A hybrid distributed source coding, decoding method and system

Abstract

The present invention relates to a hybrid distributed source coding, decoding method and system, and particularly relates to the technical field of communication data transmission. The decoding method includes: decoding the extracted subsequence codewords of the source sequence to be decoded to obtain a reconstructed extracted subsequence; using a bilateral symmetric Context model to obtain the conditional distribution of each symbol in the unextracted subsequence according to the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded; decoding according to the conditional distribution of each symbol in the unextracted subsequence and the syndromes of the unextracted subsequence of the source sequence to be decoded to obtain a reconstructed unextracted subsequence; and combining the reconstructed extracted subsequence and the reconstructed unextracted subsequence to obtain the source sequence to be decoded. The key point of the present invention is to further improve the compression performance by using the correlation between adjacent symbols within the source sequence in a distributed source coding scheme based on a channel code.

Description

Technical Field

[0001] The present invention relates to the technical field of communication data transmission, and particularly to a hybrid distributed source coding, decoding method and system. Background Art

[0002] Distributed source coding (DSC) refers to the problem of compressing two or more correlated sources that do not communicate with each other. DSC can be used in fields such as wireless sensor networks, image recognition, and video compression. The theory of DSC is developed based on the Slepian-Wolf theorem proposed by Slepian and Wolf in 1973. This theorem states that for two discrete correlated sources, even if the two sources can only be independently encoded respectively, as long as the encoded bitstreams of the two sources are available at the decoder, then the two correlated sources can be optimally encoded. For example, there are two discrete memoryless source sequences X and Y, where sequence Y is also called the side information sequence. At the encoding end, the side information sequence Y is losslessly encoded at a rate not less than its entropy H(Y). And the source sequence X is encoded at a rate not less than its conditional entropy H(X|Y). At the decoding end, if the decoder of the source sequence X receives enough compressed bits about the source sequence X, then the decoder can reconstruct the source sequence X by performing joint decoding on the received bits about X and the side information sequence Y. We also call this the asymmetric Slepian-Wolf problem.

[0003] To solve the asymmetric Slepian-Wolf problem, some scholars have focused on the technology of source codes. For example, Grangetto et al. extended the traditional arithmetic code to the distributed case (i.e., the distributed arithmetic code scheme). At the encoding end, this scheme first constructs a Context-based statistical model to utilize the correlation between adjacent symbols within the source sequence. Then, the encoder of the source sequence X encodes at a rate not less than its conditional entropy H(X|Y) by allowing the probability intervals to overlap. Although this overlap enables the encoder to achieve a higher compression ratio, it also generates ambiguity. At the decoding end, the uncertainty caused by this ambiguity may make multiple message sequences correspond to the same codeword. To eliminate this ambiguity, the joint decoder utilizes the information of the side information sequence Y to recover the source sequence X. This scheme effectively utilizes the correlation within the source sequence by constructing a Context-based statistical model and demonstrates good compression performance. However, for a source sequence with memory, the distributed arithmetic code scheme will produce highly distorted conditional distributions when using a high-order Context model to estimate the conditional probability of the source sequence X (for example, in a binary distribution, the probability of one symbol is much greater than that of the other symbol). These highly distorted conditional distributions will prevent the probability intervals of the distributed arithmetic code scheme from being further expanded, thus making the distributed arithmetic code scheme unable to fully utilize the correlation between sources to improve the compression performance.

[0004] In addition to the techniques based on source coding, the prior art also has techniques using channel codes to solve the asymmetric Slepian-Wolf problem. In 2003, Pradhan and Ramchandran proposed the first practical channel coding-based scheme for dealing with the asymmetric Slepian-Wolf problem, which is also known as the distributed source coding scheme using syndromes. In this scheme, X represents the source sequence. Since the side information sequence Y is related to X, and this correlation can be regarded as the Y sequence being obtained by passing the source sequence X through a virtual binary symmetric channel (BSC). Therefore, the Y sequence can be regarded as a "noisy" version of the source sequence X. The encoder encodes the source sequence X into the syndrome of the channel code and sends it to the decoder. After receiving the syndrome, the decoder "corrects errors" on the side information sequence Y to recover the source sequence X. This scheme has demonstrated its great application potential through various implementations of systems using Turbo codes and low-density parity-check (LDPC) codes. These channel code-based DSC schemes have all achieved good compression performance by effectively utilizing the correlation between sources. However, in practical Slepian-Wolf applications, there is correlation between adjacent symbols within most source sequences. This correlation is very important for improving the overall performance of the DSC-based scheme. But these channel code-based DSC schemes do not consider how to utilize the correlation between adjacent symbols within the source sequence to further improve the compression performance. Summary of the Invention

[0005] The object of the present invention is to provide a hybrid distributed source coding, decoding method and system, which can utilize the correlation between adjacent symbols within the source sequence to further improve the compression performance of the DSC scheme based on channel codes.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A hybrid distributed source decoding method includes:

[0008] Obtaining the extracted subsequence codeword of the source sequence to be decoded, the syndrome of the unextracted subsequence of the source sequence to be decoded, and the side information sequence of the source sequence to be decoded;

[0009] Decoding the extracted subsequence codeword of the source sequence to be decoded to obtain a reconstructed extracted subsequence;

[0010] Using a bilateral symmetric Context model, according to the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded, obtaining the conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded;

[0011] Jointly decode according to the conditional distributions of the symbols in the unextracted subsequence of the source sequence to be decoded and the syndrome of the unextracted subsequence of the source sequence to be decoded, to obtain a reconstructed unextracted subsequence;

[0012] Merge the reconstructed extracted subsequence and the reconstructed unextracted subsequence to obtain the source sequence to be decoded.

