number of iterative decoding times is greater than a preset minimum number of iterations and the sign bits of the hard decision result and the first-order decoding result are the same, and the number of iterative decoding times reaches a preset maximum number of iterations.
[0031] Furthermore, the flip decision specifically includes:
[0032] If the index value of the first-order decoding result is 0, the log-likelihood ratio value corresponding to the index in the received codeword remains unchanged;
[0033] If the index value of the first-order decoding result is 1, the log-likelihood ratio value corresponding to the index in the received codeword is negated.
[0034] Furthermore, the hardware structure of the Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm includes a fast projection module, a first-order decoding module, a comparison module and an aggregation module, wherein:
[0035] The first-order decoding module includes a Hadamard matrix multiplication module, a maximum value finding module and a generation matrix multiplication module;
[0036] The fast projection module is used to perform fast projection on the received codeword to obtain the first-order projection component of the received codeword;
[0037] The first-order decoding module decodes the first-order projection component of the received codeword based on the fast Hadamard transform to obtain a first-order decoding result;
[0038] The Hadamard matrix multiplication module iteratively calculates the first-order projection component of the received codeword based on the Hadamard matrix multiplication to obtain a decoding vector;
[0039] The maximum value finding module is used to find the position index of the maximum absolute value of the decoded vector, and calculate the binary index of the first-order projection component of the received codeword based on the position index;
[0040] The generator matrix multiplication module calculates the decoding codeword based on the binary index and the first-order generator matrix of the received codeword to obtain a first-order decoding result;
[0041] The comparison module is used to perform hard decision on the first-order projection component of the received codeword, and compare the hard decision result with the first-order decoding result based on the first-order projection component to obtain a comparison result;
[0042] The aggregation module aggregates the received codewords based on the comparison result and the first-order decoding result to obtain a decoded codeword.
[0043] The beneficial effects of the present invention are as follows: the present invention proposes an aggregation decoding algorithm based on fast iterative projection, adopts a pruning method of selecting only a fixed number of projection components for each order projection, obtains part of the first-order projection components of the received codeword after order-by-order projection, wherein the number of projection components selected at each order is related to the code length of the received codeword, and the effect of significantly reducing the number of first-order projection components is obtained; a novel early stopping strategy is proposed, after a preset threshold number of iterative decoding, by comparing the first-order decoding result with the hard decision result, if all are the same, the aggregation step can be skipped, and the output codeword is directly obtained from the hard decision of the input codeword, thereby obtaining the effect of simplifying the decoding process; in addition, based on the proposed The decoding method designs a Reed-Muller decoder hardware architecture based on a fast iterative projection aggregation decoding algorithm. By calculating the log-likelihood ratio index associated with the first-order projection in advance, the high-order codeword is directly projected to the first order, the iterative projection step is omitted, and the decoding process is simplified. The decoder using the method of the present invention has a reduced number of first-order code decoding times, lower complexity and fewer projection components than the original iterative projection aggregation decoding method, which can be reflected in the hardware performance. Lower latency can be achieved, and the performance loss is almost negligible, which can achieve lower latency, higher throughput and higher area efficiency, and meet the low latency and high reliability requirements proposed by uRLLC. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 It is a flow chart of steps of a Reed-Muller decoder based on a fast iterative projection aggregation decoding algorithm of the present invention;
[0045] Figure 2 It is a structural block diagram of a Reed-Muller decoder based on a fast iterative projection aggregation decoding algorithm of the present invention;
[0046] Figure 3 It is a hardware architecture diagram of the Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm of the present invention;
[0047] Figure 4It is a fast projection module architecture diagram of the Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm of the present invention;
[0048] Figure 5 It is a Hadamard matrix multiplication module architecture diagram of the Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm of the present invention;
[0049] Figure 6 It is a maximum value finding module architecture diagram of the Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm of the present invention;
[0050] Figure 7 It is a schematic diagram of a generator matrix multiplication module architecture of a Reed-Muller decoder based on a fast iterative projection aggregation decoding algorithm of the present invention;
[0051] Figure 8 It is a comparison module architecture diagram of the Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm of the present invention;
[0052] Fig. 9 It is a simulation diagram of the RM (7, 2) and RM (6, 3) frame error rate performance of the Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm of the present invention in an additive white Gaussian noise channel;
[0053] Fig.10 It is the RM(7,2) timing diagram of the Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm of the present invention with the overall decoder parallelism being 4. DETAILED DESCRIPTION
[0054] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only provided for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.
