A decoding optimization method for distributed joint source channel coding system

Abstract

A decoding optimization method for a distributed joint source channel coding system relates to communication data transmission. Steps: Initialize the operation of the LDPC code joint decoder, calculate the initial channel likelihood information of each variable node for each distributed source, and obtain the unsent part by calculating the a posteriori LLR value of the corresponding part of the relevant source and the spatial correlation; calculate the LLR value of the variable node and the check node of the LDPC code, and use the a posteriori LLR value of the variable node as the external information of the BCJR decoder to update the unsent part of the next iteration; calculate the amount of information of the forward speculation and backward speculation of the BCJR decoder, and add the two to output as the external information of the LDPC code joint decoder; perform hard decision operation on the a posteriori LLR value of the LDPC code joint decoding to obtain the decoding codeword; and judge the decoding result. Realize information interaction and improve the system decoding performance.

Description

Technical Field

[0001] The invention relates to communication data transmission, and in particular to a decoding optimization method of a distributed joint source channel coding system based on LDPC code. Background Art

[0002] In wireless sensor networks and video mobile communications, the transmitting nodes are mostly densely distributed, have large data sets, and have limited storage and computing capabilities. Distributed Source Coding (DSC) uses a separate encoding and joint decoding mode. Each non-communicating node is separately encoded, and the encoded information is sent to the central node for joint decoding, transferring the computational complexity to the decoding end, reducing the computational complexity of the node encoding end, and meeting the power consumption requirements of distributed scenarios. In 1973, D. Slepian and J. Wolf (D. Slepian and J. Wolf. Noiseless coding of correlated information sources [J]. IEEE Transactions on Information Theory, 1973, 19 (4): 471-480.) proposed and proved that separate coding can achieve the same compression efficiency as joint coding. The theoretical limit of this compression is called the Slepian-Wolf limit.

[0003] At present, the more popular DSC algorithms are all based on channel codes, such as Turbo codes, Low Density Parity Check (LDPC) codes and Polar codes. Channel codes have certain anti-interference capabilities and can ensure the accuracy of source decoding. In 1974, Wyner proved that for discrete correlated sources, the performance of the channel coding used reaches the Shannon limit, and the compression efficiency of the DSC designed using this channel code can approach the Slepian-Wolf limit. There are two ways to implement DSC using channel codes. One is a DSC scheme based on the syndrome, which takes the division of cosets as the basic idea. In decoding, the sequence is output by searching the coset, and it has the best compression performance under an ideal channel. The other is a DSC scheme based on the check bit. This coding scheme takes into account the noisy channel and uses the supervisory codewords generated by the coding and the related information of the distributed source for joint decoding. Compared with the syndrome-based scheme, it is more in line with the actual complex communication application scenarios.

[0004] Distributed Joint Source-Channel Coding (DJSCC) combines channel coding and DSC, and has both compression and error-resistance functions, making it more suitable for practical application scenarios. According to the classification of compression coding schemes for distributed sources, DJSCC technology can be divided into asymmetric compression and non-asymmetric (including symmetric) compression. For asymmetric compression, one of the distributed sources can be obtained by lossless or lossy decoding at the receiving end and participate in joint decoding as side information of other distributed sources; for "non-asymmetric" compression, all distributed sources achieve compression effects, and the source code rate can be adaptively adjusted while the total rate remains unchanged to ensure the balance of rate distribution between nodes. At present, more research is based on asymmetric DJSCC, and there is relatively little research on "non-asymmetric".

[0005] In distributed scenarios, in addition to the feedback information of the channel, discrete correlated information sources have correlations in time and space. These statistical characteristics can be used to optimize modulation technology and coding schemes to obtain additional performance gains. Traditional distributed correlated information source schemes only mine the spatial correlation of information sources. In order to further mine the temporal correlation of distributed information sources, many scholars have studied the time domain characteristics of information sources by establishing Markov correlated information sources. Summary of the invention

[0006] The purpose of the present invention is to provide a decoding optimization method for a distributed joint source channel coding system based on LDPC codes, which fully exploits the time-space correlation of distributed correlated sources and improves the decoding accuracy of the system.