[0013] Optionally, the method for obtaining the conditional distributions of the symbols in the unextracted subsequence of the source sequence to be decoded by using a bilaterally symmetric Context model according to the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded specifically includes:

[0014] According to the bilaterally symmetric Context model, the reconstructed extracted subsequence, and the side information sequence of the source sequence to be decoded, obtain the conditional distributions of the symbols in the side information sequence of the source sequence to be decoded under the condition that the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded are known;

[0015] When scalar scaling is not important, determine the conditional distributions of the symbols in the side information sequence of the source sequence to be decoded as the joint distributions of the symbols in the side information sequence of the source sequence to be decoded, and obtain the joint distributions of the symbols in the unextracted subsequence of the source sequence to be decoded according to the joint distributions of the symbols in the side information sequence of the source sequence to be decoded;

[0016] Obtain the conditional distributions of the symbols in the unextracted subsequence of the source sequence to be decoded according to the joint probability distribution and the joint distributions of the symbols in the unextracted subsequence of the source sequence to be decoded; the joint probability distribution is the joint probability distribution of the side information sequence of the source sequence to be decoded and the reconstructed extracted subsequence.

[0017] Optionally, the method for jointly decoding according to the conditional distributions of the symbols in the unextracted subsequence of the source sequence to be decoded and the syndrome of the unextracted subsequence of the source sequence to be decoded to obtain a reconstructed unextracted subsequence specifically includes:

[0018] Process the conditional distributions of the symbols in the unextracted subsequence of the source sequence to be decoded according to the syndrome of the unextracted subsequence of the source sequence to be decoded and the transition probability of a binary symmetric channel, to obtain the processed conditional distributions of the symbols in the unextracted subsequence of the source sequence to be decoded;

[0019] Jointly decode according to the processed conditional distributions of the symbols in the unextracted subsequence of the source sequence to be decoded and the syndrome of the unextracted subsequence of the source sequence to be decoded, to obtain a reconstructed unextracted subsequence.

[0020] Optionally, process the conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded according to the syndrome of the unextracted subsequence of the source sequence to be decoded and the transition probability of the binary symmetric channel, and obtain the processed conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded, which specifically includes:

[0021] For any symbol in the unextracted subsequence of the source sequence to be decoded, input the conditional distribution of the symbol and the syndrome of the unextracted subsequence of the source sequence to be decoded into the BP algorithm to obtain the posterior information of the symbol;

[0022] Remove the internal information of the posterior information of the symbol according to the conditional distribution of the symbol and the transition probability of the binary symmetric channel to obtain the removed distribution;

[0023] According to the bilateral symmetric Context model, use the removed distribution and the conditional distribution of the symbol to obtain the processed conditional distribution of the symbol.

[0024] Optionally, perform joint decoding according to the processed conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded and the syndrome of the unextracted subsequence of the source sequence to be decoded to obtain a reconstructed unextracted subsequence, which specifically includes:

[0025] Use an LDPC decoder to perform joint decoding according to the processed conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded and the syndrome of the unextracted subsequence of the source sequence to be decoded to obtain a reconstructed unextracted subsequence.

[0026] A hybrid distributed source decoding system, comprising:

[0027] An acquisition module, configured to acquire the extracted subsequence codeword of the source sequence to be decoded, the syndrome of the unextracted subsequence of the source sequence to be decoded, and the side information sequence of the source sequence to be decoded;

[0028] An arithmetic decoder, configured to decode the extracted subsequence codeword of the source sequence to be decoded to obtain a reconstructed extracted subsequence;

[0029] A probability estimation module, configured to use a bilateral symmetric Context model to obtain the conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded according to the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded;

[0030] A reconstructed unextracted subsequence determination module, configured to perform joint decoding according to the conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded and the syndrome of the unextracted subsequence of the source sequence to be decoded to obtain a reconstructed unextracted subsequence;

[0031] A merging module, configured to merge the reconstructed extracted subsequence and the reconstructed unextracted subsequence to obtain the source sequence to be decoded.

[0032] A hybrid distributed source coding method, applied to the above-mentioned hybrid distributed source decoding method, the hybrid distributed source coding method includes:

[0033] Obtain a source sequence to be encoded;

[0034] Determine the side information sequence of the source sequence to be encoded;

[0035] Perform an extraction operation on the source sequence to be encoded to obtain an extracted subsequence and an unextracted subsequence of the source sequence to be encoded;

[0036] Perform lossless compression on the extracted subsequence of the source sequence to be encoded to obtain the codeword of the extracted subsequence of the source sequence to be encoded;

[0037] Encode the unextracted subsequence of the source sequence to be encoded to obtain the syndrome of the unextracted subsequence of the source sequence to be encoded.

[0038] Optionally, the performing lossless compression on the extracted subsequence of the source sequence to be encoded to obtain the codeword of the extracted subsequence of the source sequence to be encoded specifically includes:

[0039] Use an arithmetic encoder based on the Context model to perform lossless compression on the extracted subsequence of the source sequence to be encoded to obtain the codeword of the extracted subsequence of the source sequence to be encoded.

[0040] Optionally, the encoding the unextracted subsequence of the source sequence to be encoded to obtain the syndrome of the unextracted subsequence of the source sequence to be encoded specifically includes:

[0041] Use an LDPC encoder to encode the unextracted subsequence of the source sequence to be encoded to obtain the syndrome of the unextracted subsequence of the source sequence to be encoded.

