CN110719112A

CN110719112A - Deep learning-based parameter adaptive RS code decoding method

Info

Publication number: CN110719112A
Application number: CN201910861946.XA
Authority: CN
Inventors: 梁煜; 安翔宇; 张为
Original assignee: Tianjin University
Current assignee: Tianjin University
Priority date: 2019-09-12
Filing date: 2019-09-12
Publication date: 2020-01-21
Anticipated expiration: 2039-09-12
Also published as: CN110719112B

Abstract

The invention relates to a parameter self-adaptive RS code decoding method based on deep learning, which comprises the following steps: carrying out RS code coding on the information sequence X to obtain a coded information sequence u; carrying out BPSK modulation on the information sequence u to obtain a transmission sequence s; the transmission sequence passes through a Gaussian white noise channel to obtain a receiving sequence y; constructing a reliability matrix by using the receiving sequence y, and calculating to obtain a reliability sequence R; dividing the reliability sequence R into a training set and a testing set; training a neural network; respectively testing the training set of R and the corresponding receiving sequence y by using LCC decoding algorithms with different eta values to find eta optimal solutions corresponding to the training set receiving sequence y one by one; and taking the SNR and the reliability sequence R as input, taking the minimum eta value obtained by testing as a classification label, and forming a data set for neural network training by the SNR and the reliability sequence R together.

Description

Deep learning-based parameter adaptive RS code decoding method

Technical Field

The invention belongs to the field of error control coding in channel coding, and relates to a parameter self-adaptive decoding algorithm based on deep learning.

Background

Error control is the basic method for controlling and correcting transmission errors caused by channel noise in digital communication by using coding and decoding technology to ensure the effectiveness and reliability of communication. The error control coding has wide application in the digital communication field and has important research significance. The Reed-Solomon (RS) code is one of error control codes, is constructed by Reed and Solomon in 1960, has simple structure, short decoding delay and rich structural characteristics, and can correct random errors and burst errors. In particular, in applications for correcting burst errors, RS codes perform better than other error control codes. Currently, RS codes have been widely applied to the fields of wireless communication, optical fiber communication, magnetic disk, solid-state storage, and the like, and are always hot spots of research in the industry and academia.

Depending on the form of the demodulator output symbols, the decoding algorithms of RS codes can be divided into two categories: a hard decision decoding algorithm and a soft decision decoding algorithm. The hard decision decoding algorithm is relatively simple and easy to implement by hardware, but the algorithm does not fully utilize channel soft information, and the decoding performance is lost. The soft-decision decoding is used for increasing decoding gain at the cost of complexity, and mainly comprises KV, BGMD and LCC algorithms. The LCC algorithm is superior to the first two in complexity and calculation period. Meanwhile, the decoding performance and complexity of the algorithm are closely related to the parameter eta of the repeated number distribution process.

In recent years, deep learning has attracted worldwide attention because of its powerful ability to solve complex tasks. The deep learning method is also applied to the fields of speech recognition, computer vision, machine translation and the like, and brings great innovation and breakthrough to the fields. Some research results are obtained by the scheme combining the deep learning and coding fields, and deep research is carried out in the fields of polar codes, BCH codes, LDPC codes and the like. Most current research considers decoding as a classification problem, and uses neural network structures (such as DNN, CNN, RNN) to realize a complete decoding process. The obvious disadvantage of this kind of scheme is that the neural network has very poor prediction capability for the unbeared codewords (codewords not included in the training set), and the performance of the decoder is greatly affected by the exponential explosion growth of the codewords possibly caused by the increase of the codeword length. In addition, some researches propose that a deep learning auxiliary traditional decoding algorithm is used for eliminating related noise, predicting key parameters and the like so as to improve the performance of the traditional algorithm.

Disclosure of Invention

The invention aims to provide a parameter self-adaptive decoding method based on deep learning, which can reduce complexity and improve efficiency. The technical scheme is as follows:

a parameter self-adaptive RS code decoding method based on deep learning comprises the following steps:

(1) and carrying out RS code coding on the information sequence X to obtain a coded information sequence u. And carrying out BPSK modulation on the information sequence u to obtain a transmission sequence s. And the transmission sequence passes through a Gaussian white noise channel to obtain a receiving sequence y. And then constructing a reliability matrix by using the receiving sequence y, and calculating to obtain a reliability sequence R. The reliability sequence R is divided into a training set and a testing set.

(2) And training the neural network. And respectively testing the training set of the R and the corresponding receiving sequence y by using LCC decoding algorithms with different eta values, and finding out eta optimal solutions corresponding to the receiving sequence y of the training set one by one. And taking the SNR and the reliability sequence R as input, taking the minimum eta value obtained by testing as a classification label, and forming a data set for neural network training by the SNR and the reliability sequence R together. The neural network structure uses an MLP structure, weight w and bias b of the network are continuously optimized through a back propagation algorithm, the neural network with high classification accuracy is obtained under the conditions of enough data sets and training algebra, and the training process is completed.

