CN100499378C - Low-density odd-parity check-code decoding method adopting uniform quantizing variable range - Google Patents

Abstract

This invention relates to a decoding method applying low density odd-even check codes of variable sphere uniform quantization including: selecting a group of parameters for decoding, initializing the soft information of the ith code cell of the LDPC code, the outer information output to the bit node I from the check node and their iterative times, decoding the variable sphere uniform quantization to the initialized and quantized soft information and outer information, totalizing all soft information and outer information of all the bit nodes to get the decodes, carrying out the following hard decision to all decode results to get the output result: if the check sum of the output result sequence is zero, or the iterative times reach to the maximum times, then the decode is finished, otherwise the time is plus with 1 to continue the decode.

Description

Adopt the low density parity check code decoding method of variable range uniform quantization

Technical field

The invention belongs to communication channel decoding technique field, the effectively and fast numerical quantization method of a kind of employing low density parity check code (LDPC sign indicating number) correcting error of information channel when particularly adopting forward error control (FEC) technology to be used for transfer of data and storage.

Background technology

Data cause various mistakes through regular meeting in storage and transmission course.The reason that produces this mistake has synchronization loss, the multipath fading in the wireless transmission, the magnetic track in the magnetic storage in random noise, the demodulating process damaged etc.This burst error is generally that aperiodicity occurs and duration length is indefinite.Because the existence of these mistakes, limited the memory capacity of memory under the rate of information throughput under the specific bandwidth and the particular area greatly.Particularly in wireless multimedia transmission system, because lot of data will and be subjected in the serious channel that disturbs of various bursts with very high reliability transmission at limited bandwidth, this problem becomes more outstanding.

In order to solve the integrity problem in transfer of data and the storage, adopt the method for chnnel coding usually.In current existing channel coding method, the low density parity check code of Ti Chuing (LDPC sign indicating number) has the most powerful error correcting capability recently, has very strong application prospect.

The LDPC sign indicating number is a kind of binary packet sign indicating number, and this sign indicating number adopts the supersparsity matrix as check matrix.The number of nonzero element is very rare in every row in the matrix (every row), and the position is random distribution.For convenience of description, the number of nonzero element is the weight of this row (row) in the definition delegation (row).For the convenience of describing, adopt the distribution of weight formula to describe this matrix.The column weight amount of same class LDPC code check matrix distributes and can be expressed as with distributed:

$\mathrm{\λ}\left(x\right)=\underset{i=2}{\overset{d_v}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{x}^{i-1}---\left(2\right)$

λ in the formula _iExpression weight be i be listed in deal shared in the matrix, d _vValue for the maximum of column weight amount in the matrix.Equally, the capable distribution of weight of same class LDPC code check matrix adopts following formula to describe:

$\mathrm{\ρ}\left(x\right)=\underset{j=2}{\overset{d_c}{\mathrm{\Σ}}}{\mathrm{\ρ}}_j{x}^{j-1}---\left(3\right)$

ρ in the formula Chinese style _jExpression weight is the row of j shared deal in matrix, d _cMaximum for row weight in the matrix.Because the LDPC sign indicating number is block code, for any legal code word V, with the product of check matrix H be zero, i.e. HV ^T=0.By this check equations as can be known, the same code element of the only corresponding LDPC sign indicating number of the nonzero element of every row in the check matrix has formed a constraint that is equivalent to duplication code.For the ease of the description in the decode procedure, defining this restriction relation is a bit node, and the exponent number of node is the weight of these row.And the nonzero element of every row in the check matrix becomes a constraint that is equivalent to check code with pairing LDPC symbol mapped.It is a check-node that this verification of same definition is closed, and the exponent number of node is the weight of this row.Each nonzero element in the matrix had both participated in the restriction relation of bit node, had participated in the restriction relation of check-node again, thereby can to define the pairing pass of matrix nonzero element be " tie line " that links these two kinds of nodes.In iterative decoding process, decoder utilizes the restriction relation of pairing check-node of the row and column of matrix and bit node to carry out iterative decoding.In iterative process, at first utilize the restriction relation of bit node to decipher, the soft information that is input as the receiving sequence correspondence of each bit node (is probability that each symbol is got " 1 " is taken from right logarithm gained again divided by the probability of getting " 0 " value, comprise two information altogether, an information is which value former symbol most probable gets, and another information has then been represented the degree of reliability of this value) and relevant check-node in the output of last once iteration; Subsequently, the output of bit node is delivered to corresponding check-node by " tie line ", utilizes the restriction relation of check-node to decipher again.In this process, a kind of output of node becomes the input of another node, and nonzero element pairing " tie line " becomes " passage " of these two kinds of node input and output exchange messages in the matrix.