[0042] A hybrid distributed source coding system, including:

[0043] An obtaining module, configured to obtain a source sequence to be encoded;

[0044] A side information sequence determination module, configured to determine the side information sequence of the source sequence to be encoded;

[0045] An extraction module, configured to perform an extraction operation on the source sequence to be encoded to obtain an extracted subsequence and an unextracted subsequence of the source sequence to be encoded;

[0046] A compression module for losslessly compressing the extracted subsequence of the to-be-encoded source sequence to obtain the codeword of the extracted subsequence of the to-be-encoded source sequence;

[0047] An encoding module for encoding the unextracted subsequence of the to-be-encoded source sequence to obtain the syndrome of the unextracted subsequence of the to-be-encoded source sequence.

[0048] According to the specific embodiments provided by the present invention, the following technical effects are disclosed: The decoding method provided by the present invention decodes the codeword of the extracted subsequence of the to-be-decoded source sequence to obtain the reconstructed extracted subsequence; uses a bilateral symmetric Context model to obtain the conditional distribution of each symbol in the unextracted subsequence according to the reconstructed extracted subsequence and the side information sequence of the to-be-decoded source sequence; performs joint decoding according to the conditional distribution of each symbol in the unextracted subsequence and the syndrome of the unextracted subsequence of the to-be-decoded source sequence to obtain the reconstructed unextracted subsequence; combines the reconstructed extracted subsequence and the reconstructed unextracted subsequence to obtain the to-be-decoded source sequence. At the decoding end, by constructing a bilateral symmetric Context model, the internal correlation of the source sequence and the inter-source correlation are fully utilized to generate the conditional distribution of the source symbols, and then the correlation between adjacent symbols within the source sequence is used to further improve the conditional distribution of the source symbols, significantly improving the compression performance of the DSC scheme based on the channel code. Description of the Drawings

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0050] Figure 1 It is a flowchart of a hybrid distributed source coding and decoding method provided by an embodiment of the present invention;

[0051] Figure 2 It is a schematic diagram of an extraction method provided by an embodiment of the present invention;

[0052] Figure 3 It is a schematic diagram of a statistical process provided by an embodiment of the present invention;

[0053] Figure 4 It is a comparison diagram of the bit error rate (BER) performance curves of the DSC scheme based on LDPC codes and the method of the present invention when the transition probability δ = 0.05 of the first-order binary Markov source sequence, different BSC crossover probabilities p, and code rate R = 1 / 4;

[0054] Figure 5 This is a comparison graph of the BER performance curves of the DSC scheme based on LDPC codes and the method of the present invention when the transfer probability δ of the first-order binary Markov source sequence in the embodiments of the present invention is 0.15, different BSC crossover probabilities p, and the code rate R = 1 / 4. Detailed implementation manners

[0055] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0056] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0057] The object of the present invention is to provide a distributed source coding and decoding method that jointly uses a channel code and a source code, fully utilizes the correlation between and within sources, and improves the compression performance of the DSC scheme based on the channel code.

[0058] To facilitate the understanding of the following specific formulas and symbols of the present invention, the definitions of the mathematical symbols, superscripts and subscripts, and English abbreviations to be involved are first given:

[0059] Π, Π T and ∏ -T respectively represent the transfer matrix of the BSC channel, the transpose of the transfer matrix, and the inverse of the transpose of the transfer matrix.

[0060] ∏(x,y) represents the transfer probability that when the input symbol of the BSC is x, the output symbol of this channel is y.

[0061] represents the y i th column of the BSC channel transfer matrix ∏.

[0062] represents the source sequence X, and the subscripts indicate starting from 1 to n. If starting from 1, the subscript 1 can be omitted.

[0063] represents the side information sequence Y.

[0064] represents the source sequence reconstructed after the decoder performs the joint decoding operation.

[0065] X1 and X2 respectively represent the decimated subsequence composed of the non-decimated symbols and the non-decimated subsequence composed of the decimated symbols after the source sequence X undergoes the decimation symbol operation.

[0066] Represents the reconstructed version of the subsequence X1 output by the decoder.

[0067] k and c respectively represent the interval of the decimated symbols and the size of the Context.

[0068] ⊙ represents performing an element-wise multiplication operation on vectors of the same dimension.

[0069] The set B represents the set composed of the decimated symbols (i.e., the symbols within the decimated subsequence X1), and B = {x k , x 2k , …, x jk , …, x Jk} ∈ A J , j = 1, 2, …, J, 1 < J * k ≤ n.

[0070] X i and Y i respectively represent the symbols of the source sequence and the side information sequence at position i.

[0071] and represent the conditional distribution and the joint distribution of the current source symbol X i (i.e., the symbol in the source sequence at position i).

[0072] represents the improved conditional distribution regarding the current source symbol X i .

[0073] y n～i represents the side information symbol not included at position i, that is, the sequence

[0074] and respectively represent the conditional distribution and the joint distribution of the current side information sequence symbol Y i (i.e., the symbol in the side information sequence at position i).

[0075] represents the conditional distribution of the current side information sequence symbol Y calculated using the left Context as and the right Context as i as a template.

[0076] Represents the posterior information generated when the Belief Propagation (BP) algorithm converges

[0077] Represents the posterior information The distribution obtained after removing the internal information.

[0078] C represents the total number of occurrences of a certain Context template.

[0079] The object of the present invention is to provide a hybrid (using channel codes and source codes simultaneously) distributed source coding and decoding method, which improves the decoding performance and compression ratio of the DSC scheme based on channel codes by simultaneously utilizing the intra-source and inter-source correlations. The present invention is only applicable to the distributed source coding and decoding scenario of binary symbols. The specific hybrid distributed source decoding method includes:

[0080] Obtain the extracted subsequence codeword of the source sequence to be decoded, the syndrome of the un-extracted subsequence of the source sequence to be decoded, and the side information sequence of the source sequence to be decoded.

[0081] Decode the extracted subsequence codeword of the source sequence to be decoded to obtain the reconstructed extracted subsequence.

