WO2006034647A1

WO2006034647A1 - Variable length encoding and decoding method

Info

Publication number: WO2006034647A1
Application number: PCT/CN2005/001587
Authority: WO
Inventors: Zhong Luo; Jing Wang; Linhua Ma; Yilin Chang
Original assignee: Huawei Technologies Co., Ltd.
Priority date: 2004-09-29
Filing date: 2005-09-28
Publication date: 2006-04-06
Also published as: CN100484245C; CN1756352A

Abstract

The invention discloses a new variable length encoding and decoding method that is directed to resolve the VLC sequence number errors that tend to happen in prior arts. In the invention, at least two types of VLCs are used alternatively for variable length encoding when performing the signal source encoding, and the bitstreams are decoded by different VLCs alternatively according to the encoding arrangement rules when performing the signal source decoding. When the sync bit of one of the VLCs is wrong, if decoding is directly performed without correcting errors, the rule of alternative VLCs will be destroyed. At this time, posterior probability can be used to judge the maximum error probability, and the maximum posterior probability can be outputted as the decoding result, so the rule of alternative VLCs will be kept. In the invention, the inserting or deleting errors will not happen, and every VLC sequence number will not change.

Description

Variable length code encoding and decoding method

TECHNICAL FIELD The present invention relates to video codec techniques, and more particularly to a variable length codec method.

BACKGROUND In source entropy compression, compressing data using variable length code (VLC) is one of the most commonly used methods. Huffinan variable length coding for a given source and source symbol [Huffman DA, "A method for construction of minimum redundancy codes", Proc. IRE, Vol 40, pp. 1098-1101, Sep. 1952. ] You can get the shortest average code length and effectively compress the source. However, the Huffman code is very sensitive to the error code. Usually, a bit error will cause the subsequent variable length codes to be decoded correctly, that is, the start and end positions of the variable length code cannot be correctly located, so the variable length code is used when there is noise interference. error diffusion is very serious, when the bit error rate ^10--4, the variable length code symbol error rate as high as ^10-2 or more.

In the art, objects to be encoded in a source are often referred to as source symbols, such as motion vectors, macroblock patterns, etc., and are represented in Table 1 by Sl, S2, S3, S4, S5. The variable length code symbol refers to a binary bit string representing a "source symbol", for example, in Table 1, for a Huffinan code, the variable length code symbol representing the source symbol S1 is "00"; representing a source symbol The variable length code symbol of S2 is "01". The variable length code symbol error refers to any bit error in the variable length code symbol, and the symbol is an error, thereby counting the variable length code symbol error probability, not the statistical bit error probability. In view of the shortcomings of Huffman codes, there are many variable length code construction schemes, but for some applications, they still cannot meet the requirements.

Variable length code symbol errors include the following:

1) Substitute error, when there is a bit error, a symbol substitution error occurs in the variable length code, that is, one variable length code symbol is erroneously decoded into another variable length code symbol;

2) deleting the error, that is, two or more variable length codes are decoded into one variable length code;

3) Insertion error, that is, a variable length code is decoded into two or more variable length code symbols.

These three errors can be divided into two categories:

(1) The symbol of the variable length code is incorrect. It means that only the symbol substitution error occurs in the variable length code, and the variable length The sequence number of the code (that is, the sequence number of any variable length code in the data stream) has not changed, and is hereinafter referred to as the first type of error.

(2) The serial number of the variable length code is incorrect. It is caused by the insertion and deletion of the variable length code. That is, because one or more variable length codes are inserted or deleted, the subsequent variable length code number will be corresponding. The ground is increased or decreased, which is hereinafter referred to as the second type of error.

When an error occurs, some variable length codes will have both the first type and the second type of error; some variable length codes are decoded correctly, there is no first type of error, but because of the insertion and deletion errors in front of them, the serial number It will increase or decrease accordingly, that is, a second type of error will occur accordingly.

