CN1905374A - Coding and decoding method of asymmetric low dense check code - Google Patents

Abstract

The invention discloses a nonsymmetrical low-density check code coding and decoding method. It mainly uses a character that message node degrees of LDPC code are irregular to allocate nodes of high degrees to information bits and nodes of low degrees to check bits and besides, at the decoding end, it can stop iteration if the information bits converge and need not wait for all bits to converge.

Description

Asymmetric low-density check code coding/decoding method

Technical field

The present invention relates to the coding and decoding method of communications field channel error correction LDPC sign indicating number, be specifically related to a kind of coding and decoding method that can reduce the LDPC sign indicating number of iterations.

Background technology

Low-density block check code (Low Density Parity-Check Code, LDPC Code) is a kind of strong methods for forward error correction coding that rediscovers over past ten years, under the long code structural environment, approached shannon limit, thereby be considered to effective substitute technology of Turbo code, probably be used to next generation mobile communication and deep space communication.

At present, the research to the LDPC sign indicating number mainly concentrates on following several direction.The first, consider the structure of LDPC sign indicating number on non-GF (2), the encoded question on polynary territory just, as GF (4), GF (8) etc.Mackay and Davey etc. have made a lot of explorations and trial (Matthew C.Davey in this direction, PHD Thesis:Error-correction using Low-DensityParity-Check Code, Gonville and Caius College, Cambridge, 1999), obtained good achievement.Check matrix on the polynary territory of structure can make performance that very big raising is arranged meticulously.The second, the LDPC sign indicating number that Gallager proposes, the column weight of its check matrix and row are heavily fixed, and this is commonly called the LDPC sign indicating number (perhaps Gallager sign indicating number) of rule; Luby, Mitzenmacher, Shokrollahi and Spielman at first propose to construct irregular binary LDPC sign indicating number (Michael G.Luby, Michael Mitzenmacher, M.AminShokrollahi, and Daniel A.Spielman, " Improved Low-DensityParity-Check Codes Using Irregular Graphs, " IEEE TRANSACTIONSON INFORMATION THEORY, VOL.47, NO.2, FEBRUARY 2001:585-598).Luby proposed in 1998, loosened the restriction to ranks weight, constructed irregular LDPC sign indicating number, and just every row (every row) weight is inequality.Result of study shows that this is for initial Gallager sign indicating number, and the performance of irregular LDPC codes has also had very big raising.The abnormal LDPC code on the more excellent non-GF (2) of performance is sought in these two research directions constantly optimum organization at present.

The decoding algorithm of LDPC all is based on sum-product algorithm at present.This algorithm is slow with respect to the MAP algorithm the convergence speed of convolution code.Generally, the MAP decoding algorithm can be restrained after five or six iteration, and sum-product algorithm needs tens of times of iteration even just can restrain for up to a hundred times.The performance that the iteration complexity that how effectively to reduce sum-product algorithm does not reduce simultaneously it again becomes a big problem of LDPC sign indicating number practical application.

Simultaneously, the coding of irregular LDPC is not considered the asymmetric protection problem in the actual application usually, for information bit and check bit at random distribute to information node, do not utilize the characteristics of irregular LDPC codes.Consider the irregular LDPC codes of the systematic code mode in the practicality, the information bit before the coding needs more error protection usually, and check bit is mainly used in the protection to information bit.

In actual applications, require the error rate of the information bit before the coding enough low, and the error rate of check bit not to be us be concerned about.The characteristics of irregular LDPC codes self just in time provide convenience for the asymmetric protection of information bit and check bit.Simultaneously, the characteristics in conjunction with LDPC decoding needn't wait all bits all to restrain, and can stop decoding and only need information bit all to restrain.

Summary of the invention

Consider above problem, proposed the present invention.The objective of the invention is to propose a kind of asymmetric LDPC coding and decoding method, it utilizes the asymmetric error-correcting performance of irregular LDPC codes, improves the error correcting capability of system information position, reduces iterations at the decoding end simultaneously, shortens decoding time.