(3) And inputting by using a test set of SNR and R, and calculating by using the trained neural network to obtain an eta optimal solution. And then, LCC decoding is carried out by using the eta value after parameter self-adaptation to obtain a decoded sequence X', and finally decoding is finished.

Drawings

FIG. 1 is a system block diagram of a deep learning-based parameter adaptive decoding algorithm

FIG. 2 training process for neural networks

Detailed Description

The invention provides a parameter self-adaptive decoding algorithm based on deep learning, which is an algorithm for assisting the LCC decoding of the traditional RS code by using the deep learning. According to analysis, a key parameter eta in the LCC algorithm is closely related to channel environment and reliability information, and the decoding performance and complexity of the LCC algorithm are determined by the size of the eta value. The problem of solving the eta optimal solution can be regarded as a simple classification problem, and the neural network has strong mode classification capability and can simply and efficiently select the eta optimal solution. The key parameter eta in the LCC algorithm is adaptively changed along with the channel environment and the reliability information of the received sequence by a deep learning method, and the decoding efficiency can be improved on the premise of not influencing the decoding performance.

The parameter adaptive decoding algorithm based on deep learning specifically comprises two stages, namely a training stage and a decoding stage.

(1) The training phase implements the process of building and training a neural network. Based on supervised learning, the invention constructs a multilayer perceptron (MLP) neural network structure, and the number of nodes and the number of layers are determined by the code length. In the aspect of a training data set, a signal-to-noise ratio and a likelihood ratio are connected as input, and a corresponding eta optimal solution obtained by a traditional decoding algorithm test is used as a classification label. After training and debugging, the neural network with excellent classification accuracy is obtained.

(2) And in the decoding stage, the signal-to-noise ratio and the likelihood ratio of the specific code word are input into the trained neural network to obtain an eta optimal solution. And then carrying out traditional LCC decoding according to the optimal solution.

The technical scheme of the invention is further specifically described below with reference to the accompanying drawings.

As shown in fig. 1, the parameter adaptive decoding algorithm based on deep learning specifically includes the following steps:

(4) and carrying out RS code coding on the information sequence X to obtain a coded information sequence u. In order to convert the encoded information into a mode suitable for transmission in the channel, BPSK modulation is performed on the information sequence u to obtain a transmission sequence s. The transmission sequence passes through a white gaussian noise channel (AWGN) to obtain a reception sequence y (y is s + n, and n is AWGN channel noise). And then constructing a reliability matrix by using the receiving sequence y, and calculating to obtain a reliability sequence R. And dividing the reliability sequence R into a training set and a testing set, wherein the ratio of the training set to the testing set is 9: 1.

(5) As in fig. 2, training of the neural network is performed. And respectively testing the training set of the R and the corresponding receiving sequence y by using LCC decoding algorithms with different eta values, and finding out eta optimal solutions (minimum values under the condition of correct decoding) corresponding to the training set receiving sequence y one by one. And taking the SNR and the reliability sequence R as input, taking the minimum eta value obtained by testing as a classification label, and forming a data set for neural network training by the SNR and the reliability sequence R together. The neural network structure uses a multilayer perceptron (MLP) structure, weight w and bias b of the network are continuously optimized through a back propagation algorithm, the neural network with high classification accuracy is obtained under the conditions of enough data sets and training algebra, and the training process is completed.

(6) And inputting by using a test set of SNR and R, and calculating by using the trained neural network to obtain an eta optimal solution. And then, LCC decoding is carried out by using the eta value after parameter self-adaptation to obtain a decoded sequence X', and finally decoding is finished.

The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims

1. A parameter self-adaptive RS code decoding method based on deep learning comprises the following steps:

(1) carrying out RS code coding on the information sequence X to obtain a coded information sequence u; carrying out BPSK modulation on the information sequence u to obtain a transmission sequence s; the transmission sequence passes through a Gaussian white noise channel to obtain a receiving sequence y; constructing a reliability matrix by using the receiving sequence y, and calculating to obtain a reliability sequence R; dividing the reliability sequence R into a training set and a testing set;

(2) training a neural network; respectively testing the training set of R and the corresponding receiving sequence y by using LCC decoding algorithms with different eta values to find eta optimal solutions corresponding to the training set receiving sequence y one by one; taking the SNR and the reliability sequence R as input, taking the minimum eta value obtained by testing as a classification label, and forming a data set for neural network training together with the SNR and the reliability sequence R; the neural network structure uses an MLP structure, weight w and bias b of the network are continuously optimized through a back propagation algorithm, the neural network with high classification accuracy is obtained under the conditions of enough data sets and training algebra, and the training process is completed.