At present, other variable length code schemes proposed for the error diffusion effect of Huffman variable length codes are only the first type of errors that reduce variable length codes, such as fast sync comma code and reverse solvable long code (Reversible). VLC, hereinafter referred to as RVLC) [Takishima, M. Wada, and H. Murakami, "Reversible variable length codes," IEEE Trans. Commun" vol. 43, pp. 158-162, Feb. Mar./Apr. 1995. ], etc., are only the first type of error to solve the variable length code in a targeted manner.

Table 5 lists the Huffman code, comma code, and RVLC's 5 source symbol encoding examples.

Table 1 Example of 5 source symbol encoding for Huffman code, comma code and RVLC

As can be seen from Table 1, the Huffman code has the shortest average code length, but it is prone to error spread, resulting in the first and second types of errors of the variable length code. As can be seen from Figure 1, a bit error will cause many subsequent source symbols to be decoded incorrectly, and insert and delete errors will occur; the question mark in the figure indicates that the corresponding code table is exceeded and cannot be decoded correctly.

RVLC is characterized by being able to decode from both the forward and reverse directions. When a variable length code decoding error is found, it can be decoded from the reverse direction to recover the variable length code as much as possible. Although RVLC can decode in the reverse direction, in the decoding of the two directions, it is difficult to determine the error from the codeword itself. Where is it, as shown in Figure 3, it is difficult to determine which of the forward and reverse decoding errors in the question mark is a true error, so it is difficult to determine which codewords are truly correctly decoded codewords. In practical applications, the rationality of the codewords solved can usually be judged by the actual content of the codec.

In contrast, comma code has the function of fast synchronization. Only one of them is shown in Table 1. There are other types of comma codes, but the functions are similar. Although comma code overcomes the error diffusion effect of Huffman code, it often has insertion and deletion errors, and it is easy to produce the second type of error of variable length code. As shown in Figure 2, one bit in the source S5 is incorrect, so that S5 is decoded into S2 and S3, which is equivalent to an insertion error. At this time, although the right side of S1, S2, and S3 continue to decode normally, Their serial numbers will be incremented by 1, which is the second type of error.

As can be seen from the above, variable length codes with synchronization function, such as comma code, can effectively reduce the first type of error, that is, the symbol substitution error of the variable length code; however, if an insertion and deletion error occurs, all subsequent changes are made. The long code will have a serial number error, which is the second type of error, which is a major factor affecting the correctness of video decoding.

Summary of the invention

In view of the above drawbacks of the prior art, the present invention provides a new variable length code encoding and decoding method to solve the problem that the second type of error is easy to occur in the existing variable length code encoding and decoding method, thereby eliminating the variable length code. The serial number of the variable length code caused by the insertion and deletion error is incorrect.

The technical solution of the present invention is to provide a variable length code encoding and decoding method, which is used for encoding and decoding a source, and the method includes the following steps:

(1) When encoding a source, alternately encode at least two types of variable length codes;

(2) When decoding the source, the coded stream is subjected to a variable length code alternate code according to the rules at the time of encoding.

In the specific implementation of the above method, a L-type variable length code is used, L is a natural number greater than or equal to 2, and is identified as LI, L2 LL, and the source may be LI, L2 in any of the following alternate manners.

LL total L class variable length code alternate coding:

(3-1) The variable length code after each L1 type variable length code is L2 type variable length code, the variable length code after each L2 type variable length code is L3 type variable length code, and so on, until LL class Variable length code, the variable length code after each LL class variable length code is an L1 type variable length code;

(3-2) The variable length code after every two L1 class variable length codes is two L2 class variable length codes, each two The variable length code behind the L2 class variable length Ma is two L3 class variable length codes, and so on, until the LL class variable length code, the variable length code after each two LL class variable length codes is two L1 type variable length code;

(3-3) Each Ml L1 class variable length code is followed by M2 L2 class variable length codes, 3 M3 L3 class variable length codes after every M2 L2 class variable length codes, and so on, until ML LLs The class variable length code, 3 艮 Ml L1 class variable length codes after each ML LL class variable length code, wherein Ml, M2, and ML are natural numbers.

