CN106603084B - A preprocessing method for punctured LDPC hard-decision decoding - Google Patents

Abstract

The present invention provides a preprocessing method for puncturing LDPC hard-decision decoding. First, the position of the coded puncturing check information is determined according to the coding puncturing algorithm; then, the generator matrix G is found according to the puncturing position. Position correlation unit; then use the received information, perform precoding processing according to the LDPC coding algorithm, and complete the check information lost by punching; finally, perform LDPC decoding work on the completed received information according to the hard decision algorithm to achieve coding Error correction performance. The invention consumes less hardware resources and is easy to implement in engineering; for the case of punching holes, it can provide a decoding gain of about 0.5dB, especially in the case of many punching positions and poor channel conditions, the error correction performance is better. Significantly.

Description

A preprocessing method for punctured LDPC hard-decision decoding

technical field

The invention belongs to the field of channel coding, and relates to a preprocessing method for puncturing LDPC hard decision decoding.

Background technique

Low-density parity-check codes (LDPC) have shown the error correction performance closest to the Shannon limit so far in communication systems, and have been widely used in optical communications, satellite communications, deep space communications, fourth-generation mobile communications, high-speed and very high-speed Rate digital subscriber lines and magnetic recording systems. But compared with other error correction codes, the encoding and decoding of LDPC codes usually requires a lot of hardware resource consumption and processing time, which greatly limits its practical application. In recent years, research efforts have mainly focused on reducing the complexity of coding and decoding, applying embedded microsystems, and improving the running speed of electronic systems. However, limitations at the hardware level make a large number of researchers pay more attention to the improvement or replacement of existing algorithms. It is a wise and reasonable choice to reduce the complexity of hardware implementation by improving or replacing the algorithm. The improved algorithm can reduce the hardware requirements to a certain extent, so as to achieve the purpose of reducing cost and being practical.

In LDPC decoding, it is mainly divided into two categories: hard-decision decoding and soft-decision decoding. Hard-decision decoding can realize decoding without any soft information. It has a simple structure, a small amount of computation, and is easy to implement in engineering. The error correction performance of soft-decision decoding is about 2-3dB lower. Soft-decision decoding has good performance, but requires a large amount of resources and has high implementation complexity.

Puncturing the LDPC code is a method for realizing variable rate LDPC. The idea is to achieve the purpose of adjusting the code rate and code length by deleting some information bits in the check information. In practical applications, it is often encountered that the code length or the code rate is adjusted to match the communication rate. At present, the soft-decision decoding method is mainly used for decoding. However, under the condition of long code length and tight hardware resources, the realization of soft-decision decoding is extremely difficult or even impossible. If the hard-decision LDPC decoding method is adopted, the decoding error correction gain will be greatly lost. , especially in the case of more puncturing information and poor channel conditions, hard-decision LDPC decoding may even lose the ability to decode and correct errors.

SUMMARY OF THE INVENTION

In order to overcome the deficiencies of the prior art, the present invention provides a preprocessing method for puncturing LDPC hard-decision decoding. According to the position of the puncturing information, combined with the generator matrix, precoding processing is performed on the received information, and then decoding is performed. code, and finally get the decoding result.

The technical scheme adopted by the present invention to solve its technical problem comprises the following steps:

(1) According to the coding puncturing algorithm, confirm the positions K1, K2...Kn of the puncturing check information after coding, and find out the coding information C, that is, the information bits, from the received information;

(2) Transform the generator matrix G according to the check matrix H, and find out the Kith column Gi in the generator matrix G corresponding to the puncturing information Ki, where the correspondence refers to the Kith information bit and the generator matrix G Ki columns correspond to each other;

(3) sum the encoded information units C in the received codewords corresponding to the non-zero elements in the Kith column of all generating matrices, and modulo 2 to obtain the precoded check information Pki; use the encoded information C to do partial re-encoding processing to Restore the verification information of the punched part;

(4) repeat steps (2), (3), traverse each value of i, respectively obtain all punching position verification information recovery values Pki;

(5) use the obtained Pki to complete the Ki-th puncturing check information in the received information respectively, and recover the complete codeword S';

(6) Take S' as the new received encoded information, and use the LDPC hard-decision decoding algorithm to decode the decoded information.

