CN105846827B - Iterative joint message source and channel interpretation method based on arithmetic code and low density parity check code - Google Patents

Abstract

The iterative joint message source and channel interpretation method based on arithmetic code and low density parity check code that the present invention is to provide a kind of.Source symbol sequence s^hCoded sequence b is obtained after AC encoder^h, D coded sequence b^hThe input message sequence b of LDPC encoder is obtained after parallel-to-serial converter, b forms codeword sequence x after passing through LDPC encoder, x is sent to awgn channel after BPSK is modulated, reception sequence r is input in the closed loop of ldpc decoder and Chase-SISO AC decoder composition and is iterated decoding, and ldpc decoder exports coding sequence after iteration several timesSequence is converted to through deserializerBy obtaining decoding symbol sebolic addressing after AC decoderThe present invention combines the very high AC of lossless compression efficiency with the very strong LDPC code of anti-error capability, so that the validity and reliability of system is all very high, while further improving reliability in the case where guaranteeing validity using IJSCD method.

Description

Iterative Joint Source-Channel Decoding Based on Arithmetic Code and Low Density Parity Check Code method

technical field

The invention relates to a joint source-channel decoding method in a wireless communication system, in particular to an iterative joint source-channel decoding method based on arithmetic codes and low-density parity-check codes.

Background technique

Source coding and channel coding are two indispensable parts of a communication system. The purpose of source coding is to remove source redundancy and improve the effectiveness of communication. The purpose of channel coding is to add redundancy to improve communication reliability. In the traditional receiver structure design, the source decoding and channel decoding are usually considered separately, and the source information is not fully utilized. The Joint Source Channel Decoding (JSCD) method considers the source and the channel as a whole, so that the source information can be fully utilized, and the reliability can be further improved on the premise of ensuring the validity.

Arithmetic coding (AC) is a lossless source coding with higher compression efficiency than Huffman code. It maps the entire symbol sequence to be coded into a codeword, and introduces the idea of decimal coding, which can be infinitely close to the theoretical compression ratio. It has been widely used in various compression standards such as image and video. Low-density Parity-check (LDPC) is a channel coding scheme comparable to Turbo codes, and its decoding performance can approach the Shannon limit. (DVB-S2) adopted. Therefore, the combination of arithmetic codes and LDPC codes has broad application prospects and practical value. The JSCD method can improve the anti-interference ability and meet the reliability requirements of communication on the basis of ensuring the effectiveness.

With the proposal of the idea of Turbo iterative decoding, the Soft-input Soft-output (SISO) decoding algorithm has received extensive attention from scholars at home and abroad. In order to realize the joint decoding of the source and channel, a source decoder structure based on the SISO algorithm is proposed. The source decoder combines the source prior information and the channel information, uses the correlation algorithm to obtain the soft information, and transmits it. to the channel decoder. At present, multimedia digital transmission has become the mainstream of communication services. The amount of data transmitted is very large and faces a complex and changeable channel environment, which puts forward higher requirements for the effectiveness and reliability of communication systems. If the joint source-channel decoding technology is applied to the multimedia communication system, the reliability of the system can be improved while ensuring the efficient transmission of data.

In 2007, M. Grangetto et al published an article entitled "Iterative decoding of serially concatenated arithmetic and channelcodes with JPEG 2000 applications" in the journal "IEEE Transactions on Image Processing", proposing an iterative algorithm based on arithmetic codes and systematic convolutional codes Joint Source Channel Decoding (Iterative Joint Source Channel Decoding, IJSCD) method, this method adds disabled symbols in the arithmetic encoder to increase the error detection capability of the arithmetic code, and the AC-SISO decoder uses the BCJR (BahlCocke Jelinek Raviv) algorithm to calculate the posterior probability of the information bit, and pass it to the convolutional code decoder, and the convolutional code decoder passes the obtained external information to the AC-SISO decoder, and finally completes the decoding through multiple iterations code. However, the disadvantage of this method is that it loses part of the compression efficiency of the arithmetic code, requires a large amount of calculation, and has high implementation complexity.