Therefore, in one aspect of the invention, proposed a kind of coding method of asymmetric loe-density parity-check code, comprised step: the check matrix H a that constructs non-rule [n, k] loe-density parity-check code; The check matrix H that the rearrangement that check matrix H a is listed as from small to large by column weight obtains resetting; Check matrix H with described rearrangement is encoded to the information code word S that imports, and wherein S is the column vector of k dimension.

In another aspect of this invention, propose a kind of interpretation method of asymmetrical loe-density parity-check code, comprised step: initialization step, preset the likelihood ratio Q of a plurality of threshold values and information bit i _IjStep of updating upwards is from likelihood ratio Q _IjComputing information node x _iState is a and verification formula A _jIn verification formula j is satisfied under the known condition of other information node distributions probability R _IjStep of updating is used probability R downwards _IjUpgrade likelihood ratio Q _IjThe trial and error decoding step, the pseudo-posterior probability of calculating bit i $E_i=P_i+\underset{j\∈\mathrm{col}\left[i\right]}{\overset{\mathrm{ij}}{\mathrm{\Σ}}}{R}_{\mathrm{ij}},$ To obtain deciphering vector x=(x ₁, x ₂..x _n), Pi represents the priori likelihood ratio of bit i; It is characterized in that described method also comprises step: that utilizes described a plurality of threshold values one of at least judges whether convergence with the hard decision value or the relevant soft value of information; And if iteration convergence, output code word x, otherwise the hard decision value and the relevant soft value of information of storing the information code word S of this iteration are used for the convergence judgement of next this iteration.

At the decoding end,, can stop decoding, reduce iterations and only need information bit all to restrain because sum-product algorithm needn't wait all bits all to restrain.

Description of drawings

Fig. 1 is the example of the LDPC sign indicating number represented with bigraph (bipartite graph).

Fig. 2 is the flow chart of asymmetric encoding.

Fig. 3 is improved sum-product algorithm realization flow figure.

Fig. 4 is the flow chart that the convergence of information bit position is judged.

Embodiment

Below in conjunction with the drawings and specific embodiments the present invention is described in further detail.

Fig. 1 has provided a simple examples representing loe-density parity-check code with bigraph (bipartite graph).The LDPC sign indicating number is a kind of based on sparse parity check matrix.1981, Tanner proposed to represent a low-density linear block codes with bigraph (bipartite graph) that from then on bigraph (bipartite graph) becomes the main tool of analyzing the LDPC sign indicating number.If a LDPC sign indicating number, information bit is long to be K, and code length is N, and check digit is M=N-K, and then the check matrix H of this sign indicating number is a matrix that size is M*N.The bigraph (bipartite graph) of H is expressed as follows: the N of a bigraph (bipartite graph) bottom node is represented N code word, becomes information node (massagenode); A M node in top is represented M verification formula, is called check-node (check node).When the check-node of the information node of bottom and top is present in same verification formula, just both are connected with limit (edge).The number of the line that will link to each other with each node is called the degree (degree) of this node.

The decoding of LDPC sign indicating number is adopted and long-pending (Sum-Product) algorithm, and whole decode procedure can be regarded the application of the BP algorithm on the bigraph (bipartite graph) of Tanner as.With Fig. 1 is example, and we are each check-node A father nodes (parent) of information node x, and each information node x is the child node (child) of check-node A.A displacement table information node (9) under the figure, a top cribbing point is represented check-node (6), and each node is represented the delegation's verification formula in the matrix H, is called a check bit.Nodes X 1, X4, X7 links to each other with node A1, has represented the first row verification formula.

