CN105978578A - Non-uniform quantification method of low-density parity-check code sum-product decoding operation numerical values - Google Patents

Abstract

The invention relates to a non-uniform quantification method of low-density parity-check code sum-product decoding operation numerical values and belongs to the communication channel decoding technical field. The method includes the following steps that: a group of multi-bit-represented uniform quantification parameters is selected for the soft information of the i-th code element of a low-density parity-check code, extrinsic information outputted from a bit node i to a check node j, and extrinsic information outputted from the check node j to the bit node i; and based on the uniform quantification parameters, the number of the quantification values of the three kinds of variables is gradually decreased sequentially until the required number of the quantification values of the three kinds of variables satisfies decoding performance requirements, and the quantification values constitute a set of three kinds of different non-uniform quantification parameters, and the three kinds of non-uniform quantification parameters are adopted to quantify the values of the three kinds of variables. With the non-uniform quantification method provided by the invention adopted, decoding performance achieved by 6-bit quantification bit width of a uniform quantification method can be achieved by 3-bit quantification bit width, and resources required by the implementation of decoding hardware can be effectively reduced. The method has a high application value.

Description

Low density parity check code and the non-uniform quantizing method of long-pending decoding operation numerical value

Technical field

The invention belongs to communication channel decoding technique field, particularly to based on low density parity check code and long-pending decoding fortune The non-uniform quantizing method of the value that counts.

Background technology

Along with the fast development of China's high-resolution earth observation systems, satellite earth observation precision promotes day by day, needs The data volume of transmission is skyrocketed through, and star ground data transmission rate demand has been risen to multiple gigabit per second by hundreds of MBPSs. And it being limited by the power of radio-frequency devices on star, star ground number passes the transfer rate of link and is risen to thousands of million by hundreds of MBPSs Bits per second is faced with huge challenge, needs the office that the chnnel coding using high-gain is nervous to alleviate junctor level surplus Face.

In existing channel coded method, low density parity check code (Low-Density Parity-Check code, Hereinafter referred to as LDPC code) have coding gain closest to the Shannon Coding Theory limit, simultaneously have error code flat low and Row decoding is suitable for the advantages such as high-speed transfer, has strong application demand in satellite data transmission field.But, existing based on FPGA core The ldpc decoder that sheet design realizes, speed is only hundreds of MBPS, will realize number in the fpga chip of resource-constrained The high-performance ldpc decoder of giga bits per second speed, needs to explore the side reducing LDPC realization of decoding complexity further Method.

Efficient iterative numerical quantifies the key being to realize the design of low complex degree ldpc decoder.The decoding that LDPC code is conventional Method is mainly based upon confidence spread and long-pending decoding algorithm, and with long-pending decoding algorithm on hardware realizes, to every one-step decoding Operation result will carry out numerical quantization process.The process of numerical quantization exactly a high accuracy numerical value set be mapped to one discrete The Finite Number set of range value, represents some quantized values with limited location bit number, and the quantized value of numerical value maximum is taken as maximum quantization Value, positive maximum quantization value and negative minimum quantization value represent quantizing range, and the difference of adjacent two quantized values is quantized interval, quantify Processing and represent the whole numerical value in an interval range with each quantized value, this interval is taken as the quantization district that this quantized value is corresponding Between, it is interval that quantized interval corresponding to maximum quantization value is taken as maximum quantization.The uniform quantization method amount of referring to the most generally used Changing the quantization method that interval is identical, its hardware implementation complexity is higher, and research high effective quantization method can be with lower hardware money Source cost reaches better performance, for solving message transmission rate and integrity problem in communication system, has extremely important Meaning.

LDPC code is a kind of block code, uses supersparsity random matrix as check matrix, and its matrix structure can be with two-way Tanner grid chart represents.Fig. 1 show a two-way Tanner figure, and in figure, each black lattice represents check matrix H The restriction relation of middle a line, corresponding LDPC code unit is constituted a pact being equivalent to check code by all nonzero elements of the most often going Bundle, this verification contextual definition is a check-node j.In figure, each black small circle represents the string in check matrix H The all nonzero elements of restriction relation, i.e. each column all correspond to the same code element in LDPC code word, and they are together with receiving symbol Constituting a constraint being equivalent to duplication code, this restriction relation is defined as a bit node i.Every in check matrix H Individual nonzero element not only participates in the restriction relation of check-node but also participates in the restriction relation of bit node, is mapped as in Tanner figure Connect " tie line " of two kinds of nodes.The output external information of check-node and bit node " links by both is internodal Line " transmission carry out mutual feedback iterative decoding.During decoding, first carry out the decoding of bit node.Each bit node connects Receive the corresponding code element Soft Inform ation of self-channel, i.e. correspondence code metasymbol and take the value that the probability of " 1 " obtains divided by the probability taking " 0 " Taking natural logrithm again, the most each bit node also receives the check-node output external information being connected with this node, utilizes this After a little information decode according to bit node restriction relation, bit node is exported external information and is delivered to connected school Testing node, the external information that each check-node utilizes transmission decodes according to check-node restriction relation, then decoding is tied The bit node being connected is given in fruit output.External information is transmitted between bit node and check-node back and forth along " tie line ", Complete the overall process of decoding.