In 2008, Feng Yuye proposed an adaptive IJSCD method based on arithmetic codes and systematic convolutional codes in his doctoral dissertation "Research on Joint Source-Channel Coding and Decoding Based on Arithmetic Codes". Compared with the IJSCD method proposed by M. Grangetto et al., the number of decoding nodes of the BCJR algorithm is adjusted adaptively. However, when the length of the information bit is large, the calculation amount of this method is still very high. Huge and difficult to achieve.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide an iterative joint source channel based on arithmetic code and low-density parity check code, which can further improve the reliability of the system while ensuring the validity, and has a small amount of calculation and a low implementation complexity. decoding method.

The object of the present invention is achieved in this way:

The source symbol sequence sh is passed through the AC encoder 1 to obtain the encoded sequence b ^h , and the D encoded sequences b ^h are ^passed through the parallel-to-serial converter 2 to obtain the input information sequence b of the LDPC encoder 3 , and b passes through the LDPC encoder 3 to form The codeword sequence x, x is modulated by BPSK and sent to AWGN channel 4, and the received sequence r is input to the closed loop composed of LDPC decoder 5 and Chase-SISO AC decoder 6 for iterative decoding. After several iterations, LDPC Decoder 5 outputs the decoding sequence Converted by serial-parallel converter 7 to obtain the sequence After passing through the AC decoder 8, the decoded symbol sequence is obtained

The received sequence r is input into the closed loop formed by the LDPC decoder 5 and the Chase-SISO AC decoder 6, and the iterative decoding specifically includes: setting the initial feedback of the Chase-SISO AC decoder 6 to the LDPC decoder 5. The value is "0", the demodulator obtains the posterior probability of the transmitted sequence according to the channel information r, and transmits it to the LDPC decoder 5, and then performs the first iterative joint decoding process; the first iterative joint decoding process; The decoding process includes: the LDPC decoder 5 uses the LLR-BP algorithm to output the decoding sequence and soft information q _i , the decoding sequence and the soft information _qi is passed to the Chase-SISO AC decoder 6, and the Chase-SISO AC decoder 6 according to the sequence The new soft information _wi is obtained by using the Chase-type algorithm with the soft information qi _, and the new soft information _wi is fed back to the LDPC decoder 5 to complete the first iterative joint decoding.

The several iterations are to continue to complete the second, third, .

The iterative joint source channel decoding (IJSCD) method based on arithmetic code (AC) and low density parity check (LDPC) code of the present invention is composed of AC encoder 1, parallel-serial converter (P/S) 2, LDPC encoder 3, AWGN channel 4, LDPC decoder 5, AC soft input and soft output (Chase-SISO AC) decoder 6 based on Chase-type algorithm, serial-to-parallel converter (S/P) 7 and AC decoder Completed in the system constituted by the encoder 8. The source symbol sequence sh is passed through the AC encoder 1 to obtain the encoded sequence b ^h , and the D encoded sequences b ^h are ^passed through the parallel-to-serial converter 2 to obtain the input information sequence b of the LDPC encoder 3 , and b passes through the LDPC encoder 3 to form The codeword sequence x, x is modulated by BPSK and sent to AWGN channel 4, and the received sequence r is input to the closed loop composed of LDPC decoder 5 and Chase-SISO AC decoder 6 for iterative decoding. After several iterations, LDPC Decoder 5 outputs the decoding sequence Converted by serial-parallel converter 7 to obtain the sequence After passing through the AC decoder 8, the decoded symbol sequence is obtained The main features of the present invention are:

(1) The LDPC decoder 5 and the Chase-SISO AC decoder 6 form a closed loop based on the iterative processing of the SISO algorithm, and the initial value fed back to the LDPC decoder 5 by the Chase-SISO AC decoder 6 is set to "0". The demodulator obtains the posterior probability of the transmitted sequence according to the channel information r, and transmits it to the LDPC decoder 5, and then performs the first iterative joint decoding process, that is, the LDPC decoder 5 uses the LLR-BP algorithm to output the decoding process. code sequence and soft information q _i , the sequence and soft information are passed to the Chase-SISO AC decoder 6, and the Chase-SISO AC decoder 6 according to the sequence and the reliability qi obtain the soft information w _i using the Chase-type algorithm _, and feed it back to the LDPC decoder 5 . So far, the first iterative joint decoding is completed, and the second, third, ..., Nth iterations are continued according to this method, until the maximum number of outer iterations is reached or

(2) A closed loop of iterative processing based on SISO algorithm composed of LDPC decoder 5 and Chase-SISO AC decoder 6, through the iterative mechanism, the channel decoder and the source decoder can exchange and transmit soft information with each other, making full use of the Source information and channel information.