[0055] Reference Figure 1 The present invention provides a Reed-Muller decoder based on a fast iterative projection aggregation decoding algorithm, comprising the following steps:
[0056] S1. Perform fast projection on the received codeword to obtain the first-order projection component of the received codeword;
[0057] Specifically, the type of codeword received in the specific embodiment of the present invention is RM code, which is usually expressed as ,in represents the total number of orders, Indicates the code length, Representation and code length Corresponding RM code parameters, for The received codeword , after projection using the fast projection method of the present invention, we get of The first-order projection component of , that is, in Each level The log-likelihood ratios are projected into a single log-likelihood ratio of order 1.
[0058] S1.1, pre-pruning the projection components of the received codeword to determine the number of retained first-order projection components;
[0059] S1.1.1, selecting a fixed number of projection components as each order projection of the received codeword based on the code length of the received codeword;
[0060] Specifically, the code length of the received codeword is , the specific embodiment of the present invention is selected The projection components are taken as each order projection of the received codeword.
[0061] S1.1.2. After projecting the received codeword step by step based on a fixed number of projection components, the number of retained first-order projection components is obtained.
[0062] Specifically, according to After projecting each order of the received codeword one by one, the number of first-order projection components finally retained is indivual.
[0063] S1.2, performing index confirmation based on the number of the first-order projection components, and storing the index in a first index variable according to the order;
[0064] Specifically, since this projection method only selects projection codewords, after iterative projection, we can get codewords, so the index of the projection Set as , let the current order be , using the first index variable To store Projection corresponds to arrive The index of each order of is calculated as follows:
[0065] when hour,
[0066]
[0067] otherwise,
[0068]
[0069] in, Represents the first index variable; Indicates the current order; Indicates projections; Indicates the length of the received codeword.
[0070] S1.3, calculating an index pair based on the first index variable, and storing the index value in an increasing order in a second index variable based on the index pair;
[0071] Specifically, for The associated projection of the order projection ( rank) contains indices from 0 to The log-likelihood ratio of , and based on the index pair, store the index value in ascending order in the second index variable In , the calculation expression of the index pair is as follows:
[0072]
[0073] in, Represents the second index variable; Represents associated projection The index of Represents associated projection The second index of Indicates the total number of orders.
[0074] S1.4, rearranging the high-order second index variables based on the low-order second index variables to obtain the associated projection arrangement of each order of the projection components;
[0075] Specifically, based on the low-order second index variable For higher-order second index variables Rearrange to obtain a higher-order correlation projection arrangement , order Increment from 1 to , which is expressed as follows:
[0076]
[0077] in, It is calculated by (6) The index value of It is calculated by (7) The index value of Indicates the index, ranging from 0 to ; Indicates The index of the second index variable of order is The value of Indicates that the index of the second index variable before modification is The value of Represents a higher-order arrangement of associated projections.
[0078] S1.5. Based on the associated projection arrangement of each order of the projection components, adjacent log-likelihood ratio values of each order of the received codeword are projected into a single value to obtain a first-order projection component of the received codeword.
[0079] Specifically, through step S1.3, we get The correlation projection of order ( The original received codeword of ) The ordering method of projections is to put adjacent The log-likelihood ratio values are a set of projections into a value, and the expression of the projection rule is as follows:
[0080]
[0081] in, Indicates projections; After rearrangement, Order of arrangement; Represents the input vector The index is The value of Represents an index whose range is , the step length is .
[0082] S2. Decoding the first-order projection component of the received codeword based on fast Hadamard transform to obtain a first-order decoding result;
[0083] Specifically, according to the projection method in step S1, it can be obtained by Degrade to The first-order projection component of The best decoding method for the first-order RM code so far is the decoding algorithm based on the fast Hadamard transform (FHT).
[0084] S2.1, iteratively calculating the first-order projection component of the received codeword based on Hadamard matrix multiplication to obtain a decoding vector;
[0085] Specifically, the calculation expression of the decoding vector is as follows:
[0086]
[0087] in, represents the decoding vector; represents the first-order projection component of the received codeword, which stores the log-likelihood ratio value; represents the Hadamard matrix.
[0088] S2.2, find the position index of the maximum absolute value of the decoded vector, and save the position index with the rightmost bit as the most significant bit;
[0089] Specifically, the position index is expressed as follows:
[0090]
[0091] in, Indicates the position index; Denotes the decoding vector Index is The absolute value of .
