5G-LDPC code closed set blind identification method based on BP iteration

Abstract

The invention relates to LDPC (Low Density Parity Check) code closed set blind identification, in particular to a BP (Back Propagation) iteration-based 5G-LDPC code closed set blind identification method. The method mainly comprises the following steps: firstly, respectively operating a punctured column and a filled column related to a matrix to be identified, setting a likelihood value of the punctured column as 0, setting a likelihood value of the filled column as a maximum value, and obtaining likelihood values of other columns through calculation, and then sequentially carrying out BP iterative calculation to obtain updated likelihood values for calculation of blind identification LLR, based on the characteristic that the correct matrix converges and the error matrix does not converge or converges slowly, the correct matrix can be identified. The method solves the problem that the traditional blind recognition method cannot perform recognition during puncturing and filling, and does not transform the original matrix, so that the characteristics of low row weight and the like of the original matrix are reserved, the recognition performance is further ensured, and the recognition efficiency is improved. Therefore, the recognition accuracy is improved when the matrixes of the alternative set are small, the similarity is high, and the number of the received code words is small.

Description

Technical Field

[0001] The present invention belongs to a channel coding blind recognition method, and specifically relates to a 5G-LDPC code closed set recognition method based on BP iteration. Background Art

[0002] In the field of channel coding, blind recognition technology can recover the coding parameters and even the specific codewords used by the sender from the received signal, so it plays an important role in automatic modulation and coding (AMC) and communication countermeasures. Among many coding schemes, LDPC codes are increasingly widely used in practice due to their free and diverse construction methods and excellent performance, but at the same time, they also bring higher blind recognition difficulties. Therefore, the blind recognition technology of LDPC codes has received more and more attention.

[0003] There are two main types of LDPC code blind recognition technology. The first is open set (full blind) blind recognition, that is, the recognizer only has a section of the received signal and has no prior knowledge of the other. At present, the main open set blind recognition methods include rank criterion method and dual vector-based method, which are difficult to recognize and it is difficult to restore the check matrix used by the sender 100%, so it is more widely used in the field of communication confrontation. The other is closed set (semi-blind) blind recognition. In addition to the received signal, the recognizer also knows the candidate set used by the sender, that is, a set consisting of several LDPC codes (such as the protocol used by the known sender, which only stipulates several LDPC codes). At this time, the recognizer returns the most likely candidate. Closed set blind recognition is less difficult, has a higher success rate, and is less complex. The disadvantage is that it requires more prior knowledge, so it is often used in automatic coding modulation technology. At this time, the sender dynamically selects the modulation and coding scheme according to the changes in channel conditions, and the receiver needs to confirm the modulation and coding scheme used by the sender, thereby improving channel utilization efficiency. Closed-set blind recognition mainly uses parity check verification, based on the log likelihood ratio (LLR) obtained from the check relation (SPC), and characterizes the maximum a posteriori probability (APP) from different angles, and selects a codeword with the largest APP. From the initial average log likelihood ratio method, to the average likelihood difference (LD) method that omits arctanh calculation, to the improved method based on cosine transform (CC), to the latest discrimination method based on two-stage APP estimation (two stage, TS), the closed-set recognition accuracy of LDPC codes is getting higher and higher.

[0004] In the field of 5G communications, due to the involvement of puncturing and padding, traditional recognition methods cannot be used. Currently, there is only one recognition method based on linear transformation. Through linear transformation, the punctured columns in the matrix to be recognized are changed to all 0s. At this time, the punctured columns are linearly independent of the received codewords, and the original blind recognition method can be used for recognition.

[0005] However, due to its own limitations, the above method often has poor recognition effect due to the large row weight of the check equation selected for blind recognition after linear transformation. On the other hand, when the matrix of the candidate set is small, the similarity is high, and the number of received codewords is small, the effect is often not ideal. Summary of the invention

[0006] Based on this, the present invention proposes a 5G-LDPC code closed set blind recognition method based on BP iteration to improve the accuracy when the candidate set matrix is ​​small, the similarity is high, and the number of received codewords is small.

[0007] The present invention provides a 5G-LDPC code closed set blind identification method based on BP iteration, comprising the following steps:

[0008] Step 1: Select a check matrix from the 5G-LDPC check matrix candidate set, and divide the received codeword according to the size of the selected check matrix;

[0009] In the initial check matrix used for BP decoding, the likelihood value of the corresponding punctured column is 0, the likelihood value of the filled column is infinity, and the likelihood values ​​of the remaining bits are filled in with the likelihood value calculation results of the corresponding received codeword;

[0010] Step 2: Perform a certain number of BP iterations to update the likelihood values ​​of the variable nodes and check nodes in the check matrix;

[0011] Step 3: Based on the obtained new likelihood value of each bit, the index value of the current check matrix is ​​obtained using the LLR blind recognition method;

[0012] Step 4: Repeat steps 1-3 for each check matrix in the 5G-LDPC check matrix candidate set, and select the LDPC check matrix with the largest index value as the identification result.