[0082] Use a bilateral symmetric Context model to obtain the conditional distribution of each symbol in the un-extracted subsequence of the source sequence to be decoded according to the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded.

[0083] Perform joint decoding according to the conditional distribution of each symbol in the un-extracted subsequence of the source sequence to be decoded and the syndrome of the un-extracted subsequence of the source sequence to be decoded to obtain the reconstructed un-extracted subsequence.

[0084] Merge the reconstructed extracted subsequence and the reconstructed un-extracted subsequence to obtain the source sequence to be decoded.

[0085] In practical applications, the use of a bilateral symmetric Context model to obtain the conditional distribution of each symbol in the un-extracted subsequence of the source sequence to be decoded according to the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded specifically includes:

[0086] According to the bilateral symmetric Context model, the reconstructed extracted subsequence, and the side information sequence of the source sequence to be decoded, obtain the conditional distribution of each symbol in the side information sequence of the source sequence to be decoded under the condition of knowing the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded.

[0087] In the case where scalar scaling is not important, determine the conditional distribution of each symbol in the side information sequence of the to-be-decoded source sequence as the joint distribution of each symbol in the side information sequence of the to-be-decoded source sequence, and obtain the joint distribution of each symbol in the unsubsampled subsequence of the to-be-decoded source sequence according to the joint distribution of each symbol in the side information sequence of the to-be-decoded source sequence. The case where scalar scaling is not important means that when the conditional distribution P(B|A) is known with the probability P(A) (i.e., the scalar), it can be regarded as equivalent to the distribution P(AB).

[0088] Obtain the conditional distribution of each symbol in the unsubsampled subsequence of the to-be-decoded source sequence according to the joint probability distribution and the joint distribution of each symbol in the unsubsampled subsequence of the to-be-decoded source sequence; the joint probability distribution is the joint probability distribution of the side information sequence of the to-be-decoded source sequence and the reconstructed subsampled subsequence.

[0089] In practical applications, jointly decode according to the conditional distribution of each symbol in the unsubsampled subsequence of the to-be-decoded source sequence and the syndrome of the unsubsampled subsequence of the to-be-decoded source sequence to obtain a reconstructed unsubsampled subsequence, which specifically includes:

[0090] Process the conditional distribution of each symbol in the unsubsampled subsequence of the to-be-decoded source sequence according to the syndrome of the unsubsampled subsequence of the to-be-decoded source sequence and the transition probability of the binary symmetric channel to obtain the processed conditional distribution of each symbol in the unsubsampled subsequence of the to-be-decoded source sequence.

[0091] Jointly decode according to the processed conditional distribution of each symbol in the unsubsampled subsequence of the to-be-decoded source sequence and the syndrome of the unsubsampled subsequence of the to-be-decoded source sequence to obtain a reconstructed unsubsampled subsequence.

[0092] In practical applications, process the conditional distribution of each symbol in the unsubsampled subsequence of the to-be-decoded source sequence according to the syndrome of the unsubsampled subsequence of the to-be-decoded source sequence and the transition probability of the binary symmetric channel to obtain the processed conditional distribution of each symbol in the unsubsampled subsequence of the to-be-decoded source sequence, which specifically includes:

[0093] For any symbol in the unsubsampled subsequence of the to-be-decoded source sequence, input the conditional distribution of the symbol and the syndrome of the unsubsampled subsequence of the to-be-decoded source sequence into the BP algorithm to obtain the posterior information of the symbol.

[0094] Remove the internal information of the posterior information of the symbol according to the conditional distribution of the symbol and the transition probability of the binary symmetric channel to obtain the removed score.

[0095] According to the bilateral symmetric Context model, the conditional distribution of the symbol after processing is obtained by using the removed distribution and the conditional distribution of the symbol.

[0096] In practical applications, joint decoding is performed according to the conditional distribution of each symbol in the unextracted subsequence of the to-be-decoded source sequence after processing and the syndrome of the unextracted subsequence of the to-be-decoded source sequence to obtain a reconstructed unextracted subsequence, which specifically includes:

[0097] An LDPC decoder is used to perform joint decoding according to the conditional distribution of each symbol in the unextracted subsequence of the to-be-decoded source sequence after processing and the syndrome of the unextracted subsequence of the to-be-decoded source sequence to obtain a reconstructed unextracted subsequence.

[0098] An embodiment of the present invention further provides a hybrid distributed source decoding system corresponding to the above decoding method, including:

[0099] An acquisition module, configured to acquire the codeword of the extracted subsequence of the to-be-decoded source sequence, the syndrome of the unextracted subsequence of the to-be-decoded source sequence, and the side information sequence of the to-be-decoded source sequence.

[0100] An arithmetic decoder, configured to decode the codeword of the extracted subsequence of the to-be-decoded source sequence to obtain a reconstructed extracted subsequence.

[0101] A probability estimation module, configured to use a bilateral symmetric Context model to obtain the conditional distribution of each symbol in the unextracted subsequence of the to-be-decoded source sequence according to the reconstructed extracted subsequence and the side information sequence of the to-be-decoded source sequence.

[0102] A reconstructed unextracted subsequence determination module, configured to perform joint decoding according to the conditional distribution of each symbol in the unextracted subsequence of the to-be-decoded source sequence and the syndrome of the unextracted subsequence of the to-be-decoded source sequence to obtain a reconstructed unextracted subsequence.

[0103] A merging module, configured to merge the reconstructed extracted subsequence and the reconstructed unextracted subsequence to obtain the to-be-decoded source sequence.