(3) The introduction and application of the low-complexity Chase-type algorithm in the AC decoder 6 .

The invention adopts a low-complexity Chase-type algorithm to realize the SISO structure of the arithmetic code, and performs iterative joint decoding of the arithmetic code and the LDPC code, which improves the reliability of the system and takes into account the complexity of realization.

The advantages of the present invention are:

Considering that the realization of the AC-SISO decoder based on the BCJR algorithm is at the expense of the AC coding efficiency, the present invention adopts the Chase-type algorithm to design the AC-SISO decoder. Using the Chase-type algorithm does not need to add forbidden symbols in the source set, and has no effect on the coding efficiency of AC.

Aiming at the problems that the AC-SISO decoder based on the BCJR algorithm has a large amount of computation and high implementation complexity, the Chase-type algorithm adopted in the present invention can effectively reduce the computation amount of the AC-SISO decoder. The AC-SISO decoder based on the Chase-type algorithm only selects the sequence satisfying the following conditions as the decoding sequence from the candidate sequences Q=2α ( ^α generally takes 6):

1) The length of the decoded symbol sequence is equal to the length of the source symbol sequence;

2) has the largest posterior probability.

Since there are not many multiplication operations in the decoding process, and the amount of operations is only related to the selection of the bit number α with the lowest reliability, it has nothing to do with the length of the sequence. Therefore, the AC-SISO decoder based on the Chase-type algorithm has lower computational complexity and is easy to implement.

The present invention combines the current AC with very high lossless compression efficiency and the LDPC code with very strong error resistance, so that the system has high effectiveness and reliability, and at the same time adopts the IJSCD method, which can ensure the effectiveness. to improve the reliability of the system.

Description of drawings

Fig. 1 is a system block diagram corresponding to the present invention;

2 is a schematic diagram of message transfer in the iterative joint decoding process of AC and LDPC codes corresponding to the present invention;

3 is a bit error rate curve of AC and LDPC code iterative joint decoding corresponding to the present invention;

FIG. 4 is a packet loss rate curve of iterative joint decoding of AC and LDPC codes corresponding to the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings:

FIG. 1 is a block diagram of a system corresponding to the present invention. The system consists of an AC encoder 1, a parallel-to-serial converter (P/S) 2, an LDPC encoder 3, an AWGN channel 4, an LDPC decoder 5, and a Chase-type based encoder. The AC soft input and soft output (Chase-SISOAC) decoder 6 of the algorithm, a serial-to-parallel converter (S/P) 7 and an AC decoder 8 are constituted. Each module in Figure 1 is defined as follows:

1 is the arithmetic code encoder;

2 is a parallel-serial converter (P/S);

3 is the LDPC code encoder;

4 is the Additive White Gaussian Noise (AWGN) channel;

5 is the LDPC decoder;

6 is the AC-SISO decoder using the Chase-type algorithm;

7 is a serial-to-parallel converter (S/P), and its function is opposite to that of module 2;

8 is an arithmetic code decoder.

For ease of description, each symbol in Figure 1 is defined as follows:

s ^h [L]: symbol sequence of length L input by the source, 1≤h≤D, D represents the number of symbol sequences input to AC encoder 1 in parallel;

b ^h [k _h ]: the codeword sequence of length k _h output by the AC encoder 1;

b[K]: the codeword sequence of length K output by the parallel-serial converter 2;

x[N]: codeword sequence of length N output by LDPC encoder 3;

r[N]: Sequence of length N output by AWGN channel 4;

q[K]: the sequence of length K output by the LDPC decoder 5;

q ^h [k _h ]: the sequence of length k _h output by the serial-parallel converter 7;

_{wh [k h ]: a sequence of length k h} ^output _by the Chase-SISO AC decoder 6;

w[K]: the sequence of length K output by the parallel-serial converter 2 and sent to the LDPC decoder 5;

After several iterations, the LDPC decoder 5 outputs a decoding sequence with a length of K;

After the serial-parallel converter 7, the output sequence of length k _h ;

The AC decoder 8 outputs a sequence of decoded symbols of length L.