In the iteration, the x node is activated afterwards q each time _Ij ^aPass to the A node that is attached thereto, a=1/0 as its confidence level.q _Ij ^aBe to remove A _jOuter x _iOn the information that other check-nodes that participate in provide, x _iConfidence level at state a.Node A _jBe activated afterwards r _Ij ^aPass to the x node that is attached thereto, a=1/0 as its confidence level.r _Ij ^aBe at information node x _iState is a and verification formula A _jIn under the known condition of other information node distributions, the probability that verification formula j satisfies.In each iteration, the confidence level of all nodes all obtains upgrading.When each iteration finishes, calculate { x _iPseudo-posterior probability e _i ^a, do and once attempt judgement, obtain adjudicating sequence Up to the judgement sequence Satisfy $H\hat{x}=0,$ Or iterations reaches the maximum that we preset, the iteration termination.Maximum iteration time can be made as ten times of average time.Because the existence of becate is arranged, decoding might converge to wrong code word, for will producing our usually said untestable fault (undetected error) in this case, but from general simulation result, the probability that such untestable fault occurs is very little.

[coding]

People's such as Luby research (Michael G.Luby, Michael Mitzenmacher, M.Amin Shokrollahi, and Daniel A.Spielman, " ImprovedLow-Density Parity-Check Codes Using Irregular Graphs; " IEEETRANSACTIONS ON INFORMATION THEORY, VOL.47, NO.2, FEBRUARY2001:585-598) show that the number of degrees of LDPC sign indicating number information node are high more, the corresponding check node is just many more, thereby error correcting capability is strong more, and is fast more in the convergence rate of iterative decoding process.Generally, the at first error correction of information node of high-order (number of degrees are maximum), and then be that the lower information node of exponent number is revised, and the general last correction of the information node of lowest-order.Thereby whole LDPC decoding presents a kind of " rough wave effect ", from the order of information node to the incremental error correction of low order node.Thereby whether reliably the number of degrees are to judge information node a kind of important measuring.The information node that the number of degrees are big can be better by correct decoding, and they offer the correct information of check-node again, and these check-nodes offer the less information node of the number of degrees to better information again.

For the irregular LDPC codes of the systematic code mode in the practicality, the information bit before the coding needs more error protection usually, and check bit is mainly used in the protection to information bit.In actual applications, require the error rate of the information bit before the coding enough low, and the error rate of check bit not to be us be concerned about.So, can earlier information node be sorted from big to small by its number of degrees, the information node that the number of degrees are big is distributed to coding information bit before then, and the little information node of the remaining number of degrees is distributed to check bit.

Below contrast Fig. 2 describes the LDCP coding method according to the embodiment of the invention in detail.

Fig. 2 shows the flow chart of asymmetric encoding.At first construct verification (supervision) the matrix H a (S201) of non-rule [n, k] LDPC sign indicating number.Ha is obtained H (S202) from little (left side) to the rearrangement that big (the right) is listed as by column weight.

Then, H is divided into the square formation A on (n-k) rank and (n-k) go the matrix B that k is listed as, satisfy AC+BS=0 (S203).Wherein C is the check code word of requirement, and it is the column vector of (n-k) dimension.And S is the information code word of input, and it is the column vector of k dimension.

Square formation A is carried out LU decompose A=LU, obtain a following triangle battle array L and upper triangular matrix U, L and U are the square formations (S204) on (n-k) rank.

Then, calculate Z=BS (S205).Separate linear equation LY=Z with the back to algorithm, obtain Y (S206).

Then separate UC=Y with the back to algorithm, obtain C (S207).

So, can obtain check code word C by information code word S, and the number of degrees of S are big, and the number of degrees of C are little.

[decoding]

As mentioned above, the information node that the number of degrees are big is distributed to the information bit before the coding, and the little information node of the number of degrees is distributed to check bit.So, aspect decoding,, generally be information bit convergence earlier, and restrain behind the check bit for sum-product algorithm.And in actual applications, more care be the convergence situation of information bit.So, after detecting information bit and all restraining, just can stop iterative decoding, and needn't wait for that all bits all restrain.Can reduce iterations and decoding delay like this, can not reduce the error rate of information bit simultaneously again.Describe interpretation method of the present invention in detail below with reference to Fig. 3 and Fig. 4.