The beneficial effects of the present invention are:

(1) The preprocessing method proposed by the present invention performs partial re-encoding processing on the received information before decoding, and restores the information of the punched part, which provides more credible information for the decoding processing. The method of decision decoding can have a decoding gain of about 0.5dB, especially when the hole is punched and the channel conditions are harsh, the effect is more obvious;

(2) The preprocessing method proposed by the present invention only re-encodes the part related to the puncturing information. Since the H matrix and the generator matrix G of the LDPC code are both sparse matrices, that is, most of the units in the matrix are 0, and a small number of units are It is 1. According to the matrix multiplication operation rule, it is only necessary to take the information of the non-zero units in the G-related units, simplify the operation, and only need to increase a small amount of hardware resources. Compared with the application of the soft-decision decoding algorithm, the consumption of hardware resources is Much less, more in line with practical application and cost considerations.

Description of drawings

Fig. 1 is the processing flow chart that the present invention carries out preprocessing method;

Fig. 2 is the application process diagram of the pretreatment method of the present invention;

FIG. 3 is a simulation comparison diagram of preprocessing performance in a specific embodiment.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and embodiments, and the present invention includes but is not limited to the following embodiments.

The technical solution adopted by the present invention to solve the technical problem is to perform precoding processing on the received information in combination with the generator matrix according to the position of the punching information, and then decode it, and finally obtain the decoding result, which specifically includes the following steps:

(1) Initialization

According to the coding puncturing algorithm, confirm the positions K1, K2...Kn of the puncturing check information after coding, and find the coded information C, that is, the information bits, from the received information.

(2) Determine the corresponding generator matrix G unit

According to the transformation of the check matrix H, the generator matrix G is obtained, and the Ki-th column Gi in the generator matrix G corresponding to the puncturing information Ki is found, wherein the correspondence refers to the mutual relationship between the Ki-th information bit and the Ki-th column of the generator matrix G. for the corresponding relationship.

(3) Precoding processing

The coded information elements C in the received codewords corresponding to the non-"0" elements in the Ki-th column of the generator matrix are summed modulo 2 to obtain the precoded check information Pki. According to the coding principle S=C×G, where S is the code word after coding, C is the coding information in the received information, and G is the generator matrix. Since the received codeword S is formed by splicing the encoded information C and the verification information J, the encoded information C in the received codeword is used to perform partial re-encoding processing to restore the verification information of the punctured part, that is, precoding processing.

(4) Repeat steps (2) and (3) to obtain all recovery values Pki

Accumulate i from 1 to n, repeat steps (2) and (3), and obtain the recovery value Pki of all punching position verification information respectively.

(5) Completing the punching position verification information and restoring the code word S'

Use the obtained Pki to complete the Ki-th puncturing check information in the codeword respectively, and recover the complete codeword S', that is, the received information + the puncturing information recovery value.

(6) LDPC hard decision decoding

Take S' as the new received encoded information, and then use the LDPC hard-decision decoding algorithm to decode the decoding information according to the hard-decision decoding steps.

The specific implementation of the preprocessing method of the present invention will be described below by taking a (1250, 2500) quasi-cyclic LDPC code with a code rate of 1/2 as an example. Since the (1250,2500) LDPC code is not a code with a relatively regular code length, in the actual system, the communication rate is 62.5kbps, plus the CRC check after encoding, in order to match the communication rate, the LDPC is punctured. , in this case, the specific steps of applying the preprocessing of the present invention to carry out decoding processing are as follows:

Step 1 Initially determine the punching position and coding information

In order to meet the requirements of rate matching, after the system completes the encoding, the last 150bits of the parity bits are punctured. Since the (1250,2500) LDPC code is composed of a 50bits*50bits cyclic code, 50bits is used as a unit for processing. , we select the 23rd, 24th, and 25th puncturing units, and group the received data, every 50 bits is a group, and the first 25 groups are the received encoded information C1, C2, ... C25.

Step 2 Determine 3 groups of punched corresponding generator matrix G unit groups

The G matrix corresponding to the H matrix is calculated by Matlab software, and the unit G23 of the G matrix corresponding to the 23rd punching unit is found. In general, the generator matrix G needs to be obtained to perform the subsequent steps, but for some special LDPC codes, the G matrix is easy to obtain, such as quasi-cyclic LDPC (QC-LDPC) codes, lower triangular structure LDPC codes, Bidiagonal structure LDPC codes, especially some LDPC codes that directly construct G matrix, etc. For codes that are not easy to obtain from the G matrix, you can refer to the coding part to borrow the G matrix generated by the coding, which can save a lot of time and effort.

Step 3: Perform precoding processing on the 23rd puncturing related information to obtain a restored value P23.

According to the coding principle S=C×G, where S is the code word after coding, C is the coding information in the received information, and G is the generator matrix. Since the received codeword s is formed by splicing the encoded information C and the verification information J, the encoded information C in the received codeword s is used to perform partial re-encoding processing to restore the verification information of the punctured part, that is, precoding processing.

