A multi-step decoding decision acceleration decoding method, device and storage medium

Abstract

The present invention relates to the field of communication technologies, and discloses a multi-step decoding decision accelerated decoding method, device, and storage medium. The method includes: for the decoding of linear block codes based on a neural network learning algorithm, calculating the component addition of multiple iterative decoding results as the basis for selecting the learning rate; based on different code lengths, code rates, and channel conditions, by setting the number of component times and addition coefficients of the iterative decoding results, achieving the optimal adjustment of the weights of the neural network learning algorithm, thereby improving the learning rate. The present invention uses the component addition of multiple iterative decoding results as the basis for selecting the learning rate, and realizes the optimal adjustment of the neural network weights under different code lengths, code rates, and different channel conditions by setting the selection of the number of component times and addition coefficients of the iterative decoding results, thereby improving the training learning rate.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and in particular, to a multi-step decoding decision acceleration decoding method, device, and storage medium. Background Art

[0002] With the rapid development of communication, which involves supporting high speed, low latency, high reliable transmission, etc., for example, autonomous driving, etc. The as-correct-as-possible transmission of information has always been the main theme of the development of communication technologies. As an indispensable anti-interference technology, channel error correction code technology, linear block codes are widely used in communication systems with their low complexity performance to ensure the correct transmission of information. The learning algorithms of neural networks have proven their powerful classification and fitting capabilities in application scenarios such as speech, image, and natural language processing. Combining the neural network learning algorithm with linear block code decoding has proven that the performance can be improved compared to the original decoding algorithm. Especially for some linear block codes that were difficult to use soft decision iterative decoding in the past, good decoding performance has been achieved after using the neural network learning algorithm for decoding.

[0003] The existing technologies focus on using neural networks to construct encoding and decoding systems, and basically perform decoding processing according to conventional processing methods, lacking targeted research on improving the efficiency of decision processing in the decoding process, thus forming related technologies. For the linear block code decoding using the neural network learning algorithm, as the length of the linear block code increases, its training convergence speed becomes very slow, which makes the initial model training and subsequent application enhancement training time costs very large. Summary of the Invention

[0004] To solve the above problems, the present invention proposes a multi-step decoding decision acceleration decoding method, device, and storage medium, which can improve the training learning rate and is applicable to a wide range of linear block codes.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A multi-step decoding decision acceleration decoding method, comprising:

[0007] For the linear block code decoding based on the neural network learning algorithm, calculate the component addition of the decoding results of multiple iterations as the basis for selecting the learning rate;

[0008] Based on different code lengths, code rates, and channel conditions, by setting the number of component times of the iterative decoding result and the addition coefficient, realize the optimal adjustment of the weights of the neural network learning algorithm, thereby improving the learning rate.

[0009] Further, it includes the following steps:

[0010] Step 1, set the initial step size of the weights of the neural network learning algorithm ;

[0011] Step 2, calculate the output after the end of the t-th iteration of the linear block code decoding based on the neural network learning algorithm , and determine whether the decoding is successful. If successful, end; otherwise, execute the next step;

[0012] Step 3, calculate the loss function of the cross entropy in the t-th iteration , where is the binary block code;

[0013] Step 4, based on the loss function of the cross entropy , calculate the accumulated value of the t-th iteration ;

[0014] Step 5, based on the accumulated value perform step size adjustment to obtain the updated step size ;

[0015] Step 6, based on the updated step size adjust the weights of the neural network learning algorithm, execute the next iteration and go back to Step 2.

[0016] Furthermore, the weights of the neural network learning algorithm include:

[0017]

[0018] where, is the weight of the j -th variable in the t-th iteration, is the weight between the check node and the variable node in the t-th iteration, is the weight between the check node and the variable node in the t-th iteration, is a positive integer, is the value in the check set participated by the variable node .