[0007] In order to facilitate understanding of the following specific formulas of the present invention, the meanings of the English abbreviations, mathematical symbols and marks involved are first given:

[0008] 1)v j represents the j-th variable node;

[0009] 2)c i represents the i-th check node;

[0010] 3) L S,ch (j) represents the distributed information source S and variable node v j The corresponding log-likelihood ratio (LLR) value of the initial state of the channel;

[0011] 4) L S,c (i,j) represents the distributed information source S check node c i To variable node v j The LLR value passed;

[0012] 5) L S,v (j,i) represents the distributed information source S variable node vj To the check node c i The LLR value passed;

[0013] 6) L S,v (j) is the variable node v of the distributed information source S j The posterior LLR value of

[0014] 7) L S,Bv (j) represents the BCJR decoder node v of the distributed source S j Amount of information

[0015] 8) Represents the variable node v j A set of connected check nodes;

[0016] 9) Indicates that the check node c is removed i With variable node v j A set of connected check nodes;

[0017] 10) Represents the check node c i A collection of connected variable nodes;

[0018] 11) Indicates removing the variable node v j With check node c i A collection of connected variable nodes;

[0019] 12) R S (j) represents the channel receiving information of the jth variable node of the distributed information source S;

[0020] 13)σ 2 represents the channel noise variance,

[0021] 14)sign() represents the sign operation;

[0022] 15)tanh() represents the tangent cosine function;

[0023] 16)tanh -1 () represents the inverse function of tangent cosine function;

[0024] 17)Γ S,t Represents the set of variable nodes sent by the distributed information source S;

[0025] 18)Γ S,nt represents the set of variable nodes that are not sent by the distributed information source S;

[0026] 19)ρ represents the spatial correlation of related information sources;

[0027] 20)Pij It represents the probability of state i transferring to state j, which is used to describe the time correlation inherent in the information source.

[0028] The present invention comprises the following steps:

[0029] 1) Initialize the LDPC code joint decoder operation. For each distributed source, calculate the initial channel likelihood information of each variable node. The unsent part is obtained by calculating the a posteriori LLR value and spatial correlation of the corresponding part of the relevant source.

[0030] 2) Calculate the LLR values ​​of the variable nodes in the LDPC code;

[0031] 3) Calculate the LLR value of the check node in the LDPC code;

[0032] 4) Calculate the a posteriori LLR values ​​of the variable nodes in the LDPC code;

[0033] 5) Calculate the amount of information for the forward and backward predictions of the BCJR decoder respectively;

[0034] 6) Calculate the amount of information of each variable node of the BCJR decoder;

[0035] 7) According to the positive or negative condition of the posterior LLR value of the variable node in step 4), a hard decision operation is performed on the code word bit. If it is negative, the code word is judged as 1, otherwise, it is judged as 0;

[0036] 8) Determine the decoding result. If the decoding sequence satisfies the check equation or reaches the maximum number of iterations, stop decoding and output the decoding sequence; otherwise, return to step 1) to update the channel likelihood information of the unsent part of each distributed information source and continue the decoding operation.

[0037] In step 1), the LDPC code joint decoding operation is initialized, and for each distributed information source, the specific method of calculating the initial channel likelihood information of each variable node can be: for each distributed information source, calculating the initial channel likelihood information L of each variable node ch , where the unsent part is obtained by calculating the a posteriori LLR value and spatial correlation of the corresponding part of the related source; without loss of generality, assume that the system has two Markov related sources S 1 With S 2 , with the source S 1 For example, the specific operation of calculating the initial likelihood information of the channel is performed according to the following formula:

[0038]

[0039] The channel likelihood information L of the unsent part of each distributed source chIn each decoding iteration, the a posteriori LLR value of the corresponding part of the relevant source is calculated with the spatial correlation. As the number of decoding iterations increases, the a posteriori LLR value becomes more accurate, and the channel likelihood information L of the unsent part is updated with it. ch The more accurate the spatial correlation, the more precisely the performance gain can be obtained.

[0040] In step 2), the LLR value of the variable node in the LDPC code is calculated, and the specific operation is performed according to the following formula:

[0041]

[0042] In step 3), the LLR value of the check node in the LDPC code is calculated, and the specific operation is performed according to the following formula:

[0043]

[0044] In step 4), the a posteriori LLR values ​​of the variable nodes in the LDPC code are calculated, and the specific operation is performed according to the following formula:

[0045]

[0046] In step 5), the information amount of the forward speculation and backward speculation of the BCJR decoder is calculated respectively, and the specific operation is performed according to the following formula;

[0047]

[0048]

[0049] The a posteriori LLR value of the LDPC code joint decoding is passed to the BCJR decoder as input information, that is, as the external information of the BCJR decoder, to realize the information interaction of the decoder; at the same time, the time correlation carried by the Markov source, that is, the state transition relationship between the previous and next code elements of the information sequence, is utilized, and the influence of the previous and next code elements on the current code element is used as the information amount of forward speculation and backward speculation respectively, and the two are added and output as the external information of the LDPC code joint decoder to obtain the performance gain of time correlation.