[0092] S2.3, calculating a binary index of a first-order projection component of the received codeword based on the position index;
[0093] Specifically, the calculation expression of the binary index is as follows:
[0094]
[0095] in, Represents a binary index; Indicates based on The judgment is 0 or 1; Denotes the decoding vector Index is The value of Indicates the code length.
[0096] S2.4. Calculate the decoding codeword based on the binary index and the first-order generation matrix of the received codeword to obtain a first-order decoding result.
[0097] Specifically, the first-order decoding result is expressed as follows:
[0098]
[0099] in, Indicates the first-order decoding result; express The first-order generating matrix of ; express The generator matrix of Indicates 1; express The generator matrix of express The generator matrix of express The generator matrix of express The generator matrix of .
[0100] S3, performing hard decision on the first-order projection component of the received codeword, and comparing the hard decision result and the first-order decoding result based on the first-order projection component to obtain a comparison result;
[0101] Specifically, the present invention obtains through step S1 projection, which is expressed as ; Make a hard decision on the projection and get the hard decision result ; The first-order decoding results of all received code words are obtained by fast Hadamard transform decoding in step S2, which is expressed as ,Will and By comparison, we can see from the hard decision rule that if the log-likelihood ratio is greater than 0, the hard decision result is 0; if the log-likelihood ratio is less than 0, it is 1; this is the same as the sign bit rule of binary numbers. Moreover, the projection The log-likelihood ratio value has been quantized, so only the first-order decoding result needs to be With projection By comparing the sign bit of the hard decision result with the first-order decoding result, the comparison result can be obtained. Record the number of different sign bits. If It means that the hard decision result is the same as the first-order decoding result; if This means that the hard decision result is different from the first-order decoding result.
[0102] S4, aggregating the received codeword based on the comparison result and the first-order decoding result to obtain a decoded codeword;
[0103] S4.1. When the number of iterative decoding is less than the preset minimum number of iterations, the received codeword is flipped according to the first-order decoding result, and the average of the flipped received codeword is taken to obtain the decoded codeword through hard decision; in a specific embodiment of the present invention, the minimum number of iterations is set to 2. The flipping decision specifically includes:
[0104] If the index value of the first-order decoding result is 0, the log-likelihood ratio value corresponding to the index in the received codeword remains unchanged;
[0105] If the index value of the first-order decoding result is 1, the log-likelihood ratio value corresponding to the index in the received codeword is negated.
[0106] S4.2, when the number of iterative decoding is greater than the preset minimum number of iterations but less than the preset maximum number of iterations, if the hard decision result is the same as the first-order decoding result, the aggregation step of the current round is skipped, and the iterative decoding is terminated, and the hard decision result of the current round is used as the final decoded codeword; the minimum number of iterations in the specific embodiment of the present invention is set to .
[0107] S4.3, when the number of iterative decoding is greater than the preset minimum number of iterations but less than the preset maximum number of iterations, if the hard decision result and the first-order decoding result are different, the received codeword is flipped according to the first-order decoding result to obtain the flipped received codeword. The specific flipping strategy can refer to step S4.1, and then the average value of the flipped received codeword is taken and the decoded codeword is obtained through hard decision.
[0108] S4.4, when the number of iterative decoding reaches the preset maximum number of iterations, if the hard decision result is the same as the first-order decoding result, the aggregation step of the current round is skipped, and the iterative decoding is terminated, and the hard decision result of the current round is used as the final decoding codeword;
[0109] S4.5, when the number of iterative decoding reaches the preset maximum number of iterations, if the hard decision result and the first-order decoding result are different, the received codeword is flipped according to the first-order decoding result to obtain the flipped received codeword; the specific flipping strategy can refer to step S4.1. Then, the average value of the flipped received codeword is taken and the final decoded codeword is obtained through hard decision.
[0110] S5. The decoded codeword is used as a new round of received codeword for iterative decoding until a condition for terminating iterative decoding is met to obtain a final decoded codeword.
[0111] Specifically, the conditions for terminating iterative decoding include that the number of iterative decoding times is greater than a preset minimum number of iterations and the sign bits of the hard decision result and the first-order decoding result are the same, and the number of iterative decoding times reaches a preset maximum number of iterations.