When using A and B variable length codes, the source can be alternately coded according to any of the following alternate modes:

(1-1) The variable length code after each class A variable length code can only be a class B variable length code, and the variable length code after each class B variable length code can only be a class A variable length code;

(1-2) Each of the two Class A variable length codes is followed by two Class B variable length codes, and each of the two Class B variable length codes is followed by two Class A variable length codes;

(1-3) Each M class A variable length code is followed by N class B variable length codes, and each N class B variable length code is followed by M [A class A variable length code, where M and N are natural numbers.

According to a preferred embodiment of the present invention, in the class A variable length code, "0" is used as the synchronization bit, "1" is used as the cutoff bit, and at least one information bit X is between any two adjacent sync bits. There is at least one information bit X between the cut-off bit and the previous sync bit, and the value of the information bit X can be "0" or "Γ; in the B-type variable length code, "1" is used as The sync bit, with "01" as the cutoff bit, has at least one information bit X before the cutoff bit and before any sync bit, and at least one information bit X between the adjacent two sync bits, in front of the cutoff bit There is at least one information bit X between the synchronization bits, and the value of the information bit X can be "0" or "1". Of course, the definition forms of the A and B variable length codes can also be interchanged.

In the step (1) of the method of the present invention, when encoding the source, the probability of the source symbol is further counted, and the source symbol with the highest probability of occurrence is represented by the shortest code; The source symbol with the second highest rate is represented by the same or the shortest length of the code. For the probability of occurrence: the source symbol is small, it is represented by the longest code.

In the step (2) of the method described in the third of the present invention, when the source is decoded, if the synchronization bit of a variable length code is in error, the maximum error of the variable length code is determined, and the one with the highest probability is used. The hometown of the error correction scheme is used as the decoding output. In the step (2) of the method of the present invention, when decoding the source, if the synchronization bit of a variable length code is in error, the posterior probability can be used to determine the maximum error probability of the variable length code, and the posterior error is used. The result of the most probable one is the decoded output.

In the codec scheme of the present invention, there are two possibilities for variable length code errors caused by bit errors, one is information bit error, which does not affect the synchronization of variable length code symbols; the other case is synchronization bit error, When the synchronization bit error occurs, if the decoding is directly performed without correcting, it will violate the rule that the A and B variable length codes should be alternately output; at this time, the posterior probability can be used to judge the maximum error probability, and the posterior probability is the largest. The output is decoded to maintain the rule that the A and B variable length codes should be alternately output, so that no insertion or deletion errors occur, and the serial number of each variable length code can remain unchanged, that is, the second type does not occur. error.

BRIEF DESCRIPTION OF THE DRAWINGS The invention will be further described with reference to the accompanying drawings and embodiments in which:

Figure 1 is a schematic diagram of error propagation of Huffman horses;

2 is a schematic diagram of error diffusion of a comma code;

Figure 3 is a schematic diagram of error propagation of the RVLC code.

detailed description

In the video codec scheme, such as MPEG2, MPEG4, H.264, etc., the syntax units of the coding, such as DCT coefficients, motion vector symbols, etc., have strict correspondence with the blocks and macroblocks of the image. The decoded variable length code not only requires the source symbol itself to be decoded correctly, but also requires that the sequence number of the source symbol be decoded correctly so that the video decoder can reconstruct the image correctly.