1, initialization procedure (S301)

If what receive has noise cancellation signal for r is arranged _n, to n=1,2 ... .N, the decision value of every bit is initialized as r _nHard decision value x _nIf ${\mathrm{<figure-callout\; id="0"\; label="p\; i"\; filenames="A20051008492000141.png,A20051008492000151.png"\; state="\{\{state\}\}">p</figure-callout>}}_i^{}=P(x_i=0),$ ${\mathrm{<figure-callout\; id="1"\; label="p\; i"\; filenames="A20051008492000141.png,A20051008492000171.png"\; state="\{\{state\}\}">p</figure-callout>}}_i^{}=P(x_i=1)=1-{\mathrm{<figure-callout\; id="0"\; label="p\; i"\; filenames="A20051008492000141.png,A20051008492000151.png"\; state="\{\{state\}\}">p</figure-callout>}}_i^{}$ Be the prior information that channel provides before the iterative decoding, and note $P_i=\ln \left(\frac{p_i^0}{p_i^1}\right)$ The priori likelihood ratio that has shown information bit i.q _Ij ¹, q _Ij ⁰Be under the condition of other calibrating reliability ten-fours except that verification formula j, information bit t _i=1/0 probability, and note $Q_{\mathrm{ij}}=\ln \left(\frac{q_{\mathrm{ij}}^0}{q_{\mathrm{ij}}^1}\right)$ Shown its corresponding likelihood ratio.For certain information bit i, use $P_i=\ln \left(\frac{p_i^0}{p_i^1}\right)$ Initialization is also stored its all Q _Ijr _Ij ¹, r _Ij ⁰Be hypothesis information bit t _i=1/0 time, it is q that other and check digit j have 1/0 probability distribution of the information bit i ' of (just participating in this verification formula j) that the limit links _{I ' j} ¹And q _{I ' j} ⁰, the probability of check digit j=1 (j of verification formula just satisfies).And note $R_{\mathrm{ij}}=\ln \left(\frac{r_{\mathrm{ij}}^0}{r_{\mathrm{ij}}^1}\right)$ Show its corresponding likelihood ratio.e _i ¹, e _i ⁰The external information that to be information node calculate in iteration each time claims that it is the pseudo-posterior probability of bit i.And note $E_i=\ln \left(\frac{e_i^0}{e_i^1}\right)$ Show its corresponding likelihood ratio.Secondly, be defined as follows function: $f\left(x\right)=\ln \left(\frac{1+x}{1-x}\right),$ x∈(-1，1)

$g\left(x\right)=f^{-1}\left(x\right)=\frac{e^x-1}{e^x+1},$

x∈R

Usually, f (x), g (x), ln (x) and these complicated nonlinear functions of exp (x) use the method for tabling look-up to realize on hardware circuit, use fixing RAM in advance and store its corresponding functional value.

2, iterative process

Step1. upwards upgrade: (S302 upgrades R _Ij)

Pass to the R of receiving sequence bit i by verification formula j _IjBe information node x _iState is a and verification formula A _jIn under the known condition of other information node distributions, the probability that verification formula j satisfies.Have through derivation:

$R_{\mathrm{ij}}=f\left(\exp \left(\underset{i^{\&prime;}\&Element;\mathrm{row}\left[j\right]\backslash \left\{i\right\}}{\mathrm{\&Sigma;}}\ln \left(g\left(Q_{i^{\&prime;}j}\right)\right)\right)\right)$

Successively carried out g, ln, ∑, exp, five kinds of arithmetic operations of f, wherein ∑ is that hardware add operation and all the other four kinds all are table lookup operations.By the Q that has stored _{I ' j}Calculate and storage R _Ij