[0050] In step 6), the information amount of each variable node of the BCJR decoder is calculated, and the specific operation is performed according to the following formula:

[0051]

[0052] In step 7), according to the positive or negative condition of the posterior LLR value of the variable node in step 4), a hard decision operation is performed on the codeword bit. The specific method can be: for the jth variable node, if The code word is judged as 1, otherwise, it is judged as 0.

[0053] The beneficial effects of the present invention are as follows:

[0054] The information amount of each variable node of the BCJR decoder is used as the external information of the LDPC code joint decoding, and participates in the LDPC code joint decoding to improve the decoding accuracy. The LDPC code joint decoder uses the posterior LLR value of the previous iteration of the relevant source and the spatial correlation to continuously update the likelihood information L of the unsent part in each decoding iteration. ch , more accurately obtain the performance gain of spatial correlation; at the same time, the BCJR decoder mines the time correlation of the source itself by calculating the amount of information of forward speculation and backward speculation; the two decoders provide each other with external information to realize information interaction, fully mining the time-space correlation of distributed Markov correlated sources and improving the decoding performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 The decoding block diagram of the cascaded LDPC code joint decoder and the BCJR decoder for two related sources.

[0056] Figure 2 The information transmission between the cascaded LDPC code joint decoder and the BCJR decoder for two related sources.

[0057] Figure 3 It is a comparison diagram of BER performance curves of the BP decoding method with traditional spatial correlation, the cascaded SP-BCJR decoding method and the method of the present invention when the spatial correlation ρ is 0.04 and the time correlation is different.

[0058] Figure 4 It is a comparison diagram of BER performance curves of the cascaded BP-BCJR decoding method and the method of the present invention when the transition probability is 0.7 and the spatial correlation is different. DETAILED DESCRIPTION

[0059] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments will further illustrate the present invention in conjunction with the accompanying drawings.

[0060] The present invention discloses a decoding optimization method for a Markovian-correlated information source-based "asymmetric" distributed joint information source channel coding system based on LDPC code. The decoding optimization method is mainly composed of a cascade connection of an LDPC code joint decoder and a BCJR decoder, and includes the following steps: performing an LDPC code joint decoder operation initialization operation, calculating the channel initial likelihood information L of each variable node for each distributed information source, and calculating the channel initial likelihood information L of each variable node for each distributed information source. ch, where the unsent part is obtained by calculating the a posteriori LLR value of the corresponding part of the relevant source and the spatial correlation; the LLR values ​​of the variable nodes and check nodes of the LDPC code are calculated, and the a posteriori LLR values ​​of the variable nodes are used as the external information of the BCJR decoder, and are also used to update the LLR of the unsent part of the next iteration ch ; Calculate the amount of information of the forward speculation and backward speculation of the BCJR decoder, and add the two together to output as the external information of the LDPC code joint decoder; perform hard decision operation on the posterior LLR value of the LDPC code joint decoding to obtain the decoded codeword; judge the decoding result.

[0061] Without loss of generality, we assume that the communication system has two related information sources S 1 and S 2 , where the source probability is P[S 1 =0] = P[S 2 =0] = 1 / 2, the temporal correlation within the source is modeled as a first-order Markov model; the spatial correlation between the two is defined as ρ, which is usually modeled as a virtual transition probability P[S 2 ≠S 1 |S 1 ]=ρ BSC channel, where the smaller the ρ value, the stronger the spatial correlation. Assume that the source S 1 and S 2 The information sequence length is M bits, and after LDPC coding, the lengths of the information sequences are P 1 With P 2 The total length of the encoded code is n 1 =M+P 1 With n 2 =M+P 2 , the corresponding encoding rate is R 1 =M / n 1 With R 2 =M / n 2 According to the traditional DJSCC system based on check bits, the distributed sources all send part of the information bits and check bits to achieve compression and error correction functions at the same time: 1 Only the first αM parts of the information bits and the check bits are sent, where α is a constant between 0 and 1; and the source S 2 Then the last (1-α)k parts of the information bits and the check bits are sent. Therefore, the distributed information source S 1 and S 2 The compression ratios are and

[0062] Figure 1 The figure shows the decoding optimization structure of the distributed joint source channel coding system based on LDPC codes for two correlated Markov sources. The information transmission is shown in Figure 2 The specific steps are as follows:

[0063] Step 1 initializes the LDPC code joint decoding operation and calculates the distributed source S 1 and S 2 Channel likelihood information of the sent variable node part and The unsent part is obtained by calculating the a posteriori LLR value and spatial correlation of the corresponding part of the relevant source, with the source S 1 For example, the specific operation of calculating the initial likelihood information of the channel is performed according to the following formula:

[0064]

[0065] in, That is, the channel likelihood information of the variable node part that has been sent in each iteration comes from the channel and remains unchanged; while the channel likelihood information of the variable node part that has not been sent is updated by the posterior LLR value of the previous iteration of the corresponding part of the relevant source using spatial correlation.

