CN114785356A

CN114785356A - Polar code list continuous elimination decoding method and device for reducing path splitting

Info

Publication number: CN114785356A
Application number: CN202210541237.5A
Authority: CN
Inventors: 王中风; 陈弘炜; 周杨灿; 汪煜凯; 吴凯; 郑耀杰
Original assignee: Nanjing University
Current assignee: Nanjing University
Priority date: 2022-05-18
Filing date: 2022-05-18
Publication date: 2022-07-22

Abstract

The invention discloses a polar code list continuous elimination (SCL) decoding method for reducing path splitting, belonging to the technical field of channel coding. The method can reduce the splitting times during SCL decoding while ensuring that the decoding performance is not lost as much as possible, thereby reducing the calculation amount of decoding. The method is characterized in that: finding out the last frozen bit of a code word, selecting y bits from high bits to low bits at the tail part of the x-bit information bit sequence after the frozen bit, adopting an SC decoding method (without splitting), and adopting an SCL decoding method (splitting) for other bits. The decoding mode reduces the SCL decoding splitting times without obvious performance reduction, so that the method has wide application prospect.

Description

Method and device for continuously eliminating and decoding polar code list for reducing path splitting

Technical Field

The invention relates to the technical field of communication and electronic information, in particular to a method for continuously eliminating and decoding a polar code list by reducing path splitting.

Background

Polar Code was proposed in 2008 by professor Erdal Arikan university of turkish bicken, the codeword is the only theoretically proven encoding method invented so far and has a low complexity of encoding and decoding, and when the encoding length is N, the complexity is O (nlog (N)). Polar code technology is based on the phenomenon of channel polarization: as the number of combined channels approaches infinity, one portion of the channels will tend to be noise-free channels and another portion of the channels will tend to be full-noise channels. At this time, useful information is transmitted by using a noiseless channel, and appointed information is transmitted by using a full-noise channel, so that the Shannon limit can be reached.

At present, a method for separating a noiseless channel and a full-noise channel in the process of encoding and decoding by using a common polarization code is to set a frozen bit in a content to be transmitted as information agreed by two communication parties. During decoding, the frozen bits may serve as a check means, and LLR (log likelihood ratio) calculated from the received signal is combined to deduce the most likely value of the information bits.

Serial Cancellation decoding (SC) is a classic decoding method for polar codes. The Serial Cancellation List (SCL) decoding developed on this basis is the most widely applied decoding method at present. SC is superior to SCL in each case. In the SC decoding process, the decoding of the information bits only depends on the calculated LLR value, and each information bit only keeps one result, so the complexity of the decoding process is relatively low, the decoding speed is high, but the decoding accuracy of the high-order information bit has strong dependence on the decoding accuracy of the low-order information bit; in the SCL decoding process, when the number of the decoders is enough, two results of 0 and 1 are reserved for each information bit and are copied to the two decoders, and finally the reservation of the results is determined according to the path metric value PM of each decoder. Therefore, the complexity of the decoding process is relatively high, the decoding speed is relatively slow, but the decoding result is processed more globally, and the decoding accuracy is higher compared with the SC decoding.

The advantages and disadvantages of the conventional classical SC and SCL decoding are complementary, and if only the conventional SC or SCL decoding method is adopted, the lower complexity and the higher decoding accuracy in the decoding process are difficult to achieve. The invention realizes the effect of hardly generating decoding performance loss while reducing complexity by simultaneously adopting two decoding processes in one decoding process through carrying out segmentation operation on one code word.

Disclosure of Invention

Aiming at the technical problems of multiple path splitting times and high complexity of the traditional SCL, the invention provides a decoding algorithm for continuously eliminating a polar code list for reducing the path splitting times, and achieves the effects of reducing the path splitting times and the complexity and hardly causing the loss of decoding performance.