With reference to Fig. 1, the symbol sequence sh to be sent is obtained by the AC encoder 1 to obtain a coding sequence b _{h with a length of k h} ^, and the D ^{b h are} converted by a parallel-serial converter 2 to obtain an information sequence b with a length of K, that is, The length of the information bits of the LDPC code is K'. If K<K', in order to ensure the effective encoding of the LDPC code, several "0" or "1" are padded after the sequence b. b passes through the LDPC encoder 3 to form a codeword sequence x, which is modulated by BPSK and sent to the AWGN channel 4, and the received sequence r is input to the closed loop composed of the LDPC decoder 5 and the Chase-SISO AC decoder 6 for iteration Decoding, after several iterations, the LDPC decoder 5 outputs the decoding sequence Converted by serial-parallel converter 7 to obtain the sequence After passing through the AC decoder 8, the decoded symbol sequence is obtained

In conjunction with Fig. 2, Fig. 2 is a schematic diagram of message transfer in the iterative process of AC and LDPC codes, and the definitions of each symbol in Fig. 2 are as follows:

v _j (1≤j≤M): the jth check node of the LDPC code;

c _i (1≤i≤N): the i-th variable node of the LDPC code;

_{Li ij} : the message passed by the j-th check node v _j to the i-th variable node c _i ;

q _i : the soft information transmitted by the i-th variable node c _i to the Chase-SISO AC decoder 6;

w _i : the soft information passed by the Chase-SISO AC decoder 6 to the i-th variable node c _i ;

T _ji : the message that the i-th variable node c _i transmits to the j-th check node v _j .

Since the check bit generated by LDPC code encoding does not contain any source information, the variable node c _i only transmits the message of the information bit to the Chase-SISO AC decoder 6, that is, q _i satisfies 1≤i≤K; , c _i satisfies 1≤i≤K.

The LDPC decoder starts decoding after receiving the channel information sequence r=(r ₁ ,r ₂ ,...,r _N ). The decoding process is as follows:

1) Initialize the information that the variable node c _i passes to the check node v _j connected to it:

Among them, δ ² is the mean square error of Gaussian white noise, and the initial value of _wi is 0;

2) Calculate the information that the check node v _j transmits to the variable node c _i connected to it, and i∈C(j), C(j) represents the set of all variable nodes connected to the check node v _j :

Among them, C(j)\i represents the set of all variable nodes connected to the check node v _j except the variable node c _i , and t is the number of LDPC decoding iterations, which is called the inner iteration number of joint decoding;

3) Calculate the information that the variable node ci transmits to the check node v _j connected to it, and j∈V( _i ), V( _i ) represents the set of all check nodes connected to the variable node ci:

Among them, V(i)\j represents the set of all check nodes connected to the variable node c _i except the check node v _j ;

4) Calculate the hard decision information of all variable nodes:

like then the decoding sequence the ith codeword of otherwise The check matrix of the LDPC code is H, if Then the iterative joint decoding of AC and LDPC ends; if And the maximum number of inner iterations is not reached, then return to step 2) to continue decoding; if and has reached the maximum number of inner iterations, the LDPC decoder will decode the sequence and its reliability q _i =T _i ^(t) are passed to the Chase-SISO decoder as extrinsic information.

decoding sequence and its reliability qi are converted by S/P7 to obtain D decoding sequences of length _{k h} and its corresponding reliability sequence q ^h , the Chase-SISO AC decoder 6 pairs these D sequences For decoding, the decoding process is as follows:

1) Determine the sequence according to q ^h The position of the α bits with the lowest confidence in the