The supersparsity characteristic of check matrix has fully been used in the decoding of LDPC sign indicating number, restriction relation by bit node and check-node calculates and the output external information (external information i.e. the information about some code element values that obtains of the restriction relation of all other code elements that belong to a code word by code word, and adopting external information is positive feedback to occur in iterative process alternately.), and feed back mutually, carry out iterative decoding.

Current, the standard interpretation method of LDPC sign indicating number for and amass interpretation method.

This and long-pending interpretation method comprise the steps:

1) initialization:

(1) receiving terminal utilizes sequence of real numbers R ₁ ^N, to the soft information LLR (R of i code element of LDPC sign indicating number _i) be initialized as:

$\mathrm{LLR}\left(R_i\right)=\frac{2}{{\mathrm{\σ}}^2}{R}_i,1\≤i\≤N---\left(4\right)$

In the formula: σ ²Be the standard variance of interchannel noise,

(2) external information that check-node j is outputed to bit node i is initialized as zero, that is:

LLR(r _ij)＝0 (5)

In the formula: r _IjFor output to the external information of bit node i from check-node j;

(3) iterations is initialized as 1;

2) soft information and external information after the initialization in the step 1) are deciphered:

(1) the bit node i is output as after the external information of check-node j and soft information decoding:

$\mathrm{LLR}\left(q_{\mathrm{ij}}\right)=\underset{j\′\≠j}{\underset{j\′\∈\mathrm{Col}\left[i\right]}{\mathrm{\Σ}}}\mathrm{LLR}\left(r_{\mathrm{ij}\′}\right)+\mathrm{LLR}\left(R_i\right)---\left(6\right)$

Col[i in the formula] location sets of expression check matrix H i row nonzero element, q _IjBe external information from bit node i to check-node j;

(2) the external information decoding back output valve that check-node j is outputed to bit node i is:

$\mathrm{LLR}\left(r_{j,i}\right)=(-1)(\underset{i\′\∈\mathrm{Row}\left(j\right)}{\mathrm{\Π}}\mathrm{sgn}(-\mathrm{LLR}\left(q_{j,i\′}\right))\mathrm{\Ψ}(\underset{i\′\∈\mathrm{Row}\left(j\right)\backslash \left\{i\right\}}{\mathrm{\Σ}}\mathrm{\Ψ}\left(\left|\mathrm{LLR}\left(q_{j,i\′}\right)\right|\right)---\left(7\right)$

Row[j in the formula] location sets of expression check matrix H j capable nonzero element, and ψ (x) be a mathematical function of hyperbolic tangent function tanh (x) extension, as shown in the formula:

$\mathrm{\Ψ}\left(x\right)=-\log \left(\tanh \left(\frac{x}{2}\right)\right)---\left(8\right)$

3) the soft information of all bit nodes and external information summation being obtained decode results is:

$\mathrm{LLR}\left({\hat{v}}_i\right)=\underset{j\′\∈\mathrm{Col}\left[i\right]}{\mathrm{\Σ}}\mathrm{LLR}\left(r_{\mathrm{ij}\′}\right)+\mathrm{LLR}\left(R_i\right)---\left(9\right)$

4) resulting decode results is carried out following hard decision and obtains exporting the result:

${\hat{u}}_i=\left\{\begin{array}{cc}1& \mathrm{if\; LLR}\left({\hat{v}}_i\right)>0\\ 0& \mathrm{if\; LLR}\left({\hat{v}}_i\right)<0\end{array}\right.---\left(10\right)$

5) if output as a result sequence verification and be zero, or iterations reaches maximum iteration time, then decoding finishes; Otherwise iterations adds 1, changes step 2 over to).