It is as follows that LDPC code based on uniform quantization method and long-pending decoding algorithm realize flow process:

1) to decoding initialization:

Receiving terminal receives the sequence of real numbers R=[R of self-channel₀,R₁,…,R_N-1], wherein N is LDPC code code word size, Each for LDPC code code element is adjusted to the form of log-likelihood ratio, as the input Soft Inform ation of this LDPC code code element:

LLR(R_i)=2/ σ²R_i,1≤i≤N

Wherein σ²For interchannel noise standard variance.The each code element Soft Inform ation calculated is carried out (Q, Q_f) at uniform quantization Reason:

Wherein Q takes quantizing bit number, Q_fTake Soft Inform ation fractional part quantizing bit number, x code fetch unit Soft Inform ation numerical value.

2) bit node decoding is carried out:

Each bit node i utilizes each LDPC code code element Soft Inform ation of self-channel and the verification joint being attached thereto Point output external information decode, input and be output as " and " relation:

$q_j^{\left(n\right)}\left(i\right)=\underset{j^{\′}\≠j}{\underset{j^{\′}\∈Col\[i\]}{\Σ}}{r}_i^{\left(n\right)}\left(j^{\′}\right)+LLR\left(R_i\right)$

Wherein Col [i] represents in check matrix H the set of nonzero element position, r in the i-th row_i ⁽ⁿ⁾J () represents that n-th is repeatedly In generation, export the external information of bit node i from check-node j.If iteration for the first time, iterations n=1, under initial condition Check-node output external information is set to zero, i.e. r_i ⁽¹⁾(j)=0.The each bit node output external information calculated is carried out all Even quantification treatment, processing method and step 1) identical, wherein Q takes quantizing bit number, Q_fTake bit node output external information decimal Part quantizing bit number, x takes bit node output external information numerical value.

3) check-node decoding is carried out:

Each check-node j utilizes connected bit node output external information to decode, and inputs and exports Relation for " amassing ":

$r_i^{(n+1)}\left(j\right)={\mathrm{\Φ}}^{-1}\left(\underset{i^{\′}\≠i}{\underset{i^{\′}\∈Row\[j\]}{\Π}}\mathrm{\Φ}\right(q_j^{\left(n\right)}\left(i^{\′}\right)\left)\right)$

Wherein Row [j] represents the set of check matrix H jth row nonzero element position, and function phi (x) is defined as

The each check-node output external information calculated is carried out uniform quantization process, processing method and step 1) phase With, wherein Q takes quantizing bit number, Q_fTaking check-node output external information fractional part quantizing bit number, x takes check-node output External information numerical value.

4) decoding result hard decision is obtained hard decision result:

After nth iteration completes, at each bit node by soft to all input external informations and corresponding LDPC code code element Information is sued for peace and is obtained decoding result:

$q^{\left(n\right)}\left(i\right)=\underset{j\∈Col\[i\]}{\Σ}{r}_i^{\left(n\right)}\left(j\right)+LLR\left(R_i\right)$

This decoding result is carried out hard decision and obtains hard decision result:

5) if hard decision result meets verification relationIt is then legal-code, is made For final decoding result output, otherwise jump to step 2), iteration is increased to n=n+1, carries out next round iteration；If reaching to set Fixed maximum iteration time, then stop iteration, last iteration hard decision result is exported.

When above-mentioned LDPC code and long-pending decoding algorithm realize on hardware, for receiving terminal receiving sequence Soft Inform ation LLR (R_i)、 In iterative process, bit node i exports the external information of check-node j every timeCheck-node j exports bit node i External information r_i ⁽ⁿ⁾J (), is required for its numerical value doing quantification treatment and carrying out finite-precision arithmetic.The numerical quantization generally used Method is to wait the uniform quantization method of quantized interval, uniform quantization parameter (Q, Q_f) represent, Q represents uniform quantization figure place, Q_fTable Show quantization decimal digits, if numerical value to be quantified is beyond quantizing rangeThen according to real number symbol to be quantified respectively Take quantized valueIf without departing from quantizing range, then quantized value takesWherein x represents and treats Quantized value, [] represents the maximum integer less than its value.Uniform quantization method is to reach the performance close with non-quantization operations Need larger quantization scope and less quantized interval, it is desirable to represent all of quantized value with more bit number, firmly Part needs to take more resource on realizing.