2) Generate the i-th test vector Wherein Q ⁼ 2α; t ⁱ traverses all binary sequences that only allow "1" to appear in the position corresponding to the α bits with the lowest reliability, and "0" to appear in other positions and the maximum code weight does not exceed α;

3) Generate the i-th test sequence in Represents the modulo 2 sum, y _k is the decoding sequence The kth hard-decision bit of ;

4) utilize standard arithmetic decoder to decode Q test sequences ^zi , if the symbol sequence length obtained by sequence ^zi decoding is equal to L, this sequence is added to set Γ, otherwise sequence ^zi is discarded;

5) Calculate the maximum a posteriori probability (Maximum a Posteriori, MAP) corresponding to the sequence ^zi in the set Γ:

in, is the sequence obtained by the BPSK modulation of the sequence ^zi , ^si is the symbol sequence obtained by the standard arithmetic decoding of the sequence ^zi , and P(s ⁱ ) represents the prior probability of the occurrence of the symbol sequence ^si ;

6) Calculate sequence pairs The output external information w ^h :

in, is the sequence corresponding to the maximum MAP value in the set Γ, J is the number of sequences in the set Γ, σ is the experimental value, E is the α bit with the least confidence in the sequence in the corresponding position.

7) D sequence pairs After decoding by Chase-SISO AC decoder 6, D extrinsic information sequences w ^h are obtained, and after P/S2 conversion, the information w transmitted to the variable node is obtained. If the maximum number of outer iterations has been reached, the decoding ends; otherwise, it returns to the LDPC decoder to continue decoding.

The IJSCD method based on AC and LDPC in this embodiment can further improve its reliability under the premise of ensuring the effectiveness of the system, and the method has a small amount of computation and low implementation complexity. The main features are as follows:

1) AC encoder 1 adopts binary adaptive arithmetic coding, the probability of the source symbol before coding is set as equal probability distribution, the coding initial interval is [0,0×FFFF), and the following bit method is used for coding;

2) The LDPC encoder 3 adopts the system LDPC code, the check matrix adopts the edge density structure (PEG) method, and the encoding adopts the approximate lower triangular matrix encoding method;

3) The LDPC decoder 5 adopts the likelihood ratio belief propagation (Log-likelihood-rate based Belief Propagation, LLR-BP) algorithm based on the number field;

4) The Chase-SISO AC decoder 6 adopts the Chase-type algorithm.

Taking binary independent memoryless sources with probability distributions of 0.9 and 0.1 as an example, the length of the generated symbol sequence ^sh is 110, and 49 parallel input symbol sequences ^sh form a data packet; the LDPC code check matrix uses the edge density The code length is 3000, the code rate is 0.876, the average degree of the variable node is 5, the average degree of the check node is 37.58, the optimal value of σ is 1, using BPSK modulation, based on AC in the AWGN channel The bit error rate (Bit Error Rate, BER) curve and the packet loss rate (Packet Error Rate, PER) curve of the separation source channel decoding (Separate Source Channel Decoding, SSCD) and IJSCD of the LDPC code are shown in Figure 3 and Figure 4 where α represents the number of bits with the lowest reliability selected during AC-SISO decoding, β represents the number of inner iterations of LDPC decoding, and ε represents the number of outer iterations.

It can be seen from Figure 3 that the bit error rate of the IJSCD method based on AC and LDPC codes is lower than that of the SSCD method, which indicates that the IJSCD method has better decoding performance and higher reliability. For the IJSCD method, when β and ε are constant, when α=6, its performance is improved by about 0.1dB compared with α=4; when α and ε are constant, when β=50, its performance is better than β=20 Significantly better; under certain conditions of α and β, the performance is slightly improved when ε=7 compared with ε=5. It can be seen that the larger the value of α, the more times of inner and outer iterations, and the better the decoding performance of the IJSCD method based on AC and LDPC codes. When α=6, β=50, and ε=5, the IJSCD method can obtain about 0.2dB gain compared with SSCD. It can also be seen from Figure 4 that the IJSCD method can obtain a lower packet loss rate than the SSCD method, which improves the reliability of communication.