Consider the realization on hardware (as FPGA, ASIC etc.), above-mentioned LDPC decoding algorithm must carry out numerical quantization and carry out the computing of limited precision.And at present comparatively general numerical quantization algorithm is that the soft information and the external information in the iterative process of input are carried out uniform quantization, and the step of uniform quantization is step 1) and the step 2 at above-mentioned decoding) in the quantization method that carries out of soft information, external information and external information decoding back output valve be:

To the quantification of the step-lengths such as value employing of output after soft information, external information and the external information decoding, promptly with quantizing element set { (2 ^Q-1-1) ,-2 ^Q-1..., 2 ^Q-1, (2 ^Q-1-1) value that differs minimum } with it represents, exceed ± (2 ^Q-1-1) value uses respectively ± and (2 ^Q-1-1) block, wherein q is a quantizing bit number.

Above-mentioned quantization method is the uniform quantization method, but uniform quantization brings the deficiency of following two aspects: the one, and the loss on the performance; The 2nd, quantizing bit number is higher.Because the hardware size of LDPC decoding and quantizing bit number are approximated to the exponential increase relation, need pay bigger cost on the hardware resource in order to reach preferable performance.

And non-uniform quantizing adopts on to the processing of formula (8) and does not wait step-length to quantize, taken into full account the nonlinear characteristic of formula (8), improved decoding performance to a certain extent, but bring other problem: the nonlinear transformation that is adopted in the non-uniform quantizing process causes formula (6) (7) to be not easy to adopt simple addition and displacement to realize, has increased the implementation complexity of hardware.

Summary of the invention

The objective of the invention is to overcome the deficiency in the numerical quantization method of existing low density parity check code decoding, a kind of low density parity check code decoding method of new employing variable range uniform quantization is proposed, the present invention can obtain better decoding performance with less quantizing bit number, reduce the resource and the scale of hardware significantly, and reached the decoding performance when not quantizing.

The low density parity check code decoding method of the employing variable range uniform quantization that the present invention proposes is characterized in that, selects one group of parameter (Q, k _Max, k _Shift, λ _Shift), wherein, Q is a quantizing bit number, k _MaxBe maximum iteration time, λ _ShiftFor switching the iterations of quantizing range and step-length, λ _ShiftFor quantizing range and step-length being carried out the factor of companding, use described parameter to decipher: this interpretation method may further comprise the steps:

1) initialization:

(1) receiving terminal utilizes sequence of real numbers

Soft information LLR (R to i code element of LDPC sign indicating number _i) carry out initialization and be quantified as:

Q wherein _fBe the shared bit number of the fractional part of Q;

(2) external information that check-node j is outputed to bit node i is initialized as zero and be quantified as Q bit zero, that is:

LLR(r _j，i)＝0，(j，i)∈{(m，n)|H _m，n＝1} (12)

(3) iterations is initialized as 1;

2) soft information and external information after initialization in the step 1) and the quantification are deciphered:

If iterations k＜k _Shift,

(1) external information and the soft information decoding of bit node i to check-node j also is output as after the quantification:

$\mathrm{LLR}\left(q_{j,i}\right)=T_v(\underset{j\&prime;\&Element;\mathrm{Col}\left(i\right)\backslash \left\{j\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(r_{j\&prime;,i}\right)+\mathrm{LLR}\left(R_i\right)),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1\}---\left(13\right)$

In the formula: T _v() is meant look up table operations, is total to Q bit bit wide, 2 ^QThe degree of depth, the numerical value of being deposited in the table is:

T _v(x)＝ψ(|x|) (14)

Wherein, $\mathrm{\&Psi;}\left(x\right)=-\log \left(\tanh \left(\frac{x}{2}\right)\right)---\left(15\right)$

