CN111988045A

CN111988045A - Improved polarization code SCF decoder based on genetic algorithm

Info

Publication number: CN111988045A
Application number: CN202010815415.XA
Authority: CN
Inventors: 王秀敏; 马强强; 李君�; 张鸿超
Original assignee: China Jiliang University
Current assignee: China Jiliang University
Priority date: 2020-08-14
Filing date: 2020-08-14
Publication date: 2020-11-24
Anticipated expiration: 2040-08-14
Also published as: CN111988045B

Abstract

The invention provides an improved polarized code serial elimination Flip (SCF) decoder based on Genetic Algorithm (GA). On the basis of an original SCF decoder, aiming at the problem that an original Candidate overturning position Set (CFPS) has redundancy, a new CFPS is constructed by utilizing GA. The initial population of the genetic algorithm is constructed with indices of all non-frozen bits, and the fitness of each individual is taken as the channel reliability calculated with a gaussian approximation. Then the population is subjected to continuous selection, crossover and mutation operations, and the optimal individuals of the population of each generation are preserved. And finally, obtaining a new candidate overturn position set CFPS-GA by counting the frequency of each population in the vector, and carrying out SCF decoding by using the newly constructed candidate overturn position set CFPS-GA. The invention has the beneficial effects that: compared with other similar SCF decoders, the SCF decoder based on the CFPS-GA can have lower computational complexity and decoding delay on the premise of ensuring decoding performance.

Description

Improved polarization code SCF decoder based on genetic algorithm

Technical Field

The invention belongs to the field of channel coding and decoding, and relates to a serial elimination Flip (SCF) decoder of a polarization code and a genetic algorithm in an artificial intelligence technology.

Background

Polarization code since 2009 by

The present invention has been widely focused, and in the newly released 5G communication standard, a polarization code is selected as a coding scheme under the control channel of the eMBB scenario. The polarization code is the only channel coding scheme which is proved to reach the Shannon theoretical limit in theory at present. The most primitive polar code decoder is the Successive Cancellation (SC) decoder, and it is under this decoder that the polar code can reach the shannon theoretical limit. However, the SC decoder is a serial decoder, and if the decoding of the previous bit is incorrect, the decoding of the following bit is affected, which causes an error propagation phenomenon. In order to improve the decoding performance of the SC decoder, a Sequential Cancellation List (SCL) decoder is proposed. Unlike SC decoders, SCL decoders retain up to L decoding paths during decoding, and then select the final path as a decoding result through path metric values or Cyclic Redundancy Check (CRC). The SCL decoder greatly improves the decoding performance of the polar code, so that the polar code has the same decoding performance as other error correcting codes. But, a plurality of decoding paths are reserved when the SCL decoder decodes, and the decoding characteristic enables the SCL decoder to have higher decoding time delay and higher calculation complexity.

In order to improve the decoding performance of the SC decoder and reduce the complexity of the SCL decoder, the scholars propose a decoder named SCF. In the SCF decoder, SC decoding is performed first. The results from the SC decoder are then CRC checked. And when the result obtained by SC decoding does not pass CRC, sequencing the absolute values of the log-likelihood ratios corresponding to all the non-frozen bits, then overturning the bit with the minimum absolute value of the log-likelihood ratio, and continuously decoding the code word after the overturning bit by using an SC decoder. T is_maxIndicating the maximum number of such attempts. In actual communication, errors caused by channel noise are few, but most of the errors are caused by errors caused by channel noise. SCF decoders find these through constant flip attemptsErrors caused by channel noise further improve the decoding performance of the polar code. Although the SCF decoder achieves a balance between decoding performance and decoding complexity as compared with the SC and SCL decoders, in the above-mentioned SCF decoding algorithm, the Candidate Flip Position Set (CFPS) is composed of indexes of all non-frozen bits. However, it has been found experimentally that some subchannels are very reliable and no flip-over attempts are required for decoding the bits of these subchannels. Then, in order to reduce the size of the CFPS and the search space of the candidate flip position, some researchers have proposed using a Critical Set (CS) as the CFPS of the SCF decoder. The CS acquisition method is to find out all rate-1 nodes in the polar code tree and then take the index corresponding to the first non-frozen bit of the rate-1 node as the element in the CS. The SCF decoder based on the CS effectively improves the decoding performance of the SCF decoder and reduces the computational complexity of the original SCF decoder. Meanwhile, some scholars also put forward a concept of a Middle-level subchannel (MBC) set by using channel characteristics. The Monte Carlo simulation result shows that the SCF decoder based on the MBC set has the same decoding performance and computational complexity as the SCF decoder based on the CS.