Step2 upgrades downwards: (S303 upgrades Q _Ij)

$Q_{\mathrm{ij}}=\underset{j^{\&prime;}\&Element;\mathrm{col}\left[i\right]\backslash \left\{i\right\}}{\mathrm{\&Sigma;}}{R}_{i{j}^{\&prime;}}$

By the R that has stored _{Ij '}Calculate and storage Q _Ij

Step3 trial and error decoding: (S304)

Next calculate the pseudo-posterior probability E of bit i according to following formula _i

$E_i=P_i+\underset{j\&Element;\mathrm{col}\left[i\right]}{\mathrm{\&Sigma;}}{R}_{\mathrm{ij}}$

Pseudo-posterior probability E _iBeing to be used for decision bit i when current iteration finishes, is 0 (1) possible probability.Can judge i bit=1 (0) according to symbol, obtain current decoding x _iAfter all bits are translated, obtain deciphering vector x=(x ₁, x ₂..x _n).

When the previous round iterative decoding finishes, attempt deciphering judgement, as follows:

If Hx=0 (S305) stops decoding so, output x=(x ₁, x ₂..x _n) as effective output valve (S306); Otherwise, if reach the iterations (S305) that presets, stop iteration so, output decode results (S306).

Otherwise (S305, not), according to iterative decoding information, for example the hard decision value or the relevant soft value of information are carried out the convergence judgement (S307) of information bit, if convergence stops iteration so, output decode results (S306); Otherwise the hard decision value of storing the information code word S of this iteration is changeed step1 then with the relevant soft value of information and is begun next round iteration (S308).

According to one embodiment of present invention, detect the convergence of information bit with the hard decision algorithm.Just, detect the hard decision result of the information bit of adjacent twice sum-product algorithm output, two results are got XOR, if different bits is then thought to stop computing by iteration convergence less than the certain proportion of total bit.

According to another embodiment of the present invention, detect the convergence of information bit, and proposed threshold criterion on the absolute sense and difference criterion on the relative meaning with the soft-decision algorithm.On physical concept, likelihood ratio has reflected the asymmetry of the posterior probability of 0,1 code element that decoding is adjudicated.Likelihood ratio is 0, illustrates that 0,1 code element etc. is general, is the most symmetrical situation, but that do the reliability of hard decision this moment is the poorest, and decoding is equivalent to throw guessing blindly as the coin.And the absolute value of likelihood ratio is big more, and 0,1 code element does not wait general (uneven more) more, illustrates the hard decision of this node reliable more.When the likelihood ratio absolute value was tending towards infinite, the hard decision value of decoder was almost set up with probability 1, and error correction is the most reliable.Generally, the number of degrees of information node are big more, and the likelihood ratio absolute value of its output is big more.And the hazardous noise of information node is more little, and the likelihood ratio absolute value of its output is big more.So the likelihood ratio absolute value is to judge that information node is whether reliably a kind of to measure more accurately.

Threshold criterion on the absolute sense is meant if the average of information bit likelihood ratio absolute value (perhaps minimum value) is higher than certain threshold value, then stops iteration.The equal value representation of absolute value the average degree of information bit decision reliability, and the minimum value of absolute value has been represented the information bit reliability of poor soft-decision.Generally, these two increasing functions that value is an iterations.When they during greater than threshold value, just show that the judgement of information bit is enough reliable, so can judge and restrain.

Threshold criterion on the relative meaning is meant if the difference of the average (perhaps minimum value) of the information bit likelihood ratio absolute value of adjacent twice iteration output less than certain threshold value, then stops iteration.The average of likelihood ratio absolute value and minimum value all are the increasing functions of the bounded of iterations.When iterative decoding was restrained, these two values all converged to some values, thereby the difference of the average (perhaps minimum value) of the information bit likelihood ratio absolute value of adjacent twice iteration output can go to zero.Thereby, the foundation that the difference on this relative meaning provides convergence to judge.