The scope of its quantification is $[-(2^{Q-1}-1)/2^{Q_f},(2^{Q-1}-1)/2^{Q_f}],$ Quantization step is

(2) check-node j is outputed to the external information decoding of bit node i and quantize after output valve be:

$\mathrm{LLR}\left(r_{j,i}\right)=(-1)(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)}{\mathrm{\&Pi;}}\mathrm{sgn}(-\mathrm{LLR}\left(q_{j,i\&prime;}\right))T_u\left(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)\backslash \left\{i\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(q_{j,i\&prime;}\right)\right),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1\}---\left(16\right)$

Wherein

$\mathrm{sgn}\left(x\right)=\left\{\begin{array}{cc}+1,& x\&GreaterEqual;0\\ -1,& x<0\end{array}\right.---\left(17\right)$

In the formula, T _u() is meant look up table operations, is total to Q bit bit wide, 2 ^QThe degree of depth, the numerical value of being deposited in the table is:

T _u(x)＝ψ(x) (18)

The scope of its quantification is $[-(2^{Q-1}-1)/2^{Q_f},(2^{Q-1}-1)/2^{Q_f}],$ Quantization step is

As iterations k 〉=k _Shift,

(1) external information and the soft information decoding of bit node i to check-node j also is output as after the quantification:

$\mathrm{LLR}\left(q_{j,i}\right)=T_v(\frac{1}{{\mathrm{\&lambda;}}_{\mathrm{shift}}}*(\underset{j\&prime;\&Element;\mathrm{Col}\left(i\right)\backslash \left\{j\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(r_{j\&prime;,i}\right)+\mathrm{LLR}\left(R_i\right))),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1\}---\left(19\right)$

In the formula: T _v() is meant look up table operations, is total to Q bit bit wide, 2 ^QThe degree of depth, the scope of its quantification is $\left[\begin{array}{cc}-\frac{1}{{\mathrm{\&lambda;}}_{\mathrm{shift}}}\&times;(2^{Q-1}-1)/2^{Q_f}& \frac{1}{{\mathrm{\&lambda;}}_{\mathrm{shift}}}\&times;(2^{Q-1}-1)/2^{Q_f}\end{array}\right],$ Quantization step is

(2) check-node j is outputed to the external information decoding of bit node i and quantize after output valve be:

$\mathrm{LLR}\left(r_{j,i}\right)=(-1)(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)\backslash \left\{i\right\}}{\mathrm{\&Pi;}}\mathrm{sgn}(-\mathrm{LLR}\left(q_{j,i\&prime;}\right))T_u({\mathrm{\&lambda;}}_{\mathrm{shift}}*\left(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)\backslash \left\{i\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(q_{j,i\&prime;}\right)\right)),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1\}---\left(20\right)$

T _u() is meant look up table operations, is total to Q bit bit wide, 2 ^QThe degree of depth, the scope of its quantification is $\left[\begin{array}{cc}-{\mathrm{\&lambda;}}_{\mathrm{shift}}\&times;(2^{Q-1}-1)/2^{Q_f}& {\mathrm{\&lambda;}}_{\mathrm{shift}}\&times;(2^{Q-1}-1)/2^{Q_f}\end{array}\right],$ Quantization step is

3) the soft information of all bit nodes and external information summation being obtained decode results is:

$\mathrm{LLR}\left(q_i\right)=\underset{j\&prime;\&Element;\mathrm{Col}\left(i\right)}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(r_{j\&prime;,i}\right)+\mathrm{LLR}\left(R_i\right))(1\&le;i\&le;N)---\left(21\right)$

4) resulting decode results is carried out following hard decision and obtains exporting the result:

${\hat{x}}_i=\left\{\begin{array}{cc}1& \mathrm{if\; LLR}\left(q_i\right)\&GreaterEqual;0\\ 0& \mathrm{otherwise}\end{array}\right.(1\&le;i\&le;N)---\left(22\right)$

6) if output as a result sequence verification and be zero, or iterations reaches maximum iteration time, then decoding finishes; Otherwise iterations adds 1, changes step 2 over to).

