CN102638332A - Improved LT (language translation) decoding method - Google Patents

Abstract

The invention discloses an improved LT (language translation) decoding method, mainly solving a problem that a coded symbol stops decoding because a coded symbol with the degree of 1 does not exist in circulation iteration start in a present BP (error back propagation) decoding algorithm. The improved LT decoding method comprises the following steps of: (1) establishing a bidirectional bigraph according to corresponding relation of an input symbol and a coding symbol; (2) finding out a coded symbol with the degree of 1 in the bidirectional bigraph, and decoding a value of the input symbol; (3) updating the value of the coded symbol; (4) dropping the coded symbol; (5) finding out two coded symbols with degree difference of 1 in the bidirectional bigraph, and decoding a value of the input symbol; (6) updating the value of the input symbol; (7) degrading the coded symbol; and (8) circulating and decoding till the values of the all input symbols are decoded. According to the improved LT decoding method disclosed by the invention, decoding operation is carried out under the condition that an existing BP decoding algorithm stops decoding without receiving more coded symbols and the decoding efficiency is improved.

Description

A kind of improved LT interpretation method

Technical field

The invention belongs to field of channel coding in the communication system, relate to coding and decoding method, particularly a kind of improved LT interpretation method is used in the LT decoding under the application background of characteristics such as to have propagation delay time big, and packet loss is big, and link is easily broken.

Technical background

LT sign indicating number (Luby Transform Codes) belongs to digital fountain sign indicating number, is one type of novel forward error correction coding technology.The LT sign indicating number is first code word with the notion round Realization of digital fountain sign indicating number, is referred to as general erasure codes, is applicable to the erasure channel environment based on packet communication.The LT sign indicating number has adopted the thought of random coded, is not limited code of a kind of code check, is also referred to as no code check sign indicating number.The LT sign indicating number can produce unlimited many coded data packet by end in the source, can correctly recover source file until receiving terminal.Because its superior performance under erasure channel is propagated in multicast, broadcast communication, satellite communication, fields such as data security storage are with a wide range of applications.

The notion of LT sign indicating number is proposed in 2002 by people such as M.Luby, and it is unlimited from the coded data packet sequence that K the online coding of source data packet produces, each coded data packet equal some source data packet of independently choosing at random XOR and.As long as the recipient receives that wherein m coded data packet just can successfully be recovered whole K source data packet with high probability through decoding arbitrarily, thereby introduces certain decoding expense a=m/K-1.

A key property of LT sign indicating number is exactly the on-fixed code check, and it makes transmitting terminal can determine to send the length of data neatly, thereby makes full use of channel capacity.But have in the application of strict delay requirement in real-time video flow transmission etc., transmitting terminal can only send with relatively-stationary code check.In fact, for each original data block, coding side can only generate the coded identification of limited quantity, then to next initial data block encoding.Very poor when channel quality, when packet loss was serious, many recipients can not receive that enough coded identifications decode, and the decoding performance of receiving terminal can receive very big influence.

Along with the development of LT sign indicating number, the LT sign indicating number has been introduced the application background of characteristics such as to have propagation delay time big, and packet loss is big, and link is easily broken, and proposed scheme low-density check matrix L DPC sign indicating number and the cascade of LT sign indicating number.In these application backgrounds, use the LT sign indicating number neither to need handshake procedure, also do not need feedback channel, simplified communication protocol, as long as guarantee to receive the coded data packet that termination is received some, just can high probability reduction original document.Yet the problem that existing LT sign indicating number faces is: the coded identification number of transmission is limited; For limited bandwidth or the very poor recipient of channel conditions, do not receive enough coded data packet usually, consider from the angle of receiving terminal; If the coded identification number is limited; Can't receive enough degree again and be 1 packet, so just can't accomplish decoding, cause communication failure.