[0019] Furthermore, in Step 2, calculating the output after the end of the t-th iteration of the linear block code decoding based on the neural network learning algorithm includes:

[0020]

[0021] In the formula, is an operation, and there is ; is the log-likelihood function, is the t -th iteration of the check node and the variable node The calculated value of the variable between is the variable node participating in the check set.

[0022] Furthermore, in step 3, calculate the loss function of cross-entropy in the t-th iteration , including:

[0023]

[0024] In the formula, is a binary block code with a code length of N and an information bit length of N - M, where N and M are positive integers; is the th codeword.

[0025] Furthermore, in step 4, based on the loss function of cross-entropy , calculate the cumulative value in the t-th iteration, including:

[0026]

[0027] In the formula, is the cumulative influence factor in the t-th iteration, is the th cumulative influence factor in the t-th iteration, .

[0028] Furthermore, in step 5, perform step size adjustment based on the cumulative value to obtain the updated step size , including:

[0029] Adopt the first terms of the cumulative value in the t-th iteration as the current step size:

[0030]

[0031] In the formula, is the step size for different iteration times, and the subscript is the iteration number;

[0032] Set a positive deviation value and a minimum value ;

[0033] If , then:

[0034]

[0035] If , then:

[0036]

[0037] If , then:

[0038] .

[0039] Further, based on the updated step size adjust the weights of the neural network learning algorithm, including:

[0040] .

[0041] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the above multi-step decoding decision accelerated decoding method is implemented.

[0042] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above multi-step decoding decision accelerated decoding method is implemented.

[0043] The beneficial effects of the present invention are as follows:

[0044] The present invention uses the component addition of the iterative decoding results as the basis for learning rate selection, and realizes the optimal adjustment of the neural network weights under different code lengths, code rates, and different channel conditions by setting the number of component times of the iterative decoding results and the selection of the addition coefficient, thereby improving the training learning rate. The present invention is applicable to a wide range of linear block codes, including low-density parity-check codes (LDPC). BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a bidirectional graph of a linear block code in Embodiment 1 of the present invention.

[0046] Figure 2 is a flowchart of a multi-step decoding decision accelerated decoding method in Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] In order to have a clearer understanding of the technical features, objectives, and effects of the present invention, the specific embodiments of the present invention are now described. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0048] Embodiment 1

[0049] Since the training convergence speed of linear block code decoding using neural network learning algorithms becomes very slow as the length of the linear block code increases, the initial model training and subsequent application enhancement training time costs are very high.

[0050] The linear block code decoding method based on neural network learning algorithms is as follows:

[0051] Taking binary block code as an example, where N is the code length, N - M is the information bit length, and N and M are positive integers. Any codeword , is transmitted in the form of , where , , is the binary field, . Suppose the codeword sent by the sender is . After transmission mapping and binary phase shift keying (BPSK) modulation, it passes through a noisy channel and finally reaches the receiver. The receiver demodulates it and outputs the decision signal , and sends it to the channel decoder. Therefore, is the received signal (or input signal) of the channel decoder. The main task of the decoder is to perform error correction decoding on the received sequence , filter out channel errors, and recover the sent codeword from the received signal with as small a decoding error probability as possible.

[0052] Let represent the conditional probability distribution function of the channel output. The log-likelihood ratio (LLR) function of the code element can be calculated:

[0053]

[0054] In the formula, .

[0055] Suppose a binary block code has a parity-check matrix , where h ij is the H th i row and j th column element of the parity-check matrix Figure 1 is its bipartite graph representation. The codeword is represented as a set of information nodes , is represented as a set of check nodes . Only when , the node to Connected by a directed edge. Let the set represent the parity-check set that variable participates in, represent a subset that does not contain . represent the local code element information set constrained by the parity-check node , represent a subset that does not contain . The past bipartite graph was used to describe a special class of linear block codes - LDPC codes, but it is currently also used to describe general linear block codes. Especially, good results have been obtained through the neural network decoding of BCH codes on the bipartite graph, making the bipartite graph more generalized.