Characteristics of the present invention and effect:

In the process of method of the present invention based on the LDPC iterative decoding, be revealed as statistics monotonically increasing trend with the absolute value of the output information of node along with the increase of iterations, the output information of long-pending node converges on these characteristics of zero gradually along with the increase of iterations, with node and long-pending node iterative process in adopt variable quantizing range and quantization step that iteration information is quantized respectively, to adapt in the iterative process and the variation tendency of the value of node and long-pending node.When iteration initial, to be arranged to the same with quantization step with the quantizing range of node and long-pending node, surpass the scope and step-length switching times of setting when iterations after, to all enlarge a factor with the quantizing range and the step-length of node output information, simultaneously the quantizing range and the step-length of long-pending node output information are all compressed a factor, continue iteration then and go down until the decoding end.

The present invention can obtain better decoding performance with less quantizing bit number, the resource and the scale of hardware have been reduced significantly, and reach decoding performance when not quantizing, aspect performance and complexity compromise, to significantly be better than the interpretation method of existing uniform quantization and non-uniform quantizing.

Embodiment

The low density parity check code decoding method of the employing variable range uniform quantization that the present invention proposes is described in detail as follows in conjunction with the embodiments:

The major parameter of the LDPC sign indicating number of present embodiment is: code length equals 2032 bits, and code check is

The column weight amount is distributed to be λ (x)=0.0039x+0.4961x ²+ 0.2500x ³+ 0.2500x ⁷, row distribution of weight formula is ρ (x)=0.0079x ⁶+ 0.9921x ⁷, one group of parameter that present embodiment is selected is: Q=4, k _Max=64, k _Shift=6, λ _Shift=2;

The selection principle of parameter is as follows: Q is a quantizing bit number, select 4～16 according to the constraint of the requirement of decoding performance and hardware resource between positive integer value, the smaller the better under the requirement of satisfying decoding performance; k _MaxBe maximum iteration time, select the value that varies in size according to the quality of the channel conditions of LDPC applied environment, correct with the iterative decoding that guarantees LDPC under condition of severe, its value is the integer value between 20～100; k _ShiftFor switching the iterations of quantizing range and step-length, to select with the variation tendency of iterations added value according to external information in the LDPC decode procedure, basic principle is to allow k _ShiftThe value of the external information of both sides can be easier to cut into two different spans, k _ShiftValue be k _Max

Between integer value; λ _ShiftFor quantizing range and step-length being carried out the factor of companding, according to k _ShiftThe statistics multiple relation of both sides external information value determines that its span is the real number value between 1.0～16.0.

The step that present embodiment utilization above-mentioned parameter is deciphered is as follows:

1) initialization:

(1) receiving terminal utilizes sequence of real numbers

Soft information LLR (R to i code element of LDPC sign indicating number _i) carry out initialization and be quantified as:

Q wherein _f=1; Quantizing range [3.53.5], quantization step 0.5

(2) external information that check-node j is outputed to bit node i is initialized as zero and be quantified as 4 bits zero, that is:

LLR(r _j，i)＝0000，(j，i)∈{(m，n)|H _m，n＝1} (24)

(3) iterations is initialized as 1;

2) soft information and external information after initialization in the step 1) and the quantification are deciphered:

If iterations k＜6,

(1) external information and the soft information decoding of bit node i to check-node j also is output as after the quantification:

$\mathrm{LLR}\left(q_{j,i}\right)=T_v(\underset{j\&prime;\&Element;\mathrm{Col}\left(i\right)\backslash \left\{j\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(r_{j\&prime;,i}\right)+\mathrm{LLR}\left(R_i\right)),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1\}---\left(25\right)$

In the formula: T _v() is meant look up table operations, totally 4 bit bit wides, 2 ⁴=16 degree of depth, the numerical value of being deposited in the table is:

T _v(x)＝ψ(|x|) (26)

Wherein, $\mathrm{\&Psi;}\left(x\right)=-\log \left(\tanh \left(\frac{x}{2}\right)\right)---\left(27\right)$