Claims (1)

1. An iterative joint source channel decoding method based on arithmetic codes and low-density parity check codes is characterized in that: source symbol sequence s^hObtaining a coding sequence b after passing through an AC encoder (1)^hD code sequences b^hAn input information sequence b of the LDPC encoder (3) is obtained after the parallel-serial converter (2), a codeword sequence x is formed after the b passes through the LDPC encoder (3), the x is sent to an AWGN channel (4) after BPSK modulation, a received sequence r is input into a closed loop formed by an LDPC decoder (5) and a Chase-SISO AC decoder (6) for iterative decoding, and the LDPC decoder (5) outputs a decoding sequence after a plurality of iterations Converted into a sequence by a serial-to-parallel converter (7) After passing through an AC decoder (8), a decoded symbol sequence is obtained

The LDPC decoder receives the channel information sequence r ═ (r ═ r)₁,r₂,...,r_N) And starting decoding, wherein the decoding process is as follows:

1) initialization variable node c_iTo check nodes v connected thereto_jThe information of (2):

wherein, delta²Is the mean square error of Gaussian white noise, w_iIs 0;

2) calculating check node v_jTo variable node c connected thereto_iAnd i ∈ C (j), C (j) represents all parity nodes v_jSet of connected variable nodes:

wherein C (j) \\ i represents a variable-dividing node c_iAll other check nodes v_jA set of connected variable nodes, wherein t is the iteration times of LDPC decoding, and is called the internal iteration times of combined decoding;

3) compute variable node c_iTo check node v connected thereto_jAnd j ∈ V (i), V (i) represents all the same variable nodes c_iSet of connected check nodes:

wherein V (i) \\ j represents a check-removing node v_jAll except the same variable node c_iA set of connected check nodes;

4) calculating hard decision information of all variable nodes:

if T_i ^(t)Is greater than or equal to 0, then the decoding sequenceIth codeword ofOtherwiseThe check matrix of the LDPC code is H ifThe iterative joint decoding of the AC and the LDPC is finished; if it is notIf the maximum internal iteration times are not reached, returning to the step 2) to continue decoding; if it is notAnd the maximum internal iteration times are reached, the LDPC decoder decodes the sequenceAnd its confidence level q_i＝T_i ^(t)As extrinsic information to a Chase-SISO decoder;

decoding sequenceAnd its confidence level q_iD pieces of length k are obtained after S/P7 conversion_hIs decoded by the decoding sequenceAnd its corresponding confidence sequence q^hThe Chase-SISO AC decoder (6) pairs the D sequencesDecoding is carried out, and the decoding process is as follows:

1) according to q^hDetermining a sequenceThe α bit position with the lowest confidence level;

2) generating the ith test vectorWherein Q is 2^α；tⁱTraversing all binary sequences which only allow the occurrence of '1' at the position corresponding to α bits with the lowest credibility and the occurrence of '0' at the rest positions and have the maximum code weight not exceeding α;

3) generating the ith test sequenceWherein Denotes the moduli 2 and, y_kIs a decoded sequenceThe kth hard decision bit of (1);

4) using standard arithmetic decoder to test Q sequences zⁱDecoding is carried out if the sequence zⁱThe length of the decoded symbol sequence is equal to L, the sequence is added to the set gamma, otherwise the sequence z is addedⁱDiscarding;

5) computing the sequence z in the set ΓⁱCorresponding maximum a posteriori probability MAP:

wherein,is a sequence zⁱSequence obtained by BPSK modulation, sⁱIs a sequence zⁱThe sequence of symbols, P(s), obtained by standard arithmetic decodingⁱ) Representing a sequence of symbols sⁱA prior probability of occurrence;

6) computing sequence pairsOutput extrinsic information w^h：

Wherein,is the sequence with maximum MAP value in the set Γ, J is the number of sequences in the set Γ, σ is the experimental value, E is α bits with minimum confidence in the sequenceThe corresponding position in (1);

7) d sequence pairsD external information sequences w are obtained by decoding through a Chase-SISO AC decoder (6)^hThen obtaining information w transmitted to the variable node after P/S2 conversion, and finishing decoding if the maximum external iteration times is reached; otherwise, returning to the LDPC decoder to continue decoding.