The present invention proposes a new variable length code encoding and decoding method, referred to herein as an alternate variable length code (Alternate VLC, referred to as AVLC), which alternates the source by at least two types of variable length codes. This alternate variable length code can not only quickly synchronize to reduce the first type of error of the variable length code, but also effectively overcome the second type of error of the variable length code. For example, in the specific implementation, the source and the two types of variable length codes can be alternately coded in the following alternate manner:

(1-1) The variable length code after each class A variable length code can only be a class B variable length code. The length code after each class B variable length code can only be a class A variable length code, that is, ABABAB. .., or BABABA...;

(1-2) Every two class A variable length codes are followed by two class B variable length codes, and each two class B variable length codes Followed by two class A variable length codes, namely AABBAABB..., or BBAABBAA...;

(1-3) After each M class A variable length code, ^ 个 N class B variable length codes, each N class B variable length code is followed by M class A variable length codes, where M and N are natural numbers, For example ABBABB..., or BAABAABAA..., etc.;

At the time of decoding, alternate decoding is performed in the same manner as the encoding.

In the three types of variable length codes A, B, and C, the source code can be alternately coded with the variable length codes of eight, B, and C in any of the following alternate ways:

(2-1) The white variable length code of each class A variable length code is a class B variable length code, and the variable length code after each class B variable length code is a class C variable length code, and each class C variable length code The latter variable length code is a class A variable length code;

(2-2) 3 艮 two class B variable length codes after every two class A variable length codes, each of the two class B variable length codes is followed by two class C variable length codes, and each two class C variable length codes Behind the two 变 class variable length code;

(2-3) Each M class A variable length code is followed by N class B variable length codes, and each N class B variable length code is followed by P class C variable length codes, after each P class C variable length code With M class A variable length code, which, ! ^ and P are natural numbers.

In fact, the key point of the above solution is to alternate the source code by using two types and two or more types of variable length codes. Therefore, the actual solution is not limited to the above examples, and there are many types, that is, many of the above examples can be used. The term of the variable length code is alternated, and the irregular alternating of multiple types of variable length codes can also be used. For example, when using variable length codes A, B, and C, the alternate modes that can be used are: AABCAABC, ABBCABBC, ABCCABCC, ABBCCCABBCCC, and the like. The invention can be implemented by using two types, three types or any plurality of types of variable length codes, and whether the rules of variable length codes are alternated or irregularly alternated, and the difference lies only in the complex complex of decoding, and the difference in effects, and the principle does not change. .

Specifically, the L-type variable length can be used, and L is a natural number greater than or equal to 1, and is identified as L1.

L2, LL, the LI, L2 LL, and L-type variable length codes can be alternately coded according to any of the following alternatives:

(3-1) The variable length code after each L1 type variable length code is L2 type variable length code, and the variable length code after each L2 type variable length code is L3 type variable length ^ horse, and so on, until LL The variable length code after each LL class variable length code is a LL class variable length code;

(3-2) The variable length code after each two L1 type variable length codes is two L2 type variable length codes, and the variable length codes after each two L2 type variable length codes are two L3 type variable length codes, Such a push, until the LL class becomes longer Code, the variable length code after every two LL class variable length codes is two L1 class variable length codes;

(3-3) Each Ml L1 class variable length code is followed by M2 L2 class variable length codes, each M2 L2 class variable length code is followed by M3 L3 class variable length codes, and so on, J! to ML LL type variable length code, each

ML LL class variable length codes are followed by Ml L1 class variable length codes, where Ml and M2 ML are natural numbers.

In the following description, the first method, that is, the rule of using two types of variable length codes A and B is used as an example, and the type A variable length code or the B type variable length code is composed of a synchronization bit and an information bit. Its structure is shown in Table 2 and Table 3:

Table 2 Structure of Class A Variable Length Code Table 3 Structure of Class B Variable Length Code

It can be seen from Table 2 and Table 3 that the "X" in the class A and class B variable length codes is an information bit, and even if the codewords have the same length, the information bits take different values ^1 to become different codewords. For example, for 0x1 in class A, when X is 0, it is 001, for code table number A1; when X is 1, it is 011, corresponding to code table number A2.