The scope of its quantification is [(2 ^4-1-1)/2 ¹, (2 ^4-1-1)/2 ¹], quantization step is 1/2 ¹

(2) check-node j is outputed to the external information decoding of bit node i and quantize after output valve be:

$\mathrm{LLR}\left(r_{j,i}\right)=(-1)(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)}{\mathrm{\&Pi;}}\mathrm{sgn}(-\mathrm{LLR}\left(q_{j,i\&prime;}\right))T_u\left(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)\backslash \left\{i\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(q_{j,i\&prime;}\right)\right),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1\}---\left(28\right)$

Wherein

$\mathrm{sgn}\left(x\right)=\left\{\begin{array}{cc}+1,& x\&GreaterEqual;0\\ -1,& x<0\end{array}\right.---\left(29\right)$

In the formula, T _u() is meant look up table operations, totally 4 bit bit wides, 2 ⁴=16 degree of depth, the numerical value of being deposited in the table is:

T _u(x)＝ψ(x) (30)

The scope of its quantification is [(2 ^4-1-1)/2 ¹, (2 ^4-1-1)/2 ¹], quantization step is 1/2 ¹

When iterations k 〉=6,

(1) external information and the soft information decoding of bit node i to check-node j also is output as after the quantification:

$\mathrm{LLR}\left(q_{j,i}\right)=T_v(\frac{1}{{\mathrm{\&lambda;}}_{\mathrm{shift}}}*(\underset{j\&prime;\&Element;\mathrm{Col}\left(i\right)\backslash \left\{j\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(r_{j\&prime;,i}\right)+\mathrm{LLR}\left(R_i\right))),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1\}---\left(31\right)$

In the formula: T _v() is meant look up table operations, totally 4 bit bit wides, 2 ⁴=16 degree of depth, the scope of its quantification is $\left[\begin{array}{cc}-\frac{1}{2}\&times;(2^{4-1}-1)/2^1& \frac{1}{2}\&times;(2^{4-1}-1)/2^1\end{array}\right],$ Quantization step is

(2) check-node j is outputed to the external information decoding of bit node i and quantize after output valve be:

$\mathrm{LLR}\left(r_{j,i}\right)=(-1)(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)\backslash \left\{i\right\}}{\mathrm{\&Pi;}}\mathrm{sgn}(-\mathrm{LLR}\left(q_{j,i\&prime;}\right))T_u({\mathrm{\&lambda;}}_{\mathrm{shift}}*\left(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)\backslash \left\{i\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(q_{j,i\&prime;}\right)\right)),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1\}---\left(32\right)$

T _u() is meant look up table operations, totally 4 bit bit wides, 2 ⁴=16 degree of depth, the scope of its quantification are [2 * (2 ^4-1-1)/2 ¹2 * (2 ^4-1-1)/2 ¹], quantization step is 2/2 ¹

3) the soft information of all bit nodes and external information summation being obtained decode results is:

$\mathrm{LLR}\left(q_i\right)=\underset{j\&prime;\&Element;\mathrm{Col}\left(i\right)}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(r_{j\&prime;,i}\right)+\mathrm{LLR}\left(R_i\right))(1\&le;i\&le;N)---\left(33\right)$

4) resulting decode results is carried out following hard decision and obtains exporting the result:

${\hat{x}}_i=\left\{\begin{array}{cc}1& \mathrm{if\; LLR}\left(q_i\right)\&GreaterEqual;0\\ 0& \mathrm{otherwise}\end{array}\right.(1\&le;i\&le;N)---\left(34\right)$

If output as a result sequence verification and be zero, or iterations reaches maximum iteration time, then decoding finishes; Otherwise iterations adds 1, changes step 2 over to).