At present about the method for LT sign indicating number decoding, its effective decoding algorithm mainly contains two kinds of Gaussian elimination algorithm and iteration belief propagation methods.Gaussian elimination algorithm GE be one type to the general decoded mode of fountain sign indicating number, be applicable to various fountain sign indicating number code words.The shortcoming of GE algorithm is that decoding complexity is high, and its operand O (nk) increases along with the growth of block of information length k fast, the computing of long code in therefore not being suitable for.Belief propagation algorithm BP is also referred to as message pass-algorithm MP, for a kind of proper LT sign indicating number of distribution design of spending, adopt the BP decoding algorithm will on calculated performance, obtain bigger raising, but computational speed is difficult to improve for middle long code word.This is a kind of being widely used and effective LT decoding algorithm.The shortcoming of existing BP decoding algorithm is to be 1 coded identification in the initial not presence of loop iteration; Then decoding can't start, and is that 1 coded identification approach exhaustion iteration also will be terminated in the loop iteration moderate, the decoder for decoding failure; Receiving terminal continues reception information, gets into next round decoding.

Summary of the invention

The objective of the invention is to problem, a kind of improved LT code coding method is provided,, can continue decoding, improve the efficient of decoding need not receive under the more coded identification to above-mentioned prior art existence.

For realizing above-mentioned purpose, technical scheme of the present invention comprises the steps:

(1) sets up coded identification collection A, set up two-way bigraph (bipartite graph) according to the corresponding relation of coded identification collection A and incoming symbol collection B;

(2) utilize two-way bigraph (bipartite graph); The searching degree is a coded identification c of 1 from coded identification collection A; According to degree is 1 coded identification c and the annexation that the incoming symbol e of a fillet is arranged with it; Translating with spending is the value of unique incoming symbol e of linking to each other of 1 coded identification c, and said " degree " is exactly the bar number of the fillet that links to each other with coded identification;

(3) value with the incoming symbol e that translates has the value of the coded identification F of annexation to carry out XOR with it, obtains the updating value of coded identification F;

(4) according to the annexation of coded identification F and incoming symbol e, remove the fillet between coded identification F and the incoming symbol e, realize degree of falling coded identification F, and with the value of the updating value in the step (3) as the coded identification F behind the degree of falling;

(5) utilize two-way bigraph (bipartite graph); Degree of searching differs 1 two coded identification m and n again from coded identification collection A; M＞n carries out XOR to the value of these two coded identification m and n, translates the value of an incoming symbol z; A fillet is arranged between this incoming symbol z and the coded identification m, and this fillet is that fillet that coded identification m has more than coded identification n;

(6) value with the incoming symbol z that translates has the value of the coded identification H of annexation to carry out XOR with it, obtains the updating value of coded identification H;

(7) according to the annexation of coded identification H and incoming symbol z, remove the fillet between coded identification H and the incoming symbol z, realize degree of falling coded identification H, and with the value of the updating value in the step (6) as the coded identification H behind the degree of falling;

(8) value of the incoming symbol that translates of recording step (2) and step (5), if when having translated the value of all incoming symbols, then decoding finishes; If when not translating the value of all incoming symbols, then to coded identification repeating step (2)-(7) behind the degree of falling, up to the value that translates all incoming symbols.

The present invention compared with prior art has following advantage:

The present invention is a kind of improved LT interpretation method that on the basis of existing BP decoding algorithm, proposes; Utilize two-way bigraph (bipartite graph); Concentrate the searching degree to differ two coded identifications of 1 from coded identification, the value of these two coded identifications is carried out XOR, translate the value of an incoming symbol; Solved existing BP decoding algorithm owing to be the problem that 1 coded identification that coded identification causes stops to decipher in the initial not presence of loop iteration; Utilize this improved LT interpretation method under the situation that existing BP decoding algorithm stops to decipher, to decode again, and need not receive more coded identification, improved decoding efficiency.

Description of drawings

Fig. 1 is a general flow chart of the present invention;

Fig. 2 sets up the sketch map of two-way bigraph (bipartite graph) when deciphering for the present invention;

Fig. 3 is 1 coded identification and the sketch map that translates the value of an incoming symbol for the present invention concentrates the searching degree in coded identification;

The sketch map that Fig. 4 upgrades the value of coded identification for the present invention;

Fig. 5 is the sketch map of the present invention to coded identification degree of falling;

Fig. 6 differs two coded identifications of 1 for the present invention's degree of searching and translates the sketch map of the value of an incoming symbol;

The sketch map that Fig. 7 upgrades the value of coded identification for the present invention;

Fig. 8 is the sketch map of the present invention to coded identification degree of falling;

Fig. 9 circulates decoding up to the sketch map of the value that translates all incoming symbols for the present invention.