[0066] Step 2 calculates the LLR value of the variable node in the LDPC code, and the specific operation is performed according to the following formula;

[0067]

[0068] in, is the BCJR decoder node v of the previous iteration j The amount of information is output as the external information of the variable node corresponding to the LDPC code.

[0069] Step 3 calculates the LLR value of the check node in the LDPC code, and the specific operation is performed according to the following formula;

[0070]

[0071] Step 4 calculates the a posteriori LLR values ​​of the variable nodes in the LDPC code and uses them as the external information and decision soft information of the BCJR decoder. The specific operation is performed according to the following formula;

[0072]

[0073] Step 5 considers the influence of the previous and next code elements on the current code element, and calculates the information amount of the forward prediction and backward prediction of the BCJR decoder respectively. The specific operation is performed according to the following formula:

[0074]

[0075]

[0076] Step 6: Add the forward and backward information of the BCJR decoder to obtain the information of each node of the BCJR decoder, and feed it back to the LDPC code joint decoder as the external information of the next iteration of the LDPC code joint decoder. The specific operation is performed according to the following formula:

[0077]

[0078] Step 7 makes a hard decision on the codeword bits according to the positive or negative status of the posterior LLR value of the variable node in step 4, that is, for the jth variable node, if The code word is judged as 1, otherwise, it is judged as 0;

[0079] Step 8 determines the decoding result. If the decoding sequence satisfies the verification equation or reaches the maximum number of iterations, the decoding is stopped and the decoding sequence is output; otherwise, the decoding returns to step 1 to update the channel likelihood information of the unsent part of each distributed information source and continue the decoding operation.

[0080] The BER performance of the present invention is compared with that of the traditional space-related belief propagation (BP) decoding algorithm and the cascaded BP-BCJR decoding algorithm on the AWGN channel for different time-space correlations as follows. The corresponding simulation parameters are: two correlated Markov sources, AR4JA code with a code rate of 1 / 2, a frame length of 1000 bits, and a parameter α=1 / 2, that is, the two correlated sources send the first half and the second half of the information bits respectively to achieve symmetric compression.

[0081] Figure 3 The results show that the spatial correlation is ρ = 0.04, different temporal correlations, traditional spatial correlation BP decoding, cascaded BP-BCJR decoding and the decoding optimization method proposed in this invention are distributed source S 1 BER performance comparison. Traditional spatial correlation BP decoding only calculates the initial LLR value of the unsent part through spatial correlation before decoding iteration, without the involvement of time-space correlation. The two distributed sources are decoded separately through iterative decoding, which is obviously not as good as the other two cascade decoding algorithms. Cascade BP-BCJR decoding introduces BCJR decoder into traditional spatial correlation BP decoding to exploit the inherent time correlation of Markov source and obtain significant performance gain. The decoding optimization method proposed in the present invention fully exploits the time-space correlation of Markov correlation source and has the best performance. At BER of the order of 10 -5 When the transition probability is 0.7, the proposed decoding optimization method has a performance gain of about 0.25dB compared with the cascaded BP-BCJR decoding method, and a performance gain of about 0.2dB when the transition probability is 0.8.

[0082] Figure 4 Shows the same transition probability P 01 =P 10 =0.7, different spatial correlations, performance comparison between cascaded BP-BCJR decoding and the decoding optimization method proposed in this invention. BER is of the order of 10 -5 When the spatial correlation ρ=0.12 and ρ=0.04, the decoding optimization method proposed in the present invention has a performance gain of about 0.5dB and 0.25dB respectively compared with the cascaded BP-BCJR decoding method.

[0083] The above shows and describes the basic principle, main features and advantages of the present invention. The present invention is based on the decoding optimization method of the "asymmetric" distributed joint source channel coding system of LDPC code. During the decoding process, the LDPC code joint decoder uses the spatial correlation interaction a posteriori LLR value, and the BCJR decoder mines the inherent time correlation of the distributed source, and the two types of decoders realize information interaction. It can be said that the present invention fully mines the time-space correlation of the distributed correlated source, bringing additional performance gains to the system.