The difference between the class A and class B variable length codes is that "0" in the eight types of 3⁄4 long codes is the synchronization bit, "1" is the cutoff bit; "in the class B variable length code is the synchronization bit, and "01" is The cut-off bit; the serial number 0 of the B-type variable length code does not exist. From its shield, the A-type variable length code and the B-type variable length code are actually comma code, so it can effectively prevent the lengthening itself. The first type of error of the code. Of course, the definition of the A and B variable length codes can also be adjusted to each other.

Both types of variable length codes can be generated by an algorithm without storing a code table. The algorithm is as follows:

(1) Encoding algorithm

Know the code table number Code_number to find its information bit § Value Info

N = Int[log ₂ (Code_number+ 1 )] + 1;

Info = Code_number+l- 2 ^N_1 ;

Codeward = ® [Info] ₂ Ni ; Where N is the information bit length, Int[ ] indicates that the mantissa is rounded off, Info indicates the value of the information bit, Codeward is the code word, that is, the obtained binary variable length code symbol, [Info]^ indicates "Info""Information bits expressed in binary form a Class A or Class B variable length code according to the variable length code structure of Table 2 or Table 3.

(2) Decoding algorithm

The value of the known information bit Info and the information bit length N, the code table number Code_ number

Code_ number = 2 ^N + Info― 1;

In source coding, according to the statistical probability of the source symbol, the source symbol with the greatest probability is represented by the shortest code, and the code table is allocated by this principle; if the previous source is encoded with the class A code table, Then a source symbol should be encoded with a Class B code table, then a Class A code table, then a Class B code table, and so on. A and B variable length codes alternate.

If the following encoding result occurs, ...AoBiAsBzAsB^.. The corresponding codeword sequence is...1 001 00001 101 01001 01101 ... , (wherein, each subscript of A and B corresponds to Table 2 and Table 3 A corresponding code table number in the middle).

In the source decoding, according to the rules at the time of encoding, the encoded code stream is subjected to alternate decoding of eight types and IB variable length codes. The source decoding alternately decodes the output according to the class A and class B variable length codes. In order to effectively detect synchronization errors, it is possible to limit the use of the shortest length in the A and B class code tables, so that the ability to find synchronization errors is significantly improved. For example, for class A variable length codes, AQ does not participate in coding, and the shortest code length is encoded from the beginning; for class B variable length codes, 8 ₁ and does not participate in coding, and the shortest code length is encoded from B ₃ . This approach sacrifices coding efficiency in exchange for error resistance.

In the codec scheme of the present invention, there are two possibilities for variable length code errors caused by bit errors, one is information bit error, which does not affect the synchronization of variable length code symbols; the other case is synchronization bit error, synchronization bit After the error, there are cases where the rules for violating the two types of burst codes alternately output by the AVLC are found in the decoding, or there are cases where the variable length code is shorter than the shortest variable length code length. In the case of an error in the sync bit, the maximum posterior probability can be used to determine its maximum error probability and correct its error.

Coded data: ...

T N2005/001587 one 9 one

Channel error: 0x0x0x0x1x0x0x1 xlxlxlxlxlxOl

SA2 X SB1

Source decoding: 0x0x0x0x1 χθχθχΐ xlxlxlxlxlxOl

SA2 SB2

Decoding error correction method one: 0x0x0x0x1 xOxOxlxlxlxlxlxlxOl

SA3 SB1

Decoding error correction method 2: 0x0x0x0x1x0x0x1 xlxlxlxlxlxOl After the channel error occurs, the original transmitted codeword (..., 8 八 1, 881, ...), after the source decoding is (..., 8 八 2, , 881 ...), X is judged to be a class A code or a class B code, which violates the requirement of alternate output of A and B code words, so it can be determined that it is part of the previous class A code or the latter class B code. a part of. If X and SB1 codewords are combined and judged as SB2 codewords, then i people are consecutively having 2 sync codeword errors in the transmission process of SB2 codewords. However, if a part of the previous Class A code of X is determined, it is considered that only one synchronization code word ^; error occurred during the transmission. Obviously, the probability of judging X as part of the previous class A code is higher than the probability of determining that part of the latter B code is correct.