The invention comprises the following steps:

1. for a codeword with code length n, without CRC check:

(1) finding the highest frozen bit of the code word to be decoded, and dividing the code word to obtain a part containing information bits and the frozen bit and the rest part containing only the information bits, wherein the length of the latter part is set as x;

(2) continuously selecting y-bit information bits (no more than x bits) from the most significant bit of the x-bit information bits to the least significant bit;

(3) before decoding to the (n-y) th bit, SCL decoding method is adopted, and when a new information bit is encountered, the decoding results are kept to be 0 and 1 at the same time, i.e. the decoding path splitting is carried out. And if the frozen bit is encountered, only the default value of the frozen bit is reserved, and the path splitting is not carried out. Assuming that the list size is L, when the number of split paths is greater than L, only the L paths with the smallest PM value are reserved until the (n-y) th bit is encountered. At this time, a decoding path with the smallest PM is selected as a result of the SCL decoding section.

(4) For the last y-bit information bit decoded by SC, the decoding process directly calculates the LLR of each information bit according to the LLR value received by the input end and the decoding result of the SCL decoding part, and finally determines the value of the information bit according to the positive and negative LLRs of the information bit.

2. For codewords with CRC check added:

(1) before decoding to (n-y) bits, the decoding process is the same as the SCL decoding partial decoding process without CRC check. Except that after the (n-y) th bit is decoded, the L decoding paths with the smallest PM value are reserved.

(2) And carrying out SC decoding on each decoding path to obtain L decoding results.

(3) It is checked whether each decoding result passes the CRC check.

(4) If the decoding result passing the CRC check exists, the decoding result with the minimum PM value is selected as the final decoding result. Otherwise, the PM minimum is directly selected from the L decoding results as the final result.

According to the inventive concept, the following technical scheme can be obtained: a decoding device for eliminating the polar code list continuity reducing the path split uses the decoding method according to any one of the above technical solutions.

Drawings

Fig. 1 shows a codeword divided into SCL and SC decoding parts. Where i represents an information bit and f represents a freeze bit.

Fig. 2 is a coding tree structure of the SCL decoding part. Wherein PM is_i，jThe path metric value corresponding to the fact that the ith decoder has finished decoding the jth information bit. u. of_kAre estimated values.

When the representative code length is N, the m-th decoder calculates the decoding output end LLR value corresponding to the bit with the index of k.

Fig. 3 shows the decoding process of information bits without CRC check (the number of bits not split is y).

Fig. 4 shows a decoding process with CRC check on the information bits (the number of bits not split is y).

Fig. 5 is a simulation result. Wherein, the SCL _ NEW graph represents the simulation result of the decoding method related by the invention, and the SCL graph represents the simulation result of the traditional SCL decoding method.

Detailed Description

The technical scheme of the invention is explained in detail by combining the drawings and the embodiment.

Code words with code length n-512 each contain 208 information bits without using CRC. The index of the last frozen bit (index range 0-511) is 504, followed by 7 bits of information. Taking y to 4, the split stops at index 508. Assuming that the maximum number of paths L is 16, the specific steps of successive erasure decoding of the polar code list for reducing the number of path splits (as shown in fig. 3) are as follows:

the first step is as follows: the receiving end receives 512 LLRs.

The second step: and utilizing an SCL decoding tree to split two paths at the information bit, respectively corresponding to two results of which the information bit estimated value is 0 or 1, calculating a path metric value PM of each path by combining a decoding output end LLR calculated by each decoder, and keeping the L with the minimum PM as 16 paths. This step is repeated until the decoding of the bit with index 507 is finished.

The third step: and reserving the decoding path with the minimum PM, and using the decoding result for the subsequent SC decoding.

The fourth step: starting from the information bit with index 508, the estimated value of each information bit depends only on the calculated sign of the decoding output LLR, i.e. SC decoding is performed. This step is repeated until the decoding is finished.

Each code word has 208 information bits, including 16 CRC check bits, and has a code length of 512. The index of the last frozen bit (index range 0-511) is 504, followed by 7 bits of information. Taking y to 4, the split stops at index 508. Assuming that the maximum number of paths L is 16, the specific steps of successive erasure decoding of the polar code list for reducing the number of path splits (as shown in fig. 4) are as follows:

the first step is as follows: the receiving end receives 512 LLRs.