[0104] As an optional implementation manner, the reconstructed unextracted subsequence determination module specifically includes:

[0105] A soft input soft output module, configured to process the conditional distribution of each symbol in the unextracted subsequence of the to-be-decoded source sequence according to the syndrome of the unextracted subsequence of the to-be-decoded source sequence and the transition probability of the binary symmetric channel to obtain the conditional distribution of each symbol in the unextracted subsequence of the to-be-decoded source sequence after processing.

[0106] An LDPC decoder, which is used to perform joint decoding according to the conditional distribution after processing each symbol in the unsubsampled subsequence of the to-be-decoded source sequence and the syndrome of the unsubsampled subsequence of the to-be-decoded source sequence, so as to obtain a reconstructed unsubsampled subsequence.

[0107] An embodiment of the present invention further provides a hybrid distributed source coding method applied to the above decoding method. The hybrid distributed source coding method includes:

[0108] Obtain a to-be-encoded source sequence.

[0109] Determine the side information sequence of the to-be-encoded source sequence.

[0110] Perform a subsampling operation on the to-be-encoded source sequence to obtain a subsampled subsequence and an unsubsampled subsequence of the to-be-encoded source sequence.

[0111] Perform lossless compression on the subsampled subsequence of the to-be-encoded source sequence to obtain a codeword of the subsampled subsequence of the to-be-encoded source sequence.

[0112] Encode the unsubsampled subsequence of the to-be-encoded source sequence to obtain the syndrome of the unsubsampled subsequence of the to-be-encoded source sequence.

[0113] In practical applications, the performing lossless compression on the subsampled subsequence of the to-be-encoded source sequence to obtain a codeword of the subsampled subsequence of the to-be-encoded source sequence specifically includes:

[0114] Use an arithmetic encoder based on the Context model to perform lossless compression on the subsampled subsequence of the to-be-encoded source sequence to obtain a codeword of the subsampled subsequence of the to-be-encoded source sequence.

[0115] In practical applications, the encoding the unsubsampled subsequence of the to-be-encoded source sequence to obtain the syndrome of the unsubsampled subsequence of the to-be-encoded source sequence specifically includes:

[0116] Use an LDPC encoder to encode the unsubsampled subsequence of the to-be-encoded source sequence to obtain the syndrome of the unsubsampled subsequence of the to-be-encoded source sequence.

[0117] An embodiment of the present invention further provides a hybrid distributed source coding system corresponding to the above coding method, including:

[0118] An acquisition module, which is used to acquire a to-be-encoded source sequence.

[0119] A side information sequence determination module, which is used to determine the side information sequence of the to-be-encoded source sequence.

[0120] An extraction module, configured to perform an extraction operation on the source sequence to be encoded to obtain an extracted subsequence and an unextracted subsequence of the source sequence to be encoded.

[0121] A compression module, configured to perform lossless compression on the extracted subsequence of the source sequence to be encoded to obtain a codeword of the extracted subsequence of the source sequence to be encoded.

[0122] An encoding module, configured to encode the unextracted subsequence of the source sequence to be encoded to obtain a syndrome of the unextracted subsequence of the source sequence to be encoded.

[0123] The present invention provides a more specific encoding and decoding method, as Figure 1 shown below:

[0124] The specific hybrid distributed source coding method is as follows:

[0125] Step 101: Obtain the source sequence X to be encoded; the correlation between adjacent symbols (x i-1 and x i ) in the source sequence X is modeled as a first-order Markov model, and the specific operation is carried out according to the following formula:

[0126]

[0127] The smaller δ is, the stronger the correlation between adjacent symbols in the source sequence X is. At the same time, the side information sequence Y is sent to the decoding end in a lossless manner.

[0128] Step 101: Send the source sequence X into a virtual BSC with a crossover probability of p to obtain the side information sequence Y; the correlation between the source sequence X and the side information sequence Y is usually modeled as the virtual crossover probability p; similarly, the smaller p is, the stronger the correlation between the source sequence X and the side information sequence Y is.

[0129] Step 102: Extract the source sequence X to be encoded every k symbols, and divide the source sequence X into an extracted subsequence X1 composed of extracted source symbols and an unextracted subsequence X2 composed of unextracted source symbols, as Figure 2 shown. Where k can be 3.

[0130] Step 103: Use an arithmetic encoder based on the Context model to perform lossless compression on the subsequence X1 to obtain a codeword, and send the codeword to the arithmetic decoder.

[0131] Step 104: Encode the subsequence X2 using an LDPC encoder, obtain the syndrome S of the subsequence X2, and send the syndrome S to the LDPC decoder.

[0132] The method for decoding the above - mentioned distributed source coding method is specifically as follows:

[0133] Step 105: The arithmetic decoder decodes the received codeword to obtain a reconstructed extraction subsequence

[0134] Step 106: At the LDPC decoding end, the probability estimation module estimates the conditional distribution of the symbols of subsequence X2 by using the correlation between subsequence X1 and subsequence X2 and the correlation between subsequence X2 and side - information sequence Y

[0135] Step 107: Send the conditional distribution together with the syndrome S and the transition probability p of the BSC into the soft - input soft - output module; the soft - input soft - output module outputs an improved conditional distribution of the symbols of subsequence X2

[0136] Step 108: Input the improved conditional distribution and the syndrome S into the LDPC decoder together; the LDPC decoder decodes by using the BP algorithm and outputs the reconstructed sequence of subsequence X2

[0137] Step 109: Merge the reconstructed subsequence by the arithmetic decoder and the reconstructed subsequence by the LDPC decoder to output the reconstructed source sequence

[0138] In step 106, the side - information sequence Y and the subsequence are known in the probability estimation module; the probability estimation module uses the symbols adjacent to the current source symbol X i on the left and right…, x (j-1)k , x jk , x (j+1)k ,… and …, y i-2 , y i-1 , y i+1 , y i+2 … to construct a bilateral - symmetric Context model to estimate the conditional distribution of the symbols of subsequence X2 The source symbols and the side - information symbols are in one - to - one correspondence, except that the side - information symbols are the noisy versions of the source symbols. Since at the decoding end, the source symbols to be decoded are unknown, during the probability - statistical process, the corresponding side - information symbols are approximately regarded as the source symbols with added noise. Capital X i represents an uncertain random variable, while lowercase is generally used to represent a certain value. Since the symbol to be decoded currently is unknown, it is represented by capital letters. And the reconstructed subsequence x and y are known, so they are in lowercase. (xjk represents the symbol at the j-th position of the reconstructed extraction subsequence and is known).