[0013] Furthermore, in step 1, the fill node is set to an upper limit value for subsequent calculations.

[0014] Furthermore, in step 2, on the one hand, each check node calculates an updated confidence based on the messages it receives from all other connected variable nodes and passes it to the target variable node;

[0015] Specifically, the check node converts the confidence messages received from other connected variable nodes through the hyperbolic tangent function and then performs a product operation to indicate that these variables satisfy the constraints of the check equation; then, the check node converts the result into the corresponding log-likelihood ratio form and passes it to the target variable node for further iterative update;

[0016] On the other hand, each variable node calculates and updates the information to be sent to the target check node based on the likelihood value of the received initial signal and the messages transmitted from other connected check nodes;

[0017] Specifically, the message sent by the variable node to the check node is the sum of its initial belief and all messages received from the remaining connected check nodes; this updated belief information represents the variable node's current estimate of the transmitted bits and is used for the next iterative propagation.

[0018] In step 3, the index value of the check matrix is ​​the index value y for each check node i Take the arithmetic mean;

[0019] Index value y i Calculate according to the following formula:

[0020]

[0021] N v (i) represents the set of variable nodes participating in the i-th check node, and the likelihood value L(v l ∣r l ) is obtained through BP iterative calculation.

[0022] The benefit of the present invention is that through the idea of ​​BP decoding, relatively accurate confidence information is obtained for the punctured bits and the padding bits through information propagation, and blind recognition is performed based on this, which solves the problem that the original 5G-LDPC cannot be blindly recognized due to the existence of puncture and padding. And unlike the idea of ​​linear transformation, the original matrix is ​​not transformed, and the characteristics such as low row weight are retained, thereby ensuring the recognition performance. The algorithm has obvious advantages over the idea of ​​linear transformation when the matrices of the alternative set are small, the similarity is high, and the number of received codewords is small. Our method has good results on matrices built based on the 5G standard, and is up to 1.5dB higher than the previous method on the data set. The overall recognition performance is better than any previous traditional algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flow chart of a closed set blind recognition method of 5G-LDPC codes based on BP iteration in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In this example, the codeword is transmitted by exemplarily adopting BPSK modulation under an AWGN channel.

[0025] First, take a check matrix to be identified in the candidate set, and divide the received codeword according to the size of the check matrix after receiving the codeword; in the process of constructing the initial matrix for BP iteration, the likelihood value of the corresponding punctured bit in the check matrix is ​​0, the likelihood value of the filled bit is infinity, and the remaining bits are filled according to the received codeword. The likelihood value (log likelihood ratio, LLR) can be calculated according to the following formula:

[0026]

[0027] where r i Indicates the received signal value, v i represents the actual codeword symbol, σ 2 represents the noise variance of the channel.

[0028] Here we consider the traditional AWGN channel using BPSK modulation. In terms of initial information, due to the puncturing and padding that need to be considered under the 5G standard, the default puncturing nodes do not actually transmit and have no prior information, so the initial likelihood value is 0. Although the padding nodes do not transmit either, in this model, all of these nodes are set to 0 by default, which is known information. Therefore, the initial likelihood value of these nodes is infinite, and is set as an upper limit value in the actual calculation for subsequent calculations. The corresponding likelihood values ​​of the remaining nodes can also be calculated under the condition of known channel models and modulation. Based on this, the initialization of BP decoding is completed, and normal iterations can be performed.

[0029] Next, each check node calculates an updated confidence based on the messages it receives from all other connected variable nodes and passes it to the target variable node. Specifically, the check node multiplies the confidence messages received from other connected variable nodes (after hyperbolic tangent function conversion) to indicate that these variables satisfy the constraints of the check equation. The check node then converts the result into the corresponding log-likelihood ratio form and passes it to the target variable node for further iterative updates.

[0030] Each variable node calculates and updates the information to be sent to the target check node based on the confidence (log likelihood ratio, LLR value) of the received initial signal and the messages passed from other connected check nodes. Specifically, the message sent by the variable node to the check node is the sum of its initial confidence and all messages received from the remaining connected check nodes. This updated confidence information represents the variable node's current estimate of the transmitted bit, which is used for the next iterative propagation.

[0031] A more detailed BP iteration is well known to those skilled in the art and will not be described in detail here.

[0032] For example, consider a check node, which represents a parity check constraint (SPC). Consider the case where the code length is d and a certain SPC constraint is satisfied, for bit v0:

[0033]

[0034] After rearranging the above formula, we get:

[0035]

[0036] For general binary random variables with probabilities p1 and p0, there is the following relationship

[0037]

[0038]

[0039]

[0040] Here, tanh represents the hyperbolic tangent function.