[0112] Reference Figure 2 and Figure 3 The present invention provides a hardware structure of a Reed-Muller decoder based on a fast iterative projection aggregation decoding algorithm, including a fast projection module, a first-order decoding module, a comparison module and an aggregation module, wherein:
[0113] The first-order decoding module includes a Hadamard matrix multiplication module, a maximum value finding module and a generation matrix multiplication module;
[0114] The fast projection module is used to perform fast projection on the received codeword to obtain the first-order projection component of the received codeword;
[0115] The first-order decoding module decodes the first-order projection component of the received codeword based on the fast Hadamard transform to obtain a first-order decoding result;
[0116] The comparison module is used to perform hard decision on the first-order projection component of the received codeword, and compare the hard decision result with the first-order decoding result based on the first-order projection component to obtain a comparison result;
[0117] The aggregation module aggregates the received codewords based on the comparison result and the first-order decoding result to obtain a decoded codeword;
[0118] The Hadamard matrix multiplication module iteratively calculates the first-order projection component of the received codeword based on the Hadamard matrix multiplication to obtain a decoding vector;
[0119] The maximum value finding module is used to find the position index of the maximum absolute value of the decoded vector, and calculate the binary index of the first-order projection component of the received codeword based on the position index;
[0120] The generator matrix multiplication module calculates the decoding codeword based on the binary index and the first-order generator matrix of the received codeword to obtain a first-order decoding result.
[0121] The fast projection module can refer to Figure 4 , for the received codeword , and its log-likelihood ratio is expressed as , input into the routing network module, have First-order projection , the routing network module selects the Index (Index numbers range from 0 to -1) Rearrange and then calculate different projections using formula 10 .
[0122] The Hadamard matrix multiplication module can refer to Figure 5 Since the Hadamard matrix is iteratively generated and has its own regularity, the Hadamard matrix multiplication module of the present invention can be used to input the first-order projection component Perform m additions and subtractions to obtain a decoded vector, and in order to reduce the critical path, a pipeline is inserted at the result of each level of addition and subtraction in a specific embodiment of the present invention.
[0123] The maximum value finding module can refer to Figure 6, take the absolute value of the decoded vector and input it into the maximum value search module, obtain the index of the maximum absolute value of the decoded vector by pairwise comparison, save the index in binary with the rightmost bit as the most significant bit, and determine whether the log-likelihood ratio value corresponding to the index is greater than 0. If it is greater than 0, add the number 0 to the leftmost bit of the index; if the log-likelihood ratio value corresponding to the index is less than or equal to 0, add the number 1 to the leftmost bit of the index, and obtain a length of Binary index of the bit .
[0124] The matrix multiplication module can be referred to Figure 7 , the binary index and The product operation is performed on the generator matrix to obtain the first-order decoding result.
[0125] The comparison module can refer to Figure 8 , the first-order decoding result With projection By comparing the sign bit of the hard decision result with the first-order decoding result, the comparison result can be obtained. Record the number of different sign bits. If , then the decoding module is not entered, and the iterative decoding is terminated, and the hard decision result of the current round is used as the final decoding codeword; if 0 enters the aggregation module.
[0126] In order to verify the actual technical effect of the method of the present invention, the present invention proposes RM (7, 2) decoder and RM (6, 3) decoder based on the fast iterative projection aggregation decoding algorithm (Fast-IPA), and compares them with the iterative projection aggregation decoding (IPA) method. In terms of the preset number of iterative decoding, it is found during simulation that when the number of iterations is set to 2, it has a good performance, and increasing the number of iterations will not have a large performance gap. Therefore, the number of iterative decoding is set to 2. In the aggregation module of the fast iterative projection aggregation decoding, the first-order decoding result and the projection result will be compared in advance. If the results are the same, the decoding will be terminated. In order to facilitate the comparison with the iterative projection aggregation decoding algorithm, Table 1 compares the first-order decoding times of the decoders in two iterative decodings under the assumption that the termination condition does not occur.
[0127] Table 1 Comparison of the effects of Fast-IPA and IPA
[0128]
[0129] Whether it is the IPA algorithm or the Fast-IPA algorithm, the complexity of the algorithm can be measured by the number of first-order code decodings involved. It can be found from Table 1 that the Fast-IPA method proposed in the present invention can greatly reduce the number of projections, that is, reduce the number of first-order code decodings, simplify and speed up the decoding process.