[0139] Conditional distribution of the symbols of subsequence X2 It is calculated according to the following formula:

[0140]

[0141] Pr(Y n = y n , B = b) represents the joint probability distribution of the side information sequence y n and subsequence X1; since the side information sequence y n and subsequence X1 are known, Pr(Y n = y n , B = b) is a known constant; since Pr(Y n = y n , B = b) is a known constant, so the conditional distribution In the case where scalar scaling is not important, it can be used interchangeably with the joint distribution and used interchangeably.

[0142] According to the probability multiplication formula, the joint distribution of X i , y b , b Converted to the form of probability, we can get:

[0143] Pr(X i = x i , Y i = y i , Y n～i = y n～i , B = b) = Pr(Y i = y i , Y n～i = y n～i , B = b|X i = x i ) · Pr(X i = x i ) (2)

[0144] Due to the discrete memorylessness of the BSC, the channel output at time t is independent of the channel outputs at other times; in addition, although the channel inputs at other times are correlated with their respective outputs, the channel inputs at other times are independent of the channel output at time t; from these conditional independence properties, we can obtain:

[0145] Pr(Y i = y i , Y n～i = y n～i , B = b|Xi = x i ) = Pr(Y n～i = y n～i , B = b | X i = x i ) · Pr(Y i = y i | X i = x i ) (3)

[0146] Substituting equation (3) into equation (2) gives:

[0147] Pr(Y i = y i , Y n～i = y n～i , B = b | X i = x i ) · Pr(X i = x i ) = Pr(X i = x i , Y n～i = y n～i , B = b) · Pr(Y i = y i | X i = x i )(4)

[0148] Equation (4) is written in vector form as:

[0149] where y n denotes the symbols at positions 1 - n, and y n～i denotes the side - information symbols not included at position i, i.e., the sequence

[0150] Summing both sides of equation (4) over x i gives:

[0151]

[0152] Performing a dimensional - expansion operation on equation (6) gives:

[0153]

[0154] Substituting equation (5) into equation (7) gives:

[0155]

[0156] Finally, the joint distribution of the symbols of the subsequence X2 can be calculated from equation (8).

[0157] The joint distribution in Equation (8) Similarly, in the case where scalar scaling is unimportant, it can be used interchangeably with the conditional distribution ; The conditional distribution can be estimated by constructing a bilaterally symmetric Context model from the symbols Y of the current side information sequence i and the symbols..., x (j-1)k , x jk , x (j+1)k ,... and..., y i-2 , y i-1 , y i+1 , y i+2 ... on its left and right; According to the bilaterally symmetric Context model, there is:

[0158]

[0159] Therefore, Equation (8) can also be written as:

[0160]

[0161] The joint distribution in Equation (10) Similarly, in the case where scalar scaling is unimportant, it can be used interchangeably with the conditional distribution ; represents the joint distribution of the current side information sequence symbol Y calculated using the left Context as and the right Context as i as templates.

[0162] represents the joint distribution of the current source sequence symbol X calculated using the left Context as and the right Context as i as templates.

[0163] Without loss of generality, taking the period k of extracting source sequence symbols as 3 and the size of the bilaterally symmetric Context model c as 2 as an example, as Figure 3 shown: The Context template of the bilaterally symmetric Context model is y i-2 , x (j-1)k , y i+1 , x jk , y i and x (j-1)k , y i-1 , x jk , y i+2 , y i ; By using the Context template for the current side information sequence symbol Y iStatistical analysis is performed for different values of

[0164] For the Context template y of the bilaterally symmetric Context model i-2 ,x (j-1)k ,y i+1 ,x jk ,y i and x (j-1)k ,y i-1 ,x jk ,y i+2 ,y i Initialize the count values of ; Context template y i-2 ,x (j-1)k ,y i+1 ,x jk ,y i and x (j-1)k ,y i-1 ,x jk ,y i+2 ,y i The count values are respectively initialized to:

[0165]

[0166] Using the Context template

[0167] y i-2 ,x (j-1)k ,y i+1 ,x jk ,y i and x (j-1)k ,y i-1 ,x jk ,y i+2 ,y i as a sliding window, traverse from the third position point of the sequence to the third - last position point of the sequence; for each combination identical to this Context template traversed by the sliding window, the count value corresponding to this Context template is incremented by 1.

[0168] After the traversal process ends, the conditional distribution can be obtained by performing calculations on the count values of the Context template in the following manner.