The effect of present embodiment is described as follows:

Table 1 and table 2 have been listed a LDPC sign indicating number employing standard and long-pending decoding algorithm under the BIAWGN channel respectively, 4 bit uniform quantizations, 6 bit uniform quantizations and resulting decoding performance of numerical quantization method algorithm of the present invention and relevant hardware implementation complexity.By table 1 as seen, under the condition of low signal-to-noise ratio, both error-correcting performances are more or less the same; Under the condition of high s/n ratio, the error-correcting performance of quantization algorithm gained of the present invention is than 4 bit uniform quantizations, and 6 bit uniform quantizations will be got well, and are almost the same with non-quantized decoding performance.In addition, as known from Table 2, the decoding complexity of method of the present invention obviously reduces.The resource of its long-pending node processing unit of neutralization descends about 50% than the uniform quantization method resource needed of at least 6 bits of the needs that reach better performance is general.As seen, the present invention greatly reduces the hardware implementation complexity of LDPC in the decoding performance that has guaranteed LDPC.

Method of the present invention can be useful in the environment of various use LDPC, as the decoding of the LDPC in the wireless communication system, and the LDPC decoding of optical fiber and the LDPC decoding in the magnetic storage etc.

The performance of the different bit number numerical quantization of table 1. method under the BIAWGN channel

E b/N O(dB)	2.0475	1.7237	1.5144	1.4116	1.1103
						Do not quantize	4.646e-7	2.920e-5	4.878e-4	1.387e-3	1.085e-2
4 bit uniform quantizations	3.150e-5	3.503e-4	2.178e-3	5.181e-3	2.917e-2
						6 bit uniform quantizations	1.550e-6	7.068-5	8.060e-4	2.506e-3	2.097e-2
Algorithm of the present invention	4.383e-7	3.189e-5	5.528e-4	2.068e-3	1.725e-2

The FPGA of the various quantization methods of table 2 realizes the Resources list

Unit (LUT)	4 bit uniform quantizations	Algorithm of the present invention	6 bit uniform quantizations
				Degree be 2 with the node processing unit	28	30	68
Degree be 3 with the node processing unit	52	54	111
				Degree be 7 with the node processing unit	170	176	323
Degree is 7 long-pending node processing unit	145	159	323

Claims (1)

1, a kind of low density parity check code decoding method that adopts the variable range uniform quantization is characterized in that, selects one group of parameter (Q, k _Max, k _Shift, λ _Shift), wherein, Q is a quantizing bit number, k _MaxBe maximum iteration time, k _ShiftFor switching the iterations of quantizing range and step-length, λ _ShiftFor quantizing range and step-length being carried out the factor of companding, use described parameter to decipher: this interpretation method may further comprise the steps:

1) initialization:

(1) receiving terminal utilizes sequence of real numbers

, to the soft information LLR (R of i code element of LDPC sign indicating number _i) carry out initialization and be quantified as:

(1≤i≤N)(1)

Q wherein _fBe the shared bit number of the fractional part of Q;

(2) external information that check-node j is outputed to bit node i is initialized as zero and be quantified as Q bit zero, that is:

LLR(r _j，i)＝0，(j，i)∈{(m，n)|H _m，n＝1}

(3) iterations is initialized as 1;

2) soft information after initialization in the step 1) and the quantification and external information are carried out variable range uniform quantization decoding:

If iterations k＜k _Shift,

(1) external information and the soft information decoding of bit node i to check-node j also is output as after the quantification:

$\mathrm{LLR}\left(q_{j,i}\right)=T_v(\underset{j\&prime;\&Element;\mathrm{Col}\left(i\right)\backslash \left(j\right)}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(r_{j\&prime;,i}\right)+\mathrm{LLR}\left(R_i\right)),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1\}$

In the formula: T _v() is meant look up table operations, is total to Q bit bit wide, 2 ^QThe degree of depth, the numerical value of being deposited in the table is:

T _v(x)＝Ψ(|x|)

Wherein, $\mathrm{\&Psi;}\left(x\right)=-\log \left(\tanh \left(\frac{x}{2}\right)\right)$

The scope of its quantification is $[-(2^{Q-1}-1)/2^{Q_f},(2^{Q-1}-1)/2^{Q_f}],$ Quantization step is