Disclosure of Invention

The invention provides an improved polarization code SCF decoder based on a genetic algorithm, aiming at the problem that a candidate turning position set of the polarization code SCF decoder has redundancy. The SCF decoding module of the polar code consists of a standard SC decoding module, a CRC checking module and a candidate flip bit structure modeling module. And the decoder adopts a new CFPS construction method based on genetic algorithm in a candidate flip bit construction module. In an implementation of the present invention, the index of each non-frozen bit is taken as an individual of the genetic algorithm, and then the initial population of the genetic algorithm is constructed with the indices of all non-frozen bits. Because the channel environment of the invention is additive white Gaussian noise, the invention takes the channel reliability calculated by Gaussian approximation as the fitness of each individual. The selection strategy used is a tournament selection. And carrying out binary coding on the new population after the selection operation, then carrying out cross and variation operation on the newly generated population, and storing the optimal individuals of each generation. When the iteration is over, the frequency of occurrence of each non-frozen bit index is counted, and then a new GA-based CFPS is constructed with those indices having a frequency other than 0. The specific steps of the construction are as follows:

the method comprises the following steps: and initializing the population by using indexes of all non-frozen bits, and calculating the fitness of each individual by using Gaussian approximation.

Step two: and selecting the individuals by using a championship selection strategy to obtain a new population after selection.

Step three: and carrying out binary coding on the selected population, then carrying out cross and mutation operations on the individuals through a single-point cross and simple mutation strategy, and storing the optimal individuals in the new population subjected to the mutation operations in a vector path.

Step four: and repeating the second step and the third step until the iteration is finished, and then obtaining a final vector path, wherein the size of the vector is the same as the iteration times of the population.

Step five: the frequency with which each non-frozen bit index appears in the vector path is counted and then the index with a frequency other than 0 is put into the new CFPS.

The newly constructed CFPS is utilized to carry out SCF decoding, and experimental simulation shows that the SCF decoder based on the CFPS-GA can reach the upper limit of single-bit upset decoding performance. And compared with other similar SCF decoders, the SCF decoder based on the CFPS-GA has lower computational complexity and decoding time delay.

Drawings

Fig. 1 is a block diagram of an SCF decoder system of the present invention;

FIG. 2 is a flow chart of the CFPS-GA construction process of the present invention;

FIG. 3 is a schematic diagram illustrating a variant operation of the present invention;

FIG. 4 shows the average decoding complexity of different decoders at 1024 code lengths;

fig. 5 shows the average decoding delay of different decoders at 1024 code length.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings, and the following examples are provided to facilitate understanding of the present invention and are intended to be a better understanding of the present invention and are not to be construed as limiting the present invention.