[0169]

[0170]

[0171] Conditional distribution Once known, the conditional distribution of the symbols of subsequence X2 can be calculated using Equation (10)

[0172] Since in Equation (10), this term may introduce negative components and thus affect subsequent operations; in order to eliminate the negative components, the following formula is used for processing:

[0173]

[0174] In step 107, the soft input soft output module takes the conditional distribution together with the syndrome S and the transition probability p of the BSC as inputs, and outputs the improved conditional distribution The specific operation steps of the soft input soft output module are carried out as follows:

[0175] Take the conditional distribution ( which means is specified to be the distribution calculated using the left Context as and the right Context as as templates) together with the syndrome S and send them into the BP algorithm; the conditional distribution will be used to initialize the BP algorithm; the BP algorithm outputs the posterior information

[0176] generated when the BP algorithm converges or reaches the maximum number of iterations Use the conditional distribution together with the transition probability p of the BSC to remove the internal information in the posterior information

[0177]

[0178] Then

[0179]

[0180] According to the total probability formula, the improved conditional distribution can be calculated by the following formula:

[0181]

[0182] represents the information after removing the internal information calculated by Equation (13), and it is in one-to-one correspondence with the bilateral Context symbols in where the bilateral Context symbols refer to the symbols adjacent to the current source symbol X i on both sides…,x (j-1)k ,x jk ,x (j+1)k ,… and …,y i-2 ,y i-1 ,y i+1 ,yi+2 …, the size of the context is determined by c.

[0183] Through this total probability formula, we have achieved the reuse of the information of adjacent symbols of the source to update the conditional distribution. Moreover, by repeatedly using the soft input soft output module, we can achieve multiple reuses of the correlation within the source.

[0184] For different correlations within the source and correlations between sources of the present invention, the BER performance of the solution of the present invention and the DSC solution based on LDPC codes is compared. The corresponding experimental parameters are as follows: regular LDPC codes, code rate R = 1 / 4, and the extraction periods k are 3, 7, and 20.

[0185] Figure 4 It shows a comparison graph of the Bit Error Rate (BER) performance curves of the method of the present invention and the DSC solution based on LDPC codes under different BSC crossover probabilities p with the transition probability δ = 0.05. Since the method of the present invention not only realizes the utilization of the correlation within the source but also realizes multiple utilizations of the correlation within the source, its BER performance curve is significantly lower than that of the DSC solution based on LDPC codes.

[0186] Figure 5 It shows a comparison graph of the Bit Error Rate (BER) performance curves of the method of the present invention and the DSC solution based on LDPC codes under different BSC crossover probabilities p with the transition probability δ = 0.15. In the case of δ = 0.15, the correlation within the source becomes weaker, making the BER performance curve of the method of the present invention Figure 4 significantly rise compared with the BER performance curve in

[0187] Aiming at the problem that the existing distributed source coding scheme based on channel codes only utilizes the correlation between sources without considering the correlation within the source, the proposed hybrid distributed source coding and decoding method of the present invention relates to communication data transmission, which not only effectively utilizes the correlation between sources to assist decoding, but also utilizes the correlation within the source multiple times to improve the decoding performance. The sender extracts the source sequence, divides the sequence into X1 and X2 subsequences, compresses X1 using an arithmetic encoder based on Context, and calculates the syndrome of X2 at the same time. The receiver reconstructs X1 using the compressed codeword, constructs a probability estimation model using X1 and the side information sequence Y to estimate the conditional distribution of the symbols of X2, and sends the estimated conditional distribution, together with the syndrome and the channel crossover probability p, into the soft input soft output module. This module outputs the improved conditional distribution, and finally performs joint decoding using the improved conditional distribution and the syndrome to reconstruct the original X2 subsequence.

[0188] The beneficial effects of the present invention are as follows:

[0189] At the decoding end, the probability estimation module fully utilizes the conditional distribution of source symbols generated by the internal correlation of the source sequence and the correlation between sources by constructing a bilateral symmetric Context model for the points to be encoded. These conditional distributions of source symbols are used to initialize the decoding algorithm, resulting in a significant improvement in the decoding performance of the DSC scheme based on channel codes. In addition, the soft input soft output module can also improve the conditional distribution of source symbols by reusing the internal correlation of the source sequence. The improved conditional distribution can be used again to initialize the decoding algorithm, further improving the decoding performance of the DSC scheme based on channel codes. With the improvement of the decoding performance, the method proposed by the present invention can complete the compression coding and decoding of the source sequence at a lower code rate, greatly improving the overall compression ratio of the system.

[0190] The various embodiments in this specification are described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0191] Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, there will be changes in the specific implementation manner and application scope according to the idea of the present invention. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A hybrid distributed source decoding method, characterized in that Including: Obtaining the extracted subsequence codeword of the source sequence to be decoded, the syndrome of the unextracted subsequence of the source sequence to be decoded, and the side information sequence of the source sequence to be decoded; Decoding the extracted subsequence codeword of the source sequence to be decoded to obtain a reconstructed extracted subsequence; Using a bilateral symmetric Context model, based on the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded, obtaining the conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded; Performing joint decoding according to the conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded and the syndrome of the unextracted subsequence of the source sequence to be decoded, to obtain a reconstructed unextracted subsequence; Combining the reconstructed extracted subsequence and the reconstructed unextracted subsequence to obtain the source sequence to be decoded.

2. The hybrid distributed source decoding method according to claim 1, wherein The step of using a bilateral symmetric Context model, based on the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded, to obtain the conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded specifically includes: Based on the bilateral symmetric Context model, the reconstructed extracted subsequence, and the side information sequence of the source sequence to be decoded, obtaining the conditional distribution of each symbol in the side information sequence of the source sequence to be decoded under the condition of knowing the reconstructed extracted subsequence and the side information sequence of the source sequence to be decoded; In the case where scalar scaling is unimportant, determining the conditional distribution of each symbol in the side information sequence of the source sequence to be decoded as the joint distribution of each symbol in the side information sequence of the source sequence to be decoded, and obtaining the joint distribution of each symbol in the unextracted subsequence of the source sequence to be decoded according to the joint distribution of each symbol in the side information sequence of the source sequence to be decoded; Obtaining the conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded according to the joint probability distribution and the joint distribution of each symbol in the unextracted subsequence of the source sequence to be decoded; the joint probability distribution is the joint probability distribution of the side information sequence and the reconstructed extracted subsequence of the source sequence to be decoded.