[0130] The performance of the RM (7,2) decoder and the RM (6,3) decoder based on the fast iterative projection aggregation decoding algorithm are simulated in the additive white Gaussian noise (AWGN) channel. The modulation mode is binary phase shift keying (BPSK). The simulation is stopped when at least 10,000 codewords are transmitted and at least 100 errors are collected. Fig. 9 It can be found that whether it is RM(7,2) or RM(6,3), compared with the original IPA decoding method, even if the Fast-IPA method pruned 74.80% and 86.89% of the projections in the decoding process of RM(7,2) and RM(6,3), the difference in frame error rate (FER) is basically less than 0.1dB, which means that the performance loss is extremely small, and compared with the polar code with the same code length and code rate based on SCL decoding, it still has a great advantage when the list size is 16 or 32.
[0131] Reference Fig.10 , the timing diagram of the RM (7, 2) decoder based on Fast-IPA when the overall decoder parallelism is 4. The overall architecture design adopts an overlap structure. In the figure, FP represents the fast projection operation. Since the parallelism is 4, each projection operation obtains 4 first-order projections. The RM (7, 2) code based on Fast-IPA has a total of 32 projections, so it needs to be repeated 8 times. Therefore, the timing Figure 1There are 8 rows in total. In addition, in actual hardware use, it is found that after using Fast-IPA's projection algorithm to project RM (7, 2) into RM (6, 1) code, the adjacent two-bit log-likelihood ratio is projected again to obtain RM (5, 1) code, with no obvious performance loss. Therefore, the first-order decoder uses the FHT decoder of RM (5, 1). The FOD in the figure is the first-order decoding process. Due to the insertion of the pipeline, it takes 5 clock cycles to complete the first-order decoding operation. In the figure, agg means the aggregation process, which averages the results of the four first-order decoders. Mean_2 means waiting for the results of two 4-parallel decodings to be completed, and then averaging the results of these two times. Similarly, Mean_4 and Mean_8 are both operations of averaging the previous results after they are completed. In this way, 8 rows of 4-parallel decoding, that is, the decoding and aggregation operations of 32 first-order codes are completed, and the decoding result of one iterative decoding is obtained after 18 clock cycles. If the number of iterative decoding is set to 2, a complete decoding requires 36 clock cycles.
[0132] The Fast-IPA decoder proposed in the present invention is synthesized using Synopsys Design Compiler and TSMC 28nm process; the IPA decoder used for comparison is synthesized using Cadence Genus RTL and STM 28nm FD-SOI process. The synthesis results are shown in Table 2.
[0133] For Fast-IPA decoder, delay The calculation is as follows:
[0134]
[0135] in, represents the clock cycle; Indicates the clock frequency.
[0136] Throughput The calculation is as follows:
[0137]
[0138] in, Indicates the throughput rate; Indicates the degree of parallelism.
[0139] Table 2 Comprehensive comparison results
[0140]
[0141] In the table, SCL polar (64, 42) indicates that the polar code length used for comparison is 64, the information bit length is 42, the decoding method is the SCL decoding method, and L = 64 is the list size in the SCL decoding method. The meanings of other polar parameters in the table are the same as above. The code length and information bit length of the (64, 42) polar code are the same as those of the RM (6, 3) code to be compared, and the code length and information bit length of the (128, 29) polar code are the same as those of the RM (7, 2) code to be compared.
[0142] From the comprehensive results, it can be seen that the decoder based on the Fast-IPA method has significant advantages over the RM decoder based on the IPA method or the polar decoder based on SCL decoding in terms of frequency, delay, throughput, area, and area efficiency. For example, by comparing the RM (6, 3) decoders based on IPA and Fast-IPA, the decoder based on Fast-IPA has a 17.65% increase in frequency, a 82.31% reduction in clock cycle, a 85.40% reduction in delay, an 8.4-fold increase in throughput, a 60.53% reduction in area, and a 22.8-fold increase in area efficiency. In addition, the 4-parallel Fast-IPA decoder even outperforms the 8-parallel IPA decoder in all aspects.
[0143] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.