[0041] So for the bit v0, we can get the following expression:

[0042]

[0043] where tanh -1 represents the inverse hyperbolic tangent function, and generalizing the above formula yields:

[0044]

[0045] N v (i) represents the set of variable nodes participating in the i-th check node. It is not difficult to find that the above formula is the core formula for information transmission in BP decoding, that is, the information transmitted from the check node to the variable node. At the same time, there is a similar identification formula for the LLR blind recognition algorithm, that is, for each check node i, its index value y i Calculate according to the following formula:

[0046]

[0047] Where L(v l ∣r l) is calculated based on BP iteration, which avoids the problem of pruning bits being 0 and padding bits being infinite. At the same time, the BP decoding iteration part used for blind recognition in the whole process can also be used as part of the decoding iteration after the correct matrix is ​​identified, further reducing the decoding time.

[0048] After the above iterations, the confidence information of the deleted nodes and filled nodes is no longer 0 and infinite, but is updated to a new value through the above information propagation process, and in this process, for the correct matrix, the deleted nodes and filled nodes will eventually converge. At this time, the LLR blind recognition method is used to identify it by whether it meets the constraint of the check equation, that is, for the correct matrix, it will converge to a larger value, while for the wrong matrix, it will not converge or converge slowly. Based on this, by setting the appropriate number of cycles, the correct matrix can be identified based on the LLR blind recognition method.

[0049] For the correct matrix, during the BP decoding iteration, the likelihood value of the bit converges during the information transmission process, and its corresponding y i It will also gradually converge, and the information transmitted by the corresponding check node to the variable node will also converge. And because it satisfies the SPC check, the final result will make the value calculated by LLR blind recognition converge to a larger value.

[0050] For an erroneous matrix, since the BP decoding iteration will not converge or converge slowly, when the correct matrix has converged, the LLR index value calculated for the erroneous matrix will be smaller.

[0051] Therefore, at this time, the LLR blind recognition algorithm is used to finally average the LLR values ​​of each check node to obtain the recognition index value of a single matrix. Based on this, the correct check matrix can be selected to achieve the purpose of blind recognition.

[0052] In this experiment, information data is randomly generated, and the candidate set matrix is ​​generated according to the 5G standard. A total of 8 matrices are generated as candidate sets, and their specific parameters are as follows:

[0053]

[0054] Such 8 matrices are often used in actual situations, so these 8 matrices are selected to form a candidate set for blind recognition experiments. The experiment is carried out when the number of codewords is equal to 10 and the number of iterations is equal to 1.

[0055] For a matrix with code length 1536 and code rate 1 / 3, the following results are obtained:

[0056]

[0057] For a matrix with code length 3072 and code rate 1 / 3, the following results are obtained:

[0058]

[0059] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, several modifications and improvements can be made without departing from the technical solution of the present invention, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by the present invention shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A 5G-LDPC code closed set blind identification method based on BP iteration, characterized in that: The following steps are involved: Step 1: Select a check matrix from the 5G-LDPC check matrix candidate set, and divide the received codeword according to the size of the selected check matrix; In the initial check matrix used for BP decoding, the likelihood value of the corresponding punctured column is 0, the likelihood value of the filled column is infinity, and the likelihood values ​​of the remaining bits are filled in with the likelihood value calculation results of the corresponding received codeword; Step 2: Perform a certain number of BP iterations to update the likelihood values ​​of the variable nodes and check nodes in the check matrix; Step 3: Based on the obtained new likelihood value of each bit, the index value of the current check matrix is ​​obtained using the LLR blind recognition method; Step 4: Repeat steps 1-3 for each check matrix in the 5G-LDPC check matrix candidate set, and select the check matrix with the largest index value as the identification result.

2. The method according to claim 1, characterized in that In step 1, the fill node is set to an upper limit value for subsequent calculations.

3. The method according to claim 1, characterized in that In step 2, on the one hand, each check node calculates an updated confidence based on the messages it receives from all other connected variable nodes and passes it to the target variable node; Specifically, the check node converts the confidence messages received from other connected variable nodes through the hyperbolic tangent function and then performs a product operation to indicate that these variables satisfy the constraints of the check equation; then, the check node converts the result into the corresponding log-likelihood ratio form and passes it to the target variable node for further iterative update; On the other hand, each variable node calculates and updates the information to be sent to the target check node based on the likelihood value of the received initial signal and the messages transmitted from other connected check nodes; Specifically, the message sent by the variable node to the check node is the sum of its initial confidence and all messages received from the remaining connected check nodes; this updated confidence information represents the variable node's current estimate of the transmitted bits and is used for the next iterative propagation.

4. The method according to claim 1, characterized in that In step 3, the index value of the check matrix is ​​the index value y for each check node i Take the arithmetic mean; Index value y i Calculate according to the following formula: N v (i) represents the set of variable nodes participating in the i-th check node, where r l Indicates the received codeword signal value, v l represents the actual codeword symbol, the likelihood value L(v l ∣r l ) is obtained through BP iterative calculation.

Images