Fig. 4 utilizes above-mentioned judgment criterion to judge the constringent figure of information bit.At first, importing the likelihood ratio of this information code word S (k), and read the correlation of S last time (k-1), mainly is the hard decision value of S (k-1), the average La (k-1) and the minimum value Lb (k-1) of S (k-1) likelihood ratio absolute value; Threshold value r is set, M1, M2, N1, N2 (S401).The length of supposing the information code word is L.Make binary number F1, F2, F3, F4 and F5 are 0 (S402).Divide following five kinds of situations to judge:

Case1: relatively this information code word S (k) and S last time (k-1) that stored obtain the number d (S413) of both differences.If d＜rL then judges convergence, make F1=1, otherwise F1=0 (S414).Here, r is the positive number less than 1, and the r condition of the bright judgement of novel more is strong more.If r is fully little, have only the d=0 of working as so, that is to say that twice information code word is just the same and could judge convergence.So r is more little, performance is good more, but iterations also will increase.

Case2: the average La (k) that calculates this information code word S (k) likelihood ratio absolute value (S423).If La (k)＞M1 illustrates iteration convergence, make F2=1, otherwise F2=0 (S424).M1 is the positive number of presetting, and M1 is big more, and performance is good more, but iterations also will increase.

Case3: the minimum value Lb (k) that calculates this information code word S (k) likelihood ratio absolute value (S433).If Lb (k)＞M2 illustrates iteration convergence, make F3=1, otherwise F3=0 (S434).M2 also is the positive number of presetting.In like manner, M2 is big more, and performance is good more, but iterations also will increase.

Case4: the average La (k) that calculates this information code word S (k) likelihood ratio absolute value (S443) and reads the La (k-1) that has stored.If La (k)-La (k-1)＜N1 illustrates iteration convergence so, make F4=1, otherwise F4=0 (S444).N1 also is the positive number of presetting.N1 is more little, and performance is good more, but iterations also will increase.

Case5: calculate the minimum value Lb (k) of this information code word S (k) likelihood ratio absolute value, and read the Lb (k-1) that stored (S453).If Lb (k)-Lb (k-1)＜N2 illustrates iteration convergence so, make F5=1, otherwise F5=0 (S454).N2 also is the positive number of presetting.N2 is more little, and performance is good more, but iterations also will increase.

More than five kinds of judgment criterion can use separately, also can get several logic F=fun (F1, F2, F3, F4, F5), for example (with and, or or) combination.For example, get the logic OR relation of five kinds of criterions, this is the most weak convergence criterion of condition.Also can get the logical AND relation of five kinds of criterions, this is the strongest convergence criterion of condition.General, the condition of convergence is strong more, and performance is good more, but iterations is also many more.Thereby this is the balance compromise process of performance and complexity.

At last, the output of decision logic combination is F as a result, and F=1 represents convergence, and F=0 represents not restrain (S416).

The decode procedure of the above adaptive trace information bit of improvement sum-product algorithm energy in time stops the iteration for the information bit of reliable decoding, and needn't wait the decoding convergence of all bits, therefore greatly reduces decoding complexity.Simultaneously, at coding side, the number of degrees of information bit are big, and the check bit number of degrees are little, thereby the better error-correcting performance of the information bit that obtains like this, can significantly not reduce decoding performance.

The above; only be the embodiment among the present invention, but protection scope of the present invention is not limited thereto, anyly is familiar with the people of this technology in the disclosed technical scope of the present invention; the conversion that can expect easily or replacement all should be encompassed in of the present invention comprising within the scope.Therefore, protection scope of the present invention should be as the criterion with the protection range of claims.

Claims (10)

1, a kind of coding method of asymmetric loe-density parity-check code comprises step:

Construct the check matrix H a of non-rule [n, k] loe-density parity-check code;

The check matrix H that the rearrangement that check matrix H a is listed as from small to large by column weight obtains resetting;

Check matrix H with described rearrangement is encoded to the information code word S that imports, and wherein S is the column vector of k dimension.