Embodiment

Below in conjunction with accompanying drawing and concrete execution mode the present invention is done further detailed explanation:

Embodiment of the present invention is: transmitting terminal sends 4 incoming symbols, and receiving terminal is received 5 coded identifications, adopts Robust degree distribution function.

Degree distribution function commonly used in the LT sign indicating number has ideality distribution function and Robust degree distribution function.

Defining ideal degree distribution function is:

$a\left(d\right)=\left\{\begin{array}{cc}1/K,& d=1\\ 1/(d^2-1),& d=2,...,K\end{array}\right.---1)$

Definition Robust degree distribution function is:

$R=c.\ln (K/h).\sqrt{K}---2)$

$\frac{1}{K-1}.\frac{\sqrt{K}}{\ln (K/h)}\≤c\≤\frac{1}{2}.\frac{\sqrt{K}}{\ln (K/h)}---3)$

$u\left(d\right)=\left\{\begin{array}{cc}\frac{R}{\mathrm{dK}},& d=1,...,(K/R)-1\\ \frac{R\ln R/c}{K},& d=K/R\\ 0,& D=K/(R+1),...,K\end{array}\right.---4)$

$b\left(d\right)=\frac{a\left(d\right)+u\left(d\right)}{\underset{d=1}{\overset{K}{\mathrm{\Σ}}}[a\left(d\right)+u\left(d\right)]}---5)$

Wherein: d is the value of the degree chosen, and h is the probability of the acceptable incoming symbol that can't recover, the constant of c for suitably choosing; K is the number of incoming symbol; R presentation code symbol degree is 1 mean value, and a (d) is the ideality distribution function, and b (d) is a Robust degree distribution function.

With reference to Fig. 1, performing step of the present invention is following:

Step 1 is set up two-way bigraph (bipartite graph).

Referring to Fig. 2, the concrete realization of this step is following:

The degree of 1a) choosing coded identification t1 according to Robust degree distribution function is 3, and the degree of coded identification t2 is 2, and the degree of coded identification t3 is 3, and the degree of coded identification t4 is 3, and the degree of coded identification t5 is 1, t1 wherein, and t2, t3, t4, t5 are 5 coded identifications;

1b) from 4 incoming symbols, choose 3 different incoming symbol s1, s2, s3 is as the fillet of coded identification t1; From 4 incoming symbols, choose 2 different incoming symbol s1, s2 is as the fillet of coded identification t2; From 4 incoming symbols, choose 3 different incoming symbol s1, s2, s3 is as the fillet of coded identification t3; From 4 incoming symbols, choose 3 different incoming symbol s2, s3, s4 is as the fillet of coded identification t4; From 4 incoming symbols, choose the fillet of 1 incoming symbol s4 as coded identification t5, s1 wherein, s2, s3, s4 are 4 incoming symbols;

1c) with coded identification t1 and corresponding 3 incoming symbol s1, s2, the limit of s3 connects, with coded identification t2 and corresponding 2 incoming symbol s1; The limit of s2 connects, with coded identification t3 and corresponding 3 incoming symbol s1, s2; The limit of s3 connects, with coded identification t4 and corresponding 3 incoming symbol s2, s3; The limit of s4 connects, and coded identification t5 is connected with the limit of corresponding 1 incoming symbol s4, constitutes two-way bigraph (bipartite graph).

Step 2, searching degree are a coded identification of 1, translate the value of an incoming symbol.

Referring to Fig. 3, the concrete realization of this step is following:

2a) finding out a degree is 1 coded identification t5, and the characteristic of this coded identification is: it only is connected with an incoming symbol s4, and the value of incoming symbol s4 equates with the value of coded identification t5;

2b) with the value of the coded identification t5 value as incoming symbol s4, t5=0 wherein obtains the value s4=0 of incoming symbol.

Step 3, coded identification obtains updating value.

Referring to Fig. 4, the concrete realization of this step is: according to incoming symbol s4 and coded identification t4, the relation of fillet is arranged between the t5; The value of the incoming symbol s4 that translates is added to coded identification t4, on the value of t5, obtains t4 '=t4+s4=1+0=1; T5 '=t5+s4=0+0=0

Wherein, what "+" represented is XOR, and t4 ' represents the updating value of t4, and t5 ' represents the updating value of t5.

Step 4 is to coded identification degree of falling.