For example, when the channel error rate is 10 - ³ , the probability of occurrence of the decoding error correction method is Pi = (1-10 - ³ ) X CIO' ³ ) ² ; the probability of occurrence of the decoding error correction method 2 is P ₂ = (1-10- ³ ) ² X (10- ³ ). It can be seen that P^b Pt is about 1000 times larger (since the two decoding algorithms, only 3 bits are changing, so the above probability only considers 3 bits). According to the principle of decoding the maximum a posteriori probability, the SA3 and SB1 outputs should be selected, that is, the originally transmitted codewords (..., SA1, SB1, ...) become (..., SA3, SB1, ...). It can be seen that the alternate coding mode of the alternate variable length code and the structure of the codeword can effectively find and correct the error, not only synchronize the variable length code symbols, but also correct the bit errors of the synchronization bits.

If the soft information of the signal output value can be obtained from the demodulator, a more accurate posterior probability information can be obtained. Let the signal be BPSK modulation, and the soft information of the modulator output 1 = [1 ¹ ₁ 13⁄4 13⁄4 . . . ]^ ], I- G (-oo ₅ +oo), judges the output information sequence Si = [s _u s _i2 s _i3 ... s _in ], Sij E [-l , +l]. 'J :

P(S"R) =Π ^Ρ )

j=i

P(Sj| ) a p ^1σ — 3⁄4

P(S _k |R) ― ^e

By comparing the probability of soft information, it is possible to more accurately determine which combination is correct.

In the BPSK (Binary Phase Shift Keying) modulation mode, "1" indicates "0";"1" indicates "1". Let the transmitted information sequence be Si = [ _Sil s _i2 s _i3 ... s _in ], _Sij e[-l, +l] (due to the possible estimation of the transmitted information sequence, the subscript i is used to indicate ), through the additive Gaussian white noise (σ represents the variance of the noise) channel, the received signal is R = [Γ! Γ ₂ Γ ₃ ... r _n ], η e (- _∞ , + oo). P( _Sij | η) indicates the probability of judging s _y when the received signal is

Independence

It is judged that the Si sequence has a probability. J ⁼¹

Claims

Rights request

A variable length code encoding and decoding method for encoding and decoding a source, wherein the method comprises the following steps:

(1) When encoding a source, at least two types of variable length codes are alternately coded; (2) when decoding a source, variable length codes are alternately decoded according to rules at the time of encoding.

2. The variable length code encoding and decoding method according to claim 1, wherein a L-type variable length code is used, L is a natural number greater than or equal to 2, and the identifier is L1, L2 LL, and may be in any of the following alternate manners. Perform L1 and L2 LL common L-type variable length code alternate coding for the source: (1) The variable length code after each L1 variable length code is L2 variable length code, and the variable length after each L2 type variable length code The code is a L3 type variable length code, and so on, until the LL type variable length code, and the variable length code after each LL type variable length code is an L1 type variable length code;

(2) The variable length code after each two L1 class variable length codes is two L2 class variable length codes, the variable length codes after each two L2 class variable length codes are two L3 class variable length codes, and so on. Until the LL class variable length code, the variable length code after each two LL class variable length codes is two L1 class variable length codes;

(3) Each Ml L1 class variable length code is followed by M2 L2 class variable length codes, each M ² U class variable length codes followed by M3 L3 class variable length codes, and so on, until ML LL class changes Long code, each ML LL variable length code is followed by Ml L1 type variable length code, where Ml and M2 ML are natural numbers.