2, coding method as claimed in claim 1 is characterized in that, the step that information code word S is encoded comprises:

H is divided into the square formation A on (n-k) rank and (n-k) go the matrix B of k row, satisfy AC+BS=0, wherein C is the check code word of requirement, and it is the column vector that (n-k) ties up;

Ask check code word C from equation AC+BS=0.

3, coding method as claimed in claim 2 is characterized in that, asks the step of check code word to comprise:

Square formation A is carried out LU decompose A=LU, obtain a following triangle battle array L and upper triangular matrix U, L and U are the square formations on (n-k) rank;

Utilize matrix B, L, U and S, obtain check code word to algorithm according to the back.

4, coding method as claimed in claim 3 is characterized in that, the number of degrees of information code word S are big, and the number of degrees of check code word C are little.

5, a kind of interpretation method of asymmetric loe-density parity-check code comprises step:

Initialization step presets the likelihood ratio Q of a plurality of threshold values and information bit i _Ij

Step of updating upwards is from likelihood ratio Q _IjComputing information node x _iState is a and verification formula A _jIn verification formula j is satisfied under the known condition of other information node distributions probability R _Ij

Step of updating is used probability R downwards _IjUpgrade likelihood ratio Q _Ij

The trial and error decoding step, the pseudo-posterior probability of calculating bit i $E_i=P_i+\underset{j\&Element;\mathrm{co}\left[i\right]}{\mathrm{\&Sigma;}}{R}_{\mathrm{ij}},$ To obtain deciphering vector x=(x ₁, x ₂..x _n), P _iThe priori likelihood ratio of expression bit i;

It is characterized in that described method also comprises step:

That utilizes described a plurality of threshold values one of at least judges whether convergence with the hard decision value or the relevant soft value of information; And

If iteration convergence, output code word x, otherwise the hard decision value and the relevant soft value of information of storing the information code word S of this iteration, the convergence that is used for next this iteration is judged.

6, interpretation method as claimed in claim 5 is characterized in that, described a plurality of threshold values comprise first threshold value, and the described step that restrains that judges whether comprises:

The information code word of the information code word of this iteration and the last iteration stored relatively obtains the number of both differences;

If the number of described difference is long-pending less than the length of first threshold value and information code word, then judge iteration convergence.

7, interpretation method as claimed in claim 5 is characterized in that, described a plurality of threshold values comprise second threshold value, and the described step that restrains that judges whether comprises:

Calculate the average of likelihood ratio absolute value of the information code word of this iteration;

If described average greater than second threshold value, is then judged iteration convergence.

8, interpretation method as claimed in claim 5 is characterized in that, described a plurality of threshold values comprise the 3rd threshold value, and the described step that restrains that judges whether comprises:

Calculate the minimum value of likelihood ratio absolute value of the information code word of this iteration;

If described minimum value greater than the 3rd threshold value, is then judged iteration convergence.

9, interpretation method as claimed in claim 5 is characterized in that, described a plurality of threshold values comprise the 4th threshold value, and the described step that restrains that judges whether comprises:

Calculate the average of likelihood ratio absolute value of the information code word of this iteration, and read the average of likelihood ratio absolute value of the information code word of the last iteration of having stored;

If the difference of the two less than the 4th threshold value, is then judged iteration convergence.

10, interpretation method as claimed in claim 5 is characterized in that, described a plurality of threshold values comprise the 5th threshold value, and the described step that restrains that judges whether comprises:

Calculate the minimum value of likelihood ratio absolute value of the information code word of this iteration, and read the minimum value of likelihood ratio absolute value of the information code word of the last iteration of having stored;

If the difference of the two less than the 5th threshold value, is then judged iteration convergence.

Images