3. A hybrid distributed source decoding method according to claim 1, characterized in that, The step of performing joint decoding according to the conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded and the syndrome of the unextracted subsequence of the source sequence to be decoded, to obtain a reconstructed unextracted subsequence specifically includes: Processing the conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded according to the syndrome of the unextracted subsequence of the source sequence to be decoded and the transition probability of a binary symmetric channel, to obtain the processed conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded; Performing joint decoding according to the processed conditional distribution of each symbol in the unextracted subsequence of the source sequence to be decoded and the syndrome of the unextracted subsequence of the source sequence to be decoded, to obtain a reconstructed unextracted subsequence.

4. A hybrid distributed source decoding method according to claim 3, characterized in that, Processing the conditional distributions of the symbols in the unextracted subsequence of the to-be-decoded source sequence according to the syndromes of the unextracted subsequence of the to-be-decoded source sequence and the transition probability of the binary symmetric channel to obtain the processed conditional distributions of the symbols in the unextracted subsequence of the to-be-decoded source sequence, specifically including: For any symbol in the unextracted subsequence of the to-be-decoded source sequence, inputting the conditional distribution of the symbol and the syndromes of the unextracted subsequence of the to-be-decoded source sequence into the BP algorithm to obtain the posterior information of the symbol; Removing the internal information of the posterior information of the symbol according to the conditional distribution of the symbol and the transition probability of the binary symmetric channel to obtain the removed distribution; According to the bilateral symmetric Context model, using the removed distribution and the conditional distribution of the symbol to obtain the processed conditional distribution of the symbol.

5. A hybrid distributed source decoding method according to claim 3, characterized in that Jointly decoding according to the processed conditional distributions of the symbols in the unextracted subsequence of the to-be-decoded source sequence and the syndromes of the unextracted subsequence of the to-be-decoded source sequence to obtain a reconstructed unextracted subsequence, specifically including: Using an LDPC decoder to jointly decode according to the processed conditional distributions of the symbols in the unextracted subsequence of the to-be-decoded source sequence and the syndromes of the unextracted subsequence of the to-be-decoded source sequence to obtain a reconstructed unextracted subsequence.

6. A hybrid distributed source decoding system, characterized in that, Including: An acquisition module, configured to acquire the codeword of the extracted subsequence of the to-be-decoded source sequence, the syndromes of the unextracted subsequence of the to-be-decoded source sequence, and the side information sequence of the to-be-decoded source sequence; An arithmetic decoder, configured to decode the codeword of the extracted subsequence of the to-be-decoded source sequence to obtain a reconstructed extracted subsequence; A probability estimation module, configured to use a bilateral symmetric Context model to obtain the conditional distributions of the symbols in the unextracted subsequence of the to-be-decoded source sequence according to the reconstructed extracted subsequence and the side information sequence of the to-be-decoded source sequence; A reconstructed unextracted subsequence determination module, configured to jointly decode according to the conditional distributions of the symbols in the unextracted subsequence of the to-be-decoded source sequence and the syndromes of the unextracted subsequence of the to-be-decoded source sequence to obtain a reconstructed unextracted subsequence; A merging module, configured to merge the reconstructed extracted subsequence and the reconstructed unextracted subsequence to obtain the to-be-decoded source sequence.

7. A hybrid distributed source coding method, characterized in that, Applied to the hybrid distributed source decoding method according to any one of claims 1-5, the hybrid distributed source coding method includes: Obtaining a to-be-encoded source sequence; Determining the side information sequence of the to-be-encoded source sequence; Performing an extraction operation on the to-be-encoded source sequence to obtain an extracted subsequence and an unextracted subsequence of the to-be-encoded source sequence; Performing lossless compression on the extracted subsequence of the to-be-encoded source sequence to obtain the codeword of the extracted subsequence of the to-be-encoded source sequence; Encoding the unextracted subsequence of the to-be-encoded source sequence to obtain the syndromes of the unextracted subsequence of the to-be-encoded source sequence.

8. A hybrid distributed source coding method according to claim 7, characterized in that, The lossless compression of the extracted subsequence of the to-be-encoded source sequence to obtain the codeword of the extracted subsequence of the to-be-encoded source sequence specifically includes: Using an arithmetic encoder based on the Context model to perform lossless compression on the extracted subsequence of the to-be-encoded source sequence to obtain the codeword of the extracted subsequence of the to-be-encoded source sequence.

9. A hybrid distributed source coding method according to claim 7, characterized in that, The encoding of the unextracted subsequence of the to-be-encoded source sequence to obtain the syndrome of the unextracted subsequence of the to-be-encoded source sequence specifically includes: Using an LDPC encoder to encode the unextracted subsequence of the to-be-encoded source sequence to obtain the syndrome of the unextracted subsequence of the to-be-encoded source sequence.

10. A hybrid distributed source coding system, characterized in that, Applied to the hybrid distributed source decoding method according to any one of claims 1-5, the hybrid distributed source coding system includes: An acquisition module for acquiring a to-be-encoded source sequence; An side information sequence determination module for determining the side information sequence of the to-be-encoded source sequence; An extraction module for performing an extraction operation on the to-be-encoded source sequence to obtain the extracted subsequence and the unextracted subsequence of the to-be-encoded source sequence; A compression module for performing lossless compression on the extracted subsequence of the to-be-encoded source sequence to obtain the codeword of the extracted subsequence of the to-be-encoded source sequence; An encoding module for encoding the unextracted subsequence of the to-be-encoded source sequence to obtain the syndrome of the unextracted subsequence of the to-be-encoded source sequence.

Images