Claims

1. A decoding optimization method for a distributed joint source channel coding system, characterized in that The following steps are involved: 1) Initialize the LDPC code joint decoder operation. For each distributed source, calculate the initial channel likelihood information of each variable node. The unsent part is obtained by calculating the a posteriori LLR value and spatial correlation of the corresponding part of the relevant source. The specific method of performing the LDPC code joint decoding operation initialization work and calculating the channel initial likelihood information of each variable node for each distributed information source is as follows: for each distributed information source, calculating the channel initial likelihood information L of each variable node ch , where the unsent part is obtained by calculating the a posteriori LLR value and spatial correlation of the corresponding part of the related source; without loss of generality, assume that the system has two Markov related sources S1 and S2. Taking source S1 as an example, the specific operation of calculating the initial likelihood information of the channel is performed according to the following formula: Among them, L S,ch (j) represents the distributed information source S and variable node v j The corresponding log-likelihood ratio of the initial state of the channel, v j represents the jth variable node, R S (j) is the variable node v of the distributed information source S j The channel receiving information, σ 2 is the channel noise variance, ρ is the spatial correlation of the relevant source, L S,v (j) is the variable node v of the distributed information source S j The posterior LLR value of S,t represents the variable node set sent by the distributed information source S, Γ S,nt represents the set of variable nodes that are not sent by the distributed information source S; The channel likelihood information L of the unsent part of each distributed source ch In each decoding iteration, the a posteriori LLR value of the corresponding part of the relevant source is calculated with the spatial correlation. As the number of decoding iterations increases, the a posteriori LLR value becomes more accurate, and the channel likelihood information L of the unsent part is updated with it. ch The more accurate, the more precisely the performance gain of spatial correlation can be obtained; 2) Calculate the LLR value of the variable node in the LDPC code. The specific operation is performed according to the following formula: Among them, L S,v (j,i) represents the distributed information source S variable node v j The LLR value transmitted to the check node ci, L S,ch (j) represents the distributed information source S and variable node v j The corresponding log-likelihood ratio LLR value of the initial state of the channel, L S,Bv (j) represents the BCJR decoder node v of the distributed source S j The amount of information, c i represents the i-th test node, Represents the variable node v j The set of connected check nodes, Indicates that the check node c is removed i With variable node v j A set of connected check nodes; 3) Calculate the LLR value of the check node in the LDPC code. The specific operation is performed according to the following formula: Among them, L S,c (i,j) represents the distributed information source S check node c i To variable node v j The LLR value passed, tanh() represents the tangent cosine function, tanh -1 () represents the inverse function of the tangent cosine function, represents the set of variable nodes connected to the check node ci, Indicates removing the variable node v j With check node c i A collection of connected variable nodes; 4) Calculate the a posteriori LLR value of the variable node in the LDPC code. The specific operation is performed according to the following formula: 5) Calculate the amount of information for the forward and backward predictions of the BCJR decoder respectively. The specific operation is performed according to the following formula; Among them, L S,Bv (j) represents the BCJR decoder node v of the distributed source S j The amount of information, L S,v (j) is the variable node v of the distributed information source S j The posterior LLR value of The a posteriori LLR value of the LDPC code joint decoding is passed to the BCJR decoder as input information, that is, as the external information of the BCJR decoder, to realize the information interaction of the decoder; at the same time, the time correlation carried by the Markov source, that is, the state transition relationship between the previous and next code elements of the information sequence, is used to take the influence of the previous and next code elements on the current code element as the amount of forward speculation and backward speculation respectively, and the two are added and output as the external information of the LDPC code joint decoder to obtain the performance gain of time correlation. 6) Calculate the information volume of each variable node of the BCJR decoder. The specific operation is performed according to the following formula: 7) According to the positive or negative condition of the posterior LLR value of the variable node in step 4), a hard decision operation is performed on the code word bit. If it is negative, the code word is judged as 1, otherwise, it is judged as 0; 8) Determine the decoding result. If the decoding sequence satisfies the check equation or reaches the maximum number of iterations, stop decoding and output the decoding sequence; otherwise, return to step 1) to update the channel likelihood information of the unsent part of each distributed information source and continue the decoding operation.

2. A decoding optimization method for a distributed joint source channel coding system as claimed in claim 1, characterized in that In step 7), according to the positive or negative condition of the posterior LLR value of the variable node in step 4), a hard decision operation is performed on the codeword bit. The specific method is: for the jth variable node, if The code word is judged as 1, otherwise, it is judged as 0; where sign() represents the sign operation.

Images