In the G matrix, among the matrix units in the 23rd column, only the 5th, 11th, 17th, and 25th matrix units are non-zero units, and the cyclic shift times are: 17, 18, 5, and 34, respectively. Perform 17, 18, 5, and 34 right cyclic shift operations on the encoded information units C5, C11, C17, and C25 in the received codeword, respectively, and perform 17, 18, 5, and 34 cyclic shift operations to the right, and the shifted encoded information C'5, C'11 , C'17, C'25 do modulo 2 sum operation, and get the recovery value P23 of the 23rd punching unit.

Step 4 Obtain the recovery values P24 and P25 according to steps (2) and (3) respectively

According to step (2), take the 24th and 25th columns of G matrix units G24 and G25 respectively, wherein the 6th, 11th, 20th, and 21st in G24 are non-zero units, and their cyclic shift times are: 25, 13, 1, and 9 respectively. ; Among them, the 7th, 12th, 18th, and 25th in G25 are non-zero units, and their cyclic shift times are: 28, 29, 43, and 14, respectively. According to step (3), modulo 2 sum processing is performed on the corresponding coded information after shifting, respectively, to obtain the restored values P24 and P25 of the 24th and 25th single-hole units.

Step 5 Restore 2500bits codeword S'

The codewords received and analyzed have a total of 2350 bits, and the P23, P24, and P25 obtained in step (4) are added to the back of the received codeword, that is, the positions of the 23rd, 24th, and 25th punching units, and the codeword S' of 2500 bits is recovered. .

Step 6 Weighted Bit Flip Hard Decision Algorithm Decoding

The weighted bit-flip hard-decision decoding algorithm is applied, and the maximum number of iterations is selected as 20, and the recovered codeword S' is decoded to obtain the decoding result C'.

Taking the (1250,2500) quasi-cyclic puncturing LDPC code with a code rate of 1/2 as an example, applying the preprocessing method of the patented invention, the performance analysis of the decoding gain obtained by software simulation and the comparison of hard-decision decoding are shown in Figure 3 , it can be seen that compared with the hard-decision decoding method, the lower the signal-to-noise ratio (Eb/N0) of this preprocessing method, the more the error correction performance is improved. When the signal-to-noise ratio (Eb/N0) is lower than 5.5dB, the preprocessing method can improve the decoding gain by about 0.5-1.0dB.

At the same time, for this kind of code, this preprocessing method is applied to the specific practice and implemented on the CycloneIII EP3C120F84I7FPGA of Altera Corporation. The resource occupancy of soft-decision decoding, preprocessing hard-decision decoding, and hard-decision decoding is shown in Table 1. , in which the soft-decision decoding information bit adopts 6-bit quantization processing, and the hard-decision decoding adopts the weighted bit-flip hard-decision decoding algorithm. It can be seen that the hardware resources required to apply this preprocessing method are about 1/3 of the requirements of the soft-decision decoding method, and only need to increase the hardware resources by less than 1% compared with the unprocessed hard-decision decoding.

Table 1. Comparison of resource occupation of three processing methods

Approach Logic elements Registers Memory bits Preprocessing hard decisions 7078 (6%) 3846 14721 (<1%) soft decision decoding 23596 (20%) 16091 142771 (4%) hard decision decoding 5803 (5%) 3611 10625 (<1%)

Claims (1)

1. a preprocessing method for punching LDPC hard-decision decoding is characterized in that comprising the steps: (1) According to the coding puncturing algorithm, confirm the positions K1, K2, ...Ki, ...Kn, 1≤i≤n of the puncturing check information after coding, and find the coding information C, that is, the information bits, from the received information; (2) Transform to obtain the generator matrix G according to the check matrix H, find out the Ki th column Gi in the generator matrix G corresponding to the position Ki of the puncturing check information, wherein the correspondence refers to the Ki th information bit and the generated The Ki-th column of matrix G corresponds to each other; (3) sum the encoded information units C in the received codewords corresponding to the non-zero elements in the Ki-th column of all generating matrices, and add modulo 2 to obtain the pre-encoded check information Pki; use the encoded information C to do partial re-encoding processing to Restore the verification information of the punched part; (4) repeating steps (2) and (3), traversing each value of i, respectively, obtains the recovery value Pki of all punching position verification information; (5) use the obtained Pki to complete the Ki-th puncturing check information in the received information respectively, and recover the complete codeword S'; (6) Take S' as the new received encoded information, and use the LDPC hard-decision decoding algorithm to decode the decoded information.