CN-121984523-A - SC-LDPC code-based grouping parameter sharing neural sliding window decoding method

CN121984523ACN 121984523 ACN121984523 ACN 121984523ACN-121984523-A

Abstract

The invention discloses a grouping parameter sharing neural sliding window decoding method based on an SC-LDPC code, which comprises the steps of building a neural network based on a basic mode diagram corresponding to an input basic matrix, wherein the dimensions of an input layer and an output layer of the neural network are determined by the number of variable nodes of the basic mode diagram, the dimensions of a hidden layer of the neural network are determined by the number of edges of the basic mode diagram, training the neural network by using a sliding window mechanism-based loss function in a parameter and other window training mode through a gradient optimization method, and applying a parameter group obtained according to the trained neural network to each window of sliding window decoding to decode the SC-LDPC code, so that a decoding result is output, and the SC-LDPC code decoding method capable of effectively reducing the complexity and the storage cost of the neural network training and avoiding the occurrence of dimensional explosion problem is provided.

Inventors

ZHU MIN
GAO WANTING
WAN FEI
BAI BAOMING

Assignees

西安电子科技大学

Dates

Publication Date: 20260505
Application Date: 20260115

Claims (10)

1. The method for decoding the packet parameter sharing neural sliding window based on the SC-LDPC code is characterized by comprising the following steps: building a neural network based on a basic mode diagram corresponding to an input basic matrix, wherein the dimensions of an input layer and an output layer of the neural network are determined by the number of variable nodes of the basic mode diagram, and the dimensions of a hidden layer of the neural network are determined by the number of edges of the basic mode diagram; Training the neural network by using a window training mode such as parameters and a loss function based on a sliding window mechanism through a gradient optimization method; And applying the parameter set obtained according to the trained neural network to each window of sliding window decoding so as to decode the SC-LDPC code and output a decoding result.
2. The method for decoding a packet parameter sharing neural sliding window according to claim 1, wherein each hidden layer in the neural network comprises a first sub-layer and a second sub-layer, the first sub-layer corresponds to a process of transmitting a message from a variable node to a check node, and the second sub-layer corresponds to a process of transmitting a message from a check node to a variable node.
3. The method of packet parameter sharing neural sliding window decoding according to claim 2, wherein the process of message passing in the first sub-layer comprises: ; Wherein, the Represent the first A first sub-layer of the plurality of hidden layers; representation correspondence Intermediate slave variable node To check nodes Is a message of (2); represent the first A second sub-layer in the hidden layer; representation correspondence Well from check node To the variable node Is a message of (2); ; Indicating the total number of hidden layers; representation and the variable node A set of all check node connection edges associated; representation and the variable node All of the associated check nodes except the check node A collection of connection edges; representation and the variable node All check nodes connected except the check node The rest check node sets except the check node sets.
4. The method of packet parameter sharing neural sliding window decoding according to claim 2, wherein the process of message passing in the second sub-layer comprises: ; Wherein, the Represent the first A first sub-layer of the plurality of hidden layers; represent the first A second sub-layer in the hidden layer; representation correspondence Well from check node Addressed to variable nodes Is a message of (2); representation and the check node All of the variable nodes associated except the variable node A collection of connection edges; representation and the check node All variable nodes connected except the variable node The rest variable node sets except the variable node sets; representing a sign function; representation correspondence Intermediate slave variable node To the check node Is a message of (2); representing a linear function; Representing the normalization factor; representation and the check node A set of connected variable nodes; representation and the check node All of the variable nodes associated except the variable node A collection of connection edges; representing the offset.
5. The method for decoding a sliding window of a packet parameter sharing nerve according to claim 1, wherein all edges of sub-blocks belonging to the same SC-LDPC code in the window training mode of parameters and the like share the same set of parameters at the same iteration number.
6. The method of packet parameter sharing neural sliding window decoding according to claim 1, wherein the loss function comprises: ; Wherein, the Representing the loss function; An output vector representing the neural network; representing a codeword vector; representing codeword length; represent the first A neuron message; The influence degree of the data at different positions in the window on the current decoding symbol is represented; Representing the transmission of codeword sequences The elements.
7. An SC-LDPC code-based packet parameter sharing neural sliding window decoding device, comprising: The device comprises a building module, a neural network, a first matrix and a second matrix, wherein the building module is used for building a neural network based on a basic mode diagram corresponding to an input basic matrix, the dimensions of an input layer and an output layer of the neural network are determined by the number of variable nodes of the basic mode diagram, and the dimensions of a hidden layer of the neural network are determined by the number of edges of the basic mode diagram; the training module is used for training the neural network by adopting a window training mode such as parameters and a loss function based on a sliding window mechanism through a gradient optimization method; And the output module is used for applying the parameter set obtained according to the trained neural network to each window of sliding window decoding so as to decode the SC-LDPC code and output a decoding result.
8. The packet parameter sharing neural sliding window decoding device according to claim 7, wherein each hidden layer in the neural network comprises a first sub-layer and a second sub-layer, wherein the first sub-layer corresponds to a process of transmitting a message to a check node by a variable node, and the second sub-layer corresponds to a process of transmitting a message to a variable node by a check node.
9. The electronic equipment is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus; a memory for storing a computer program; A processor configured to implement the SC-LDPC code-based packet parameter sharing neural sliding window decoding method of any one of claims 1 to 6 when executing a computer program stored on a memory.
10. A computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, and when the computer program is executed by a processor, the method for decoding the SC-LDPC code-based grouping parameter sharing neural sliding window is implemented according to any one of claims 1 to 6.