In fig. 1, the system block diagram mainly consists of a polar code encoding module, a channel module and a polar code SCF decoding module. In addition, the polarization code SCF decoding module consists of a standard SC decoding module, a CRC checking module and a candidate flip bit structure modeling module based on a genetic algorithm. The channel module adopts additive white gaussian noise. The CRC check bit length in the CRC check module is 16, and the CRC generator polynomial is g (x) x¹⁶+x¹⁵+x²+1. The candidate turning position construction module adopts a genetic algorithm technology belonging to the field of artificial intelligence. As shown in fig. 2, since the information transmitted by the frozen bits is known to both the transmitting end and the receiving end, flipping these bits when SC decoding is in error does not improve the decoding performance, so the index of the non-frozen bits is used for population initialization. Since the binary encoding starts from 0 and the index starts from 1, when the individual is binary encoded, the value corresponding to the binary encoding is 1 less than the decimal value. And whether to perform crossover operations on parent individuals is determined by the crossover rate. Because new indices may be generated by interleaving operations, which may be frozen bit indices, the generated indices need to be examined and only the children of the indices that correspond to non-frozen bits can be considered valid individuals. Fig. 3 is a schematic diagram of a mutation operation in which an individual 514 of the population is first binary coded into its corresponding binary vector 100000001. flag1 denotes a randomly generated random number, p, ranging from 0 to 1_mThe variation rate is shown. As can be seen from fig. 3, the binary vector can be mutated from 1000000001 to 1000001010 by a mutation operation.

Based on the construction flow chart of fig. 2, the code length is 1024, the code rate is 0.5, and the signal-to-noise ratio is 2.5 dB. The population size and the number of non-frozen bits are the same. Because the original SCF decoding requires ordering of the log-likelihood ratios for all the non-frozen bits, its CFPS size is equal to the number of non-frozen bit bits. Under the same condition, the CFPS-GA constructed by the genetic algorithm has a smaller set, so that the decoding complexity and the decoding time delay of the SCF decoder improved based on the genetic algorithm can be effectively reduced.

In a simulation experiment, the code length is 1024, the code rate is 0.5, and a control variable method is adopted in simulation conditions. The SCF-GA is used for representing an SCF decoder of the CFPS-GA constructed based on a genetic algorithm, and the SCF-CS is used for representing an SCF decoder based on a key set. And SCO1 represents the upper decoding performance limit of a single bit-flipped decoder. Simulation experiment results show that the SCF-GA decoder and the SCO1 decoder have almost the same decoding performance under all signal-to-noise ratios. And when the signal-to-noise ratio is less than 2dB, the SCF-GA decoder has the same decoding performance as the SCF-CS and the original SCF decoder. But when the signal-to-noise ratio is larger than 2dB, the SCF-GA decoder can obtain better decoding performance. Compared with the original SCF decoder, the SCF-CS has the frame error rate of 10^-3The SCF-GA decoder has a performance gain of about 0.1 dB. In fig. 4, the present invention represents the average normalized computational complexity with an additional number of flips. As the signal-to-noise ratio increases, the complexity of various SCF decoders quickly drops to the same level as SC decoders. Since the CFPS-GA constructed by the present invention contains fewer non-frozen bit indices, the search space is smaller when deciding the flip position. It can be seen from fig. 5 that the SCF-GA decoder has a lower decoding delay over the entire signal-to-noise ratio range.

The above embodiments do not limit the present invention in any way, and all technical solutions obtained by using similar structures, methods and similar variations of the present invention are within the scope of the present invention.

Claims

1. An improved SCF decoder based on a genetic algorithm is characterized in that a polarization code SCF decoding module consists of a standard SC decoding module, a CRC (cyclic redundancy check) module and a candidate flip bit structure modeling module based on the genetic algorithm. The invention is characterized in that a candidate inversion bit construction module, which is different from the traditional construction method based on the log likelihood ratio, and the invention constructs the candidate inversion bit by utilizing the genetic algorithm, and comprises the following steps:

the first step is that indexes of all non-frozen positions are utilized to form an initial population of the genetic algorithm;

calculating the fitness of each individual by utilizing Gaussian approximation, and then performing selection operation on the individual, wherein the selection strategy is selected for championship match selection;

carrying out binary coding on the selected individuals, then carrying out cross and variation operations on the individuals, and storing the optimal individuals of each generation;

and fourthly, counting the times of taking each frozen bit index as the optimal individual of each generation, and forming a new candidate position set by using the index with the non-0 counting times as a constructed final result.