[0056] The iteration of the linear block code decoding method based on the neural network learning algorithm is as follows:

[0057] For a positive integer t, the t-th iteration is as follows:

[0058]

[0059]

[0060] Among them, is the coefficient for training the neural network decoding, is 's inverse function. Let:

[0061]

[0062]

[0063] Then:

[0064] .

[0065] Based on this, to solve the problem that the coefficient training convergence rate of the linear block code decoding method based on the neural network learning algorithm decreases as the code length increases, this embodiment provides a multi-step decoding decision accelerated decoding method. By using the component addition of the decoding results of multiple iterations as the basis for selecting the learning rate, and based on different code lengths, code rates, and channel conditions, by setting the number of component times and addition coefficients of the iterative decoding results, the optimal adjustment of the weights of the neural network learning algorithm is realized, thereby improving the learning rate.

[0066] Specifically, as Figure 2 shown, the method of this embodiment can be implemented by the following steps:

[0067] Step 1, set the initial step size of the weights of the neural network learning algorithm ;

[0068] Step 2: Calculate the output after the t-th iteration of the linear block code decoding based on the neural network learning algorithm , and determine whether the decoding is successful. If successful, end; otherwise, execute the next step;

[0069] Step 3: Calculate the loss function of cross entropy in the t-th iteration , where is the binary block code;

[0070] Step 4: Based on the loss function of cross entropy , calculate the accumulated value in the t-th iteration ;

[0071] Step 5: Adjust the step size based on the accumulated value to obtain the updated step size ;

[0072] Step 6: Adjust the weights of the neural network learning algorithm based on the updated step size , execute the next iteration and go back to Step 2.

[0073] In this embodiment, the weights of the neural network learning algorithm include:

[0074]

[0075] where is the weight of the j -th variable in the t-th iteration, is the weight between the check node and the variable node in the t-th iteration, is the weight between the check node and the variable node in the t-th iteration, is a positive integer, is the value in the check set participated by the variable node .

[0076] It should be noted that the initial step sizes of the weights of the neural network learning algorithm can be the same or different. In this embodiment, the case where the initial step sizes are the same is taken as an example for illustration.

[0077] Preferably, in Step 2, calculating the output after the t-th iteration of the linear block code decoding based on the neural network learning algorithm

[0078]

[0079] In the formula, is an operation, and there is ; is the log-likelihood function, is the t th iteration of the variable calculation value between the check node and the variable node ; is the check set participated by the variable node .

[0080] Preferably, in step 3, calculate the loss function of the cross-entropy in the t-th iteration , including:

[0081]

[0082] In the formula, is a binary block code with a code length of N and an information bit length of N - M, where N and M are positive integers; is the rd codeword.

[0083] Preferably, in step 4, based on the loss function of the cross-entropy , calculate the accumulated value in the t-th iteration, including:

[0084]

[0085] In the formula, is the accumulated influence factor in the t-th iteration, is the th accumulated influence factor in the t-th iteration, . can be the same or different.

[0086] Preferably, in step 5, perform step size adjustment based on the accumulated value to obtain the updated step size , including:

[0087] Adopt the first terms of the accumulated value in the t-th iteration as the current step size:

[0088]

[0089] In the formula, is the step size of different iteration times, and the subscript is the iteration number;

[0090] Set a positive deviation value and a minimum value ; Preferably, .

[0091] If , then:

[0092]

[0093] If , then:

[0094]

[0095] If , then:

[0096] .

[0097] Preferably, based on the updated step size adjust the weights of the neural network learning algorithm, including:

[0098] .

[0099] In summary, the method of this embodiment uses the component addition of the iterative decoding results multiple times as the basis for learning rate selection, and realizes the optimal adjustment of the neural network weights under different code lengths, code rates, and different channel conditions by setting the number of times of iterative decoding result components and the selection of addition coefficients, thereby improving the training learning rate.