Claims
1. A Reed-Muller decoder based on a fast iterative projection aggregation decoding algorithm, characterized in that: The following steps are involved: Performing fast projection on the received codeword to obtain the first-order projection component of the received codeword; Decoding the first-order projection component of the received codeword based on fast Hadamard transform to obtain a first-order decoding result; Performing a hard decision on a first-order projection component of the received codeword, and comparing the hard decision result with the first-order decoding result based on the first-order projection component to obtain a comparison result; Aggregating the received codeword based on the comparison result and the first-order decoding result to obtain a decoded codeword; Iteratively decoding the decoded codeword as a new round of received codewords until a condition for terminating iterative decoding is met to obtain a final decoded codeword; The step of quickly projecting the received codeword to obtain the first-order projection component of the received codeword specifically includes: Preliminarily pruning the projection components of the received codeword to determine the number of retained first-order projection components; Performing index confirmation based on the number of the first-order projection components and storing the index in a first index variable according to the order; Calculating an index pair based on the first index variable, and storing the index values in increasing order in a second index variable based on the index pair; Rearranging the high-order second index variables based on the low-order second index variables to obtain the associated projection arrangement of each order of the projection components; Based on the associated projection arrangement of each order of the projection components, adjacent log-likelihood ratio values of each order of the received codeword are projected into a single value to obtain a first-order projection component of the received codeword.
2. The Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm according to claim 1, characterized in that: The step of pre-pruning the projection components of the received codeword to determine the number of retained first-order projection components specifically includes: A fixed number of projection components are selected based on the code length of the received code word as each order projection of the received code word; After projecting the received codewords step by step based on a fixed number of projection components, the number of retained first-order projection components is obtained.
3. The Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm according to claim 1, characterized in that: The step of decoding the first-order projection component of the received codeword based on the fast Hadamard transform to obtain the first-order decoding result specifically includes: Iteratively calculate the first-order projection component of the received codeword based on Hadamard matrix multiplication to obtain a decoding vector; Find the position index of the maximum absolute value of the decoded vector, and save the position index with the rightmost bit as the most significant bit; Calculate a binary index of a first-order projection component of the received codeword based on the position index; The decoding codeword is calculated based on the binary index and the first-order generation matrix of the received codeword to obtain a first-order decoding result.
4. The Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm according to claim 1, characterized in that: The step of aggregating the received codeword based on the comparison result and the first-order decoding result to obtain the decoded codeword specifically includes: When the number of iterative decoding is less than the preset minimum number of iterations, the received codewords are flipped according to the first-order decoding results, and the average of the flipped received codewords is taken to obtain the decoded codeword through hard decision; When the number of iterative decoding is greater than the preset minimum number of iterations, if the hard decision result is the same as the first-order decoding result, the aggregation step of the current round is skipped, and the iterative decoding is terminated, and the hard decision result of the current round is used as the final decoded codeword; When the number of iterative decoding is greater than the preset minimum number of iterations, if the hard decision result is different from the first-order decoding result, the received codeword is flipped according to the first-order decoding result, and the average of the flipped received codeword is taken to obtain the decoded codeword through hard decision.
5. The Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm according to claim 1, characterized in that: The conditions for terminating iterative decoding include that the number of iterative decoding times is greater than a preset minimum number of iterations and the sign bits of the hard decision result and the first-order decoding result are the same, and the number of iterative decoding times reaches a preset maximum number of iterations.
6. The Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm according to claim 4, characterized in that: The flip decision specifically includes: If the index value of the first-order decoding result is 0, the log-likelihood ratio value corresponding to the index in the received codeword remains unchanged; If the index value of the first-order decoding result is 1, the log-likelihood ratio value corresponding to the index in the received codeword is negated.
7. The Reed-Muller decoder based on the fast iterative projection aggregation decoding algorithm according to claim 1, characterized in that: Its hardware structure includes fast projection module, first-order decoding module, comparison module and aggregation module, among which: The first-order decoding module includes a Hadamard matrix multiplication module, a maximum value finding module and a generation matrix multiplication module; The fast projection module is used to perform fast projection on the received codeword to obtain the first-order projection component of the received codeword; The first-order decoding module decodes the first-order projection component of the received codeword based on the fast Hadamard transform to obtain a first-order decoding result; The Hadamard matrix multiplication module iteratively calculates the first-order projection component of the received codeword based on the Hadamard matrix multiplication to obtain a decoding vector; The maximum value finding module is used to find the position index of the maximum absolute value of the decoded vector, and calculate the binary index of the first-order projection component of the received codeword based on the position index; The generator matrix multiplication module calculates the decoding codeword based on the binary index and the first-order generator matrix of the received codeword to obtain a first-order decoding result; The comparison module is used to perform hard decision on the first-order projection component of the received codeword, and compare the hard decision result with the first-order decoding result based on the first-order projection component to obtain a comparison result; The aggregation module aggregates the received codewords based on the comparison result and the first-order decoding result to obtain a decoded codeword.