(2) check-node j is outputed to the external information decoding of bit node i and quantize after output valve be:

$\mathrm{LLR}\left(r_{j,i}\right)=(-1)(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)}{\mathrm{\&Pi;}}\mathrm{sgn}(-\mathrm{LLR}\left(q_{j,i\&prime;}\right))T_u\left(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)\backslash \left\{i\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(q_{j,i\&prime;}\right)\right),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1)\}$

Wherein

$\mathrm{sgn}\left(x\right)=\left\{\begin{array}{cc}+1,& x\&GreaterEqual;0\\ -1,& x<0\end{array}\right.$

In the formula, T _u() is meant look up table operations, is total to Q bit bit wide, 2 ^QThe degree of depth, the numerical value of being deposited in the table is:

T _u(x)＝Ψ(x)

The scope of its quantification is $[-(2^{Q-1}-1)/2^{Q_f},(2^{Q-1}-1)/2^{Q_f}],$ Quantization step is

As iterations k 〉=k _Shift,

(1) external information and the soft information decoding of bit node i to check-node j also is output as after the quantification:

$\mathrm{LLR}\left(q_{i,j}\right)=T_v(\frac{1}{{\mathrm{\&lambda;}}_{\mathrm{shift}}}*(\underset{j\&prime;\&Element;\mathrm{Col}\left(i\right)\backslash \left\{j\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(r_{j\&prime;,i}\right)+\mathrm{LLR}\left(R_i\right))),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1\}$

In the formula: T _v() is meant look up table operations, is total to Q bit bit wide, 2 ^QThe degree of depth, the scope of its quantification is $\left[\begin{array}{cc}-\frac{1}{{\mathrm{\&lambda;}}_{\mathrm{shift}}}\&times;(2^{Q-1}-1)/2^{Q_f}& \frac{1}{{\mathrm{\&lambda;}}_{\mathrm{shift}}}\&times;(2^{Q-1}-1)/2^{Q_f}\end{array}\right],$ Quantization step is

(2) check-node j is outputed to the external information decoding of bit node i and quantize after output valve be:

$\mathrm{LLR}\left(r_{j,i}\right)=(-1)(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)\backslash \left\{i\right\}}{\mathrm{\&Pi;}}\mathrm{sgn}(-\mathrm{LLR}\left(q_{j,i\&prime;}\right))T_u({\mathrm{\&lambda;}}_{\mathrm{shift}}*\left(\underset{i\&prime;\&Element;\mathrm{Row}\left(j\right)\backslash \left\{i\right\}}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(q_{j,i\&prime;}\right)\right)),(j,i)\&Element;\left\{(m,n)\right|H_{m,n}=1)\}$

T _u() is meant look up table operations, is total to Q bit bit wide, 2 ^QThe degree of depth, the scope of its quantification is $\left[\begin{array}{cc}-{\mathrm{\&lambda;}}_{\mathrm{shift}}\&times;(2^{Q-1}-1)/2^{Q_f}& {\mathrm{\&lambda;}}_{\mathrm{shift}}\&times;(2^{Q-1}-1)/2^{Q_f}\end{array}\right],$ Quantization step is

3) the soft information of all bit nodes and external information summation being obtained decode results is:

$\begin{array}{cc}\mathrm{LLR}\left(q_i\right)=\underset{j\&prime;\&Element;\mathrm{Col}\left(i\right)}{\mathrm{\&Sigma;}}\mathrm{LLR}\left(r_{j\&prime;,i}\right)+\mathrm{LLR}\left(R_i\right))& (1\&le;i\&le;N)\end{array}$

4) resulting decode results is carried out following hard decision and obtains exporting the result:

${\hat{x}}_i=\left\{\begin{array}{cc}1& \mathrm{ifLLR}\left(q_i\right)\&GreaterEqual;0\\ 0& \mathrm{otherwise}\end{array}\right.$ (1≤i≤N)

5) if output as a result sequence verification and be zero, or iterations reaches maximum iteration time, then decoding finishes; Otherwise iterations adds 1, changes step 2 over to).