[0100] Embodiment 2

[0101] Based on Embodiment 1, this embodiment:

[0102] This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the multi-step decoding decision acceleration decoding method of Embodiment 1. Among them, the computer program can be in the form of source code, object code, executable file, or some intermediate form, etc.

[0103] Embodiment 3

[0104] Based on Embodiment 1, this embodiment:

[0105] This embodiment provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the multi-step decoding decision acceleration decoding method of Embodiment 1. Among them, the computer program can be in the form of source code, object code, executable file or some intermediate form, etc. The storage medium includes: any entity or device capable of carrying the computer program code, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice within the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the storage medium does not include electrical carrier signals and telecommunication signals.

[0106] It should be noted that, for the foregoing method embodiments, for the sake of simplicity of description, they are expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

Claims

1. A multi-step decoding decision acceleration decoding method, characterized in that, including: For the linear block code decoding based on the neural network learning algorithm, calculate the component addition of the iterative decoding results for multiple times as the basis for the selection of the learning rate; Based on different code lengths, code rates, and channel conditions, by setting the number of component times and the addition coefficient of the iterative decoding results, realize the optimal adjustment of the weights of the neural network learning algorithm, thereby improving the learning rate; The multi-step decoding decision acceleration decoding method includes the following steps: Step 1, set the initial step size of the weights of the neural network learning algorithm ; Step 2: Calculate the output after the t-th iteration of the linear block code decoding based on the neural network learning algorithm , and determine whether the decoding is successful. If it is successful, end; otherwise, execute the next step; Step 3: Calculate the loss function of cross entropy in the t-th iteration , where is a binary block code; Step 4: Loss function based on cross entropy , calculate the cumulative value at the t-th iteration ; Step 5. Based on the accumulated value perform step size adjustment to obtain an updated step size ; Step 6: Based on the updated step size Adjust the weights of the neural network learning algorithm, perform the next iteration, and go to Step 2; In step 5, based on the accumulated value perform step size adjustment to obtain an updated step size , including: Adopt the accumulated value of the t-th iteration The first terms are the current step size: In the formula, is the step size for different iteration times, and the subscript is the iteration time; Set a positive deviation value and the minimum value ; If , then: If , then: If , then: 。 2. The multi-step decoding decision acceleration decoding method according to claim 1, characterized in that The weights of the neural network learning algorithm include: Among them, is the weight of the j -th variable at the t-th iteration, is the weight between the check node and the variable node at the t-th iteration, is the weight between the check node and the variable node at the t-th iteration, is a positive integer, is the value in the check set participated by the variable node .

3. A multi-step decoding decision acceleration decoding method according to claim 2, characterized in that, In step 2, calculate the output after the t-th iteration of the linear block code decoding based on the neural network learning algorithm , including: In the formula, is an operation, and there is ; is the log-likelihood function, is the t -th iteration of the variable calculation value between the check node and the variable node , is the check set participated by the variable node .

4. A multi-step decoding decision acceleration decoding method according to claim 3, characterized in that In step 3, calculate the loss function of cross entropy in the t-th iteration , including: In the formula, is a binary block code with a code length of N and an information bit length of N - M, where N and M are positive integers; is the th codeword.

5. A multi-step decoding decision acceleration decoding method according to claim 4, characterized in that In step 4, the loss function based on cross entropy , calculate the accumulated value at the t-th iteration , including: In the formula, is the cumulative influence factor of the t-th iteration, is the -th cumulative influence factor of the t-th iteration, .

6. A multi-step decoding decision acceleration decoding method according to claim 1, characterized in that In step 6, based on the updated step size Adjust the weights of the neural network learning algorithm, including: 。 7. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it realizes the multi-step decoding decision acceleration decoding method according to any one of claims 1-6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it realizes the multi-step decoding decision acceleration decoding method according to any one of claims 1-6.

Images