The second step is that: and splitting two paths at the information bit by using an SCL (service class hierarchy) decoding tree, respectively corresponding to two results of which the information bit estimation value is 0 or 1, calculating a path metric value PM of each path by combining a decoding output end LLR (likelihood ratio) calculated by each decoder, and keeping L (maximum ratio) of the PM minimum to 16 paths. This step is repeated until the decoding of the bit with index 507 is finished.

The third step: and reserving all decoding paths, and respectively using the decoding result of each decoding path for the subsequent SC decoding.

The fourth step: starting from the information bit with index 508, the estimated value of each information bit depends only on the calculated sign of the decoding output LLR, i.e. SC decoding is performed. Repeating the steps until the decoding is finished, and finally obtaining the decoding result of the L-16 groups.

The fifth step: and respectively performing CRC on the 16 groups of decoding results.

And a sixth step: and judging whether a decoding result passing the CRC check exists or not. If the decoding results of the array passing the CRC check exist, selecting one result with the minimum PM value of the corresponding decoding path from the results as the final result of the decoding. And if all the decoding results do not pass the CRC, selecting one decoding result with the minimum PM value of the corresponding decoding path as a final result.

In the execution process of the decoding method, the number of splitting operations of the numerical value of the decoder is (I-y) (I is the number of information bits), which is less than the number of splitting operations of I times in the traditional SCL decoding process. Therefore, compared with the traditional SCL decoding process, the decoding method reduces the splitting times and reduces the complexity. The simulation is performed on four cases of 512-bit code length, 192-bit information + 16-bit CRC check, and maximum decoding path number L being 2, 4, 8, and 16, and the result of comparing the decoding method of the present invention with the conventional SCL decoding method is shown in fig. 5(a) (b) (c) (d). It can be seen from the figure that the decoding method related to the present invention is very close to the Frame Error Rate (FER) curve of the conventional SCL decoding method. Therefore, the decoding method achieves the effects of reducing the splitting times and complexity and hardly losing the decoding accuracy.

Claims

1. A method for decoding a list of polar codes with reduced path splitting by successive cancellation, the method comprising: for a polar code, the code word is divided into two parts, which contain both frozen bits and information bits and only information bits, with the last frozen bit as a boundary. In the part only containing information bits, the highest bit (last bit) to the lower bit selects continuous y bits, and decoding is carried out by adopting an SC (non-split) mode; and for the rest part of the code word, decoding in an SCL (normally split) mode.

2. The method for successive cancellation of lists of polar codes with reduced path splitting as recited in claim 1, wherein: if the information bit of the code word to be decoded does not carry out CRC encoding, only one path with the minimum path metric value (PM) is reserved when the decoding is carried out to the first information bit which is not split, and other paths are discarded. In the only path reserved, SC decoding is applied to the remaining information bits.

3. The method for successive cancellation of lists of polar codes with reduced path splitting as recited in claim 1, wherein: if the information bit is added with CRC, all split paths are reserved when the decoding is carried out to the first information bit which is not split. Thereafter, SC decoding is performed for each decoding path.

4. The method for successive cancellation of lists of polar codes for reducing path splitting as recited in claim 1, wherein: for the code word of which the information bit is added with CRC, if a decoding path passing through CRC check exists after decoding is finished, selecting the path with the minimum PM from the paths as a decoding result; if all paths do not pass CRC, directly selecting the path with the minimum PM from all decoding paths as a decoding result.

5. A polar code list successive elimination decoding apparatus for reducing path splitting, comprising: selecting continuous y-bit information bits from the highest bit (the last bit) to the lower bit of the polarization code, and adopting SC (non-splitting) decoding; and for the rest part of the code word, decoding in an SCL (normally split) mode. If the decoded polar code does not contain CRC codes, only one path with the minimum PM is reserved when the decoding is carried out to the first information bit without splitting, and the SC decoding is continued after other paths are discarded. If CRC is added to the decoded polarization code, when decoding is carried out to the first information bit which is not split, all split paths are reserved, then SC decoding is carried out on each decoding path, and if the decoding path passing the CRC check exists after decoding is finished, the path with the minimum PM is selected from the paths as a decoding result; if all paths do not pass CRC, directly selecting the path with the minimum PM from all decoding paths as a decoding result.