3. The variable length code encoding and decoding method according to claim 2, wherein A,

In the case of two types of variable length codes, the A and B types of variable length code alternate codes are used for the source in any of the following alternate ways:

(1) The variable length code after each class A variable length code can only be a class B variable length code, and the variable length code after each class B variable length code can only be a class A variable length code;

(2) Each of the two Class A variable length codes is followed by two Class B variable length codes, and each of the two Class B variable length codes is followed by two Class A variable length codes;

(3) Each M class A variable length code is followed by N class B variable length codes, and each N class B variable length code is followed by M class A variable length codes, where M and N are natural numbers.

The variable length code encoding and decoding method according to claim 3, wherein in the step (1), when encoding the source, In the class A variable length code, "0" is used as the synchronization bit, "1" is used as the cutoff bit, and at least one information bit x is between any two adjacent sync bits, and the one in front of the cutoff bit There is at least one information bit X between the synchronization bits, and the value of the information bit X may be "0" or "Γ; in the B-type variable length code, "1" is used as the synchronization bit, and "01" is used as the synchronization bit. The cutoff bit has at least one information bit X before the cutoff bit and before any sync bit, and at least one information bit x between the adjacent two sync bits, at least between the cutoff bit and the previous sync bit An information bit x, the value of the information bit X can be "0" or T.

The variable length code encoding and decoding method according to claim 3, wherein in the step (1), when encoding the source,

In the 变 variable length code, "1" is used as the synchronization bit, "01" is used as the cutoff bit, before the cutoff bit and before any sync bit, there is at least one information bit x, and the adjacent two sync bits There is at least one information bit X between there, and at least one information bit X between the cut-off bit and the previous sync bit, the information bit X may have a value of "0" or ";

In the class B variable length code, "0" is used as the synchronization bit, with "as the cutoff bit, at least one information bit X between any two adjacent sync bits, and the synchronization bit before the cutoff bit There is at least one information bit x between, and the value of the information bit X can be "0" or "1".

The variable length code encoding and decoding method according to any one of claims 1 to 5, wherein, when encoding the source in the step (1), for the source symbol having the highest probability of occurrence, It is represented by the shortest code; for the source symbol with the second highest probability of occurrence, it is represented by the same or the shortest length of the code, and for the source symbol with the least probability of occurrence, it is represented by the longest code.

The variable length code encoding and decoding method according to claim 6, wherein in the step (2), when the source is decoded, if the synchronization bit of a variable length code is in error, the length is determined to be longer. The maximum error of the code is possible, and the result of the one of the most probable error correction schemes is used as the decoded output.

The variable length code encoding and decoding method according to claim 6, wherein in the step (2), when the source is decoded, if the synchronization bit of a variable length code is in error, the posterior is used. The probability is used to judge the variable length code ^ maximum error possibility, and the result of the error correction scheme with the largest posterior probability is used as the decoding output.

The variable length code encoding and decoding method according to claim 8, wherein in the step (2), when decoding the source, the posterior probability information can be obtained by the following method, and the signal is BPSK modulation. , the soft information of the modulator output R = [r _{i r2} r ₃ ... r _n ], η (-∞, +∞), The output information sequence Si = [ _Sil s _i2 s _i3 ... s _in ], sy≡[-l, +l] , then: ^ e - .δ

-^(∑( _3⁄4 - ) ² )-(∑( _3⁄4 - ) ² )]

P(S _k | ) ― ^e

10. The variable length code encoding and decoding method according to claim 2, wherein

When the A, B, and C variable length codes are used, the source code may be alternately coded with the variable length codes of the eight, B, and C types according to any of the following alternate modes:

(1) The variable length code after each class A variable length code is a class B variable length code, and the variable length code after each class B variable length code is a class C variable length code, after each class C variable length code The variable length code is a class A variable length code;

(2) Every two class A variable length codes are followed by two class B variable length codes, and each two class B variable length codes are followed by two class C variable length codes, and each two class C variable length codes are followed by two A class variable length code;

(3) Each M class A variable length code is followed by N class B variable length codes, and each N class B variable length code is followed by P class C variable length codes, and each P class C variable length code is followed by M A class A variable length code, where 3⁄41 and P are natural numbers.