Referring to Fig. 5, the concrete realization of this step is following:

4a) according to the relation that fillet is arranged between coded identification t4 ' and the incoming symbol s4, remove the fillet of coded identification t4 ' and incoming symbol s4, realize that the degree of coded identification t4 ' reduces to 0 by 1;

4b) according to the relation that fillet is arranged between coded identification t5 ' and the incoming symbol s4, remove the fillet of coded identification t5 ' and incoming symbol s4, realize that the degree of coded identification t5 ' reduces to 0 by 1.

Step 5, searching degree differ two coded identifications of 1, translate the value of an incoming symbol.

Referring to Fig. 6, the concrete realization of this step is following:

5a) find out two degree and differ 1, and have only an inconsistent coded identification of fillet, like coded identification t1 and t2, wherein the degree of t1 is 3, and the degree of t2 is 2, t1＞t2, and satisfaction differs 1, and t1 and t2 have only an inconsistent condition of fillet;

5b) value of two coded identifications finding out is carried out XOR, obtain the value of an incoming symbol,, obtain the value of incoming symbol s3, i.e. s3=0 as through t1+t2=0+0=0.

Step 6 is upgraded the coded identification value.

Referring to Fig. 7, the concrete realization of this step is: according to incoming symbol s3 and coded identification t1, t3, t4 ' have the relation of fillet; The value of the incoming symbol s3 that translates is added to coded identification t1, and t3 is on the value of t4 '; Obtain t1 '=t1+s3=0+0=0, t3 '=t3+s3=0+0=0, t4 '=t4 '+s3=1+0=1; Above-mentioned t1 ' represents the updating value of t1, and t3 ' represents the updating value of t3, t4 " represent the updating value of t4 '.

Step 7 is to coded identification degree of falling.

Referring to Fig. 8, the concrete realization of this step is following:

7a) relation of fillet is arranged according to incoming symbol s3 and coded identification t1 ', remove the fillet of coded identification t1 ' and incoming symbol s3, realize that the degree of coded identification t1 ' reduces to 2 by 3;

7b) relation of fillet is arranged according to incoming symbol s3 and coded identification t3 ', remove the fillet of coded identification t3 ' and incoming symbol s3, realize that the degree of coded identification t3 ' reduces to 2 by 3;

7c) according to coded identification s3 and coded identification t4 " relation of fillet is arranged, remove coded identification t4 " with the fillet of incoming symbol s3, realize coded identification t4 " degree reduce to 1 by 2.

Step 8, the circulation decoding.

The value s3=0 of the incoming symbol that the value s4=0 of the incoming symbol that recording step (2) translates and step (5) translate, at this moment owing to the value that does not translate all incoming symbols, thus the coded identification repeating step (2) of remainder is arrived step (7),

Referring to Fig. 9, the concrete realization of this step is following:

8a) finding out a degree is 1 coded identification t4 ", the characteristic of this coded identification is: it only is connected with an incoming symbol s2, and the value of incoming symbol s2 and coded identification t4 " value equate;

8b) with coded identification t4 " value as the value of incoming symbol s2, wherein t4 "=1, obtain the value s2=1 of incoming symbol, shown in Fig. 9 (a);

8c) according to incoming symbol s2 and coded identification t1 ', t2, t3 ', t4 " between the relation of fillet is arranged; the value of the incoming symbol s2 that translates is added to coded identification t1 ', t2, t3 ', t4 " value on; Obtain t1 "=t1 '+s2=0+1=1, t2 '=t2+s2=0+1=1, t3 "=t3 '+s2=0+1=1, t4 " '=t4 "+s2=1+1=0; Wherein, t1 " represent the updating value of t1 ', t2 ' represents the updating value of t2; t3 " represent the updating value of t3 ', t4 " ' represent t4 " and updating value, shown in Fig. 9 (b);

8d) according to incoming symbol s2 and coded identification t1 " relation of fillet is arranged, remove coded identification t1 " with the fillet of incoming symbol s2, realize coded identification t1 " degree reduce to 1 by 2;

8e) relation of fillet is arranged according to incoming symbol s2 and coded identification t2 ', remove the fillet of coded identification t2 ' and incoming symbol s2, realize that the degree of coded identification t2 ' reduces to 1 by 2;