Description

SC-LDPC code-based grouping parameter sharing neural sliding window decoding method Technical Field The invention belongs to the technical field of communication, and particularly relates to a grouping parameter sharing neural sliding window decoding method based on an SC-LDPC (space coupling low density parity check) code. Background In recent years, neural network-assisted decoding algorithms exhibit excellent performance for medium-short code lengths. However, in the case of long codes, the training scale of the neural network is often huge, which may cause dimensional explosion and difficult application. To solve this problem, we focus on the processing of long codes. Spatially coupled low density parity check codes are widely subject to academic and industrial attention because they can significantly improve error correction performance in the case of long codes, and their key is the design of the spatial coupling. Thus, SC-LDPC codes are chosen to explore how to deal with the application of long or semi-infinite coupled base mode patterns in neural networks. In recent years, deep learning methods have been rapidly developed in channel decoding. In literature Learning to decode linear codes using DEEP LEARNING, a learnable weight is applied to the Sum-Product (SP) algorithm, which is improved by 1.5dB compared to the conventional SP algorithm when decoding a High Density Parity Check (HDPC) matrix. Articles Neural offset min-sum decoding and DEEP LEARNING methods for improved decoding of linear codes have studied Normalized Minimum Sum (NMS)/Offset Minimum Sum (OMS) decoders with a learnable normalization/offset factor to provide a more hardware friendly neural decoder. However, these prior art documents are all studied on HDPC and have a code length of less than 200 bits. In literature Learning to decode protograph LDPC codes, authors propose a neural Minimum Sum (MS) decoding method of a base mode pattern LDPC code that exploits the structure lifting to overcome the code length limitation. However, for SC-LDPC codes based on a base mode pattern, the code length is generally long due to the long or semi-infinite coupling of the base mode pattern, and the number of neurons per layer is excessive, resulting in a sharp increase in computation and storage requirements, which may cause a problem of dimensional explosion. Therefore, how to provide an SC-LDPC code decoding method that can effectively reduce complexity and storage cost of neural network training and avoid occurrence of a dimensional explosion problem becomes an important problem. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a packet parameter sharing neural sliding window decoding method based on an SC-LDPC code. The technical problems to be solved by the invention are realized by the following technical scheme: in a first aspect, the present invention provides a method for decoding a packet parameter sharing neural sliding window based on an SC-LDPC code, including: building a neural network based on a basic mode diagram corresponding to an input basic matrix, wherein the dimensions of an input layer and an output layer of the neural network are determined by the number of variable nodes of the basic mode diagram, and the dimensions of a hidden layer of the neural network are determined by the number of edges of the basic mode diagram; Training the neural network by using a window training mode such as parameters and a loss function based on a sliding window mechanism through a gradient optimization method; And applying the parameter set obtained according to the trained neural network to each window of sliding window decoding so as to decode the SC-LDPC code and output a decoding result. In a second aspect, the present invention provides a packet parameter sharing neural sliding window decoding device based on SC-LDPC code, including: The device comprises a building module, a neural network, a first matrix and a second matrix, wherein the building module is used for building a neural network based on a basic mode diagram corresponding to an input basic matrix, the dimensions of an input layer and an output layer of the neural network are determined by the number of variable nodes of the basic mode diagram, and the dimensions of a hidden layer of the neural network are determined by the number of edges of the basic mode diagram; the training module is used for training the neural network by adopting a window training mode such as parameters and a loss function based on a sliding window mechanism through a gradient optimization method; And the output module is used for applying the parameter set obtained according to the trained neural network to each window of sliding window decoding so as to decode the SC-LDPC code and output a decoding result. In a third aspect, the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communicat