8f) according to incoming symbol s2 and coded identification t3 " relation of fillet is arranged, remove coded identification t3 " with the fillet of incoming symbol s2, realize coded identification t3 " degree reduce to 1 by 2;

8g) according to incoming symbol s2 and coded identification t4 " fillet of ' relation of fillet is arranged, remove coded identification t4 " ' and incoming symbol s2, realize coded identification t4 " ' degree reduce to 0 by 1, shown in Fig. 9 (c);

8i) finding out a degree is 1 coded identification t3 ", the characteristic of this coded identification is: it only is connected with an incoming symbol s1, and the value of incoming symbol s1 and coded identification t3 " value equate;

8j) with coded identification t3 " value as the value of incoming symbol s1, wherein t3 "=1, obtain the value s1=1 of incoming symbol, shown in Fig. 9 (d).

Translate the value of all incoming symbols this moment, s1=1, s2=1, s3=0, s4=0, decoding finishes.

Claims (4)

1. an improved LT interpretation method comprises the steps:

(1) sets up coded identification collection A, set up two-way bigraph (bipartite graph) according to the corresponding relation of coded identification collection A and incoming symbol collection B;

(2) utilize two-way bigraph (bipartite graph); The searching degree is a coded identification c of 1 from coded identification collection A; According to degree is 1 coded identification c and the annexation that the incoming symbol e of a fillet is arranged with it; Translating with spending is the value of unique incoming symbol e of linking to each other of 1 coded identification c, and said " degree " is exactly the bar number of the fillet that links to each other with coded identification;

(3) value with the incoming symbol e that translates has the value of the coded identification F of annexation to carry out XOR with it, obtains the updating value of coded identification F;

(4) according to the annexation of coded identification F and incoming symbol e, remove the fillet between coded identification F and the incoming symbol e, realize degree of falling coded identification F, and with the value of the updating value in the step (3) as the coded identification F behind the degree of falling;

(5) utilize two-way bigraph (bipartite graph); Degree of searching differs 1 two coded identification m and n again from coded identification collection A; M＞n carries out XOR to the value of these two coded identification m and n, translates the value of an incoming symbol z; A fillet is arranged between this incoming symbol z and the coded identification m, and this fillet is that fillet that coded identification m has more than coded identification n;

(6) value with the incoming symbol z that translates has the value of the coded identification H of annexation to carry out XOR with it, obtains the updating value of coded identification H;

(7) according to the annexation of coded identification H and incoming symbol z, remove the fillet between coded identification H and the incoming symbol z, realize degree of falling coded identification H, and with the value of the updating value in the step (6) as the coded identification H behind the degree of falling;

(8) value of the incoming symbol that translates of recording step (2) and step (5), if when having translated the value of all incoming symbols, then decoding finishes; If when not translating the value of all incoming symbols, then to coded identification repeating step (2)-(7) of remainder, up to the value that translates all incoming symbols.

2. LT interpretation method according to claim 1, wherein the described corresponding relation according to coded identification collection A and incoming symbol collection B of step (1) is set up two-way bigraph (bipartite graph), carries out as follows:

1a) according to Robust degree distribution function, the number of degrees d that chooses each coded identification among the coded identification collection A at random;

1b) from incoming symbol collection B evenly picked at random d different incoming symbol as the fillet of each coded identification among the coded identification collection A;

1c) each coded identification among the coded identification collection A is connected with the limit of corresponding d different incoming symbol, constitute two-way bigraph (bipartite graph).

3. LT interpretation method according to claim 1; Wherein step (2) is described is 1 coded identification c and the annexation that the incoming symbol e of a fillet is arranged with it according to degree; Translating with spending is the value of unique incoming symbol e of linking to each other of 1 coded identification c; Be according to having only a fillet between coded identification c and the incoming symbol e, and the value equal relation of the value of incoming symbol e and coded identification c, with the value of coded identification c value as incoming symbol e.

4. LT interpretation method according to claim 1; Wherein the described value with the incoming symbol e that translates of step (3) has the value of the coded identification F of annexation to carry out XOR with it; Obtain the updating value of coded identification F; Be according to the relation that fillet is arranged between incoming symbol e and the coded identification F, the value of incoming symbol e and the value of coded identification F are carried out mould 2 and computing, its end value is as the updating value of coded identification F.

Images