During decoding, outside Soft Inform ation, check-node output external information and bit node output that receiving terminal receives Its numerical values recited of information is in different interval, and traditional (Q, Q_f) uniform quantization, to Soft Inform ation and the numerical value of two kinds of external informations Quantification treatment uses same quantizing range and quantized interval, i.e. Soft Inform ation and two kinds of external informations is used unified uniform quantization Parameter.Owing to bit node output external information is with the increase of iterations, rapidly converge near maximum quantization value, check-node Output external information, with the increase of iterations, slowly converges near maximum quantization value, and receiving sequence Soft Inform ation numerical value is in decoding During keep constant, so in given iterations, Soft Inform ation and two kinds of external informations are for each quantization of uniform quantization Interval sensitivity is different, and Soft Inform ation is high to specific quantized interval sensitivity, and bit node output external information does not quantify district to little Between sensitivity low, and check-node output external information low to big other quantized interval sensitivity.For ensureing decoding performance, generally require Select more quantized value, represent that the quantizing bit number of these quantized values is more, waste hardware resource, and quantized value is crossed the tightest Ghost image rings decoding performance, is therefore difficult to find between quantizing bit number and decoding performance compromise.

Summary of the invention

It is an object of the invention to overcome existing low density parity check code and the uniform quantization side of long-pending decoding operation numerical value The deficiency of method, proposes a kind of new low density parity check code (LDPC) and the non-uniform quantizing method of long-pending decoding operation numerical value, Can reach and the performance of higher bit uniform quantization parameter similar with less quantizing bit number, be substantially reduced hardware resource, tool There is stronger engineering practical value.

The low density parity check code of present invention proposition and the non-uniform quantizing method of long-pending decoding operation numerical value, its feature exists In the external information to code element Soft Inform ation, exporting bit node from check-node, from bit node exports check-node Information these three variable is respectively adopted different non-uniform quantizing parameters；First Soft Inform ation, the contrast to LDPC code i-th code element External information that special node i exports check-node j, the external information that check-node j exports bit node i select ratio more than a group The special uniform quantization parameter represented；Again based on this uniform quantization parameter, gradually reduce the quantization of described three kinds of variablees successively Value number, finally gives the quantized value of the three kinds of variable desirable number meeting decoding performance requirement, and these quantized values form three kinds The set of different non-uniform quantizing parameters, uses described different non-uniform quantizing parameter respectively to three kinds of variate-value amounts of carrying out Change

The method specifically includes following steps:

1) determined by emulation and meet the lowest evenness quantizing bit number of decoding performance requirement, obtain the even amount of Soft Inform ation Change parameter (Q_L,Q_fL), the uniform quantization parameter (Q of bit node output external information_q,Q_fq), check-node exports the equal of external information Even quantization parameter (Q_r,Q_fr), wherein Q_L,Q_q,Q_rRepresent the quantizing bit number of three kinds of information, Q respectively_fL,Q_fq,Q_frRepresent three respectively The quantizing bit number of the fractional part of the information of kind；

2) based on step 1) the uniform quantization parameter (Q that obtains_L,Q_fL),(Q_q,Q_fq) and (Q_r,Q_fr), the most progressively subtract Few LLR (R_i)、And r_i ⁽ⁿ⁾J the quantized value number of () is to arrivingIndividual, use Q_nonRepresent non-uniform quantizing value bit number, Q_nonValue is no more than Q_L,Q_q,Q_rThe positive integer of minima；First start and pass through performance simulation from maximum quantization value to determine successively The of each variable..., 1 non-uniform quantizing value, be embodied as step as follows:

2.1) determine and receive each code element Soft Inform ation LLR (R_i)Individual non-uniform quantizing value；The soft letter of code element Breath uniform quantization parameter (Q_L,Q_fLWill in)The arithmetic number of individual forward quantizations value is divided intoIndividual forward quantized interval,The negative realpartition of individual negative quantified value becomesIndividual negative sense quantized interval, positive smallest quantization interval and negative maximum quantization Between interval, the quantized value in region is zero；Merge interval for maximum quantization with time big quantized interval first, the quantization district that will merge In between, the intermediate value of smallest quantization interval is as theIndividual non-uniform quantizing value；By this non-uniform quantizing value, and do not merge The former uniform quantization value of quantized interval, and variableAnd r_i ⁽ⁿ⁾(the Q of (j)_q,Q_fq) and (Q_r,Q_fr) uniform quantization parameter, Carry out performance simulation, if decoding performance and step 1) in ternary when all using uniform quantization parameter the bit error rate in same quantity Level, then cancel current quantized interval and merge；Secondly maximum three quantized intervals are merged the non-uniform quantizing value obtained, again enter Row performance simulation, if the bit error rate is at the same order of magnitude when decoding performance and ternary all use uniform quantization parameter, then cancels this Secondary quantized interval merges, and by that analogy, is gradually increased the number merging quantized interval, until the N-1 time, maximum N number of quantization district Between merge after, when simulation performance all uses uniform quantization parameter than ternary, the bit error rate increases more than an order of magnitude, then cancel Current quantized interval merges, using the intermediate value of smallest quantization interval during the N-2 time merging as the of Soft Inform ationIndividual amount Change value, N is positive integer；

2.2) at Soft Inform ation LLR (R_i) use step 2.1) determine non-uniform quantizing value, external information r_i ⁽ⁿ⁾J () uses (Q_r,Q_fr) on uniform quantization parameter basis, use and step 2.1) and same procedure determine external information?Individual non-uniform quantizing value, by this non-uniform quantizing value, and does not merges quantized interval former uniform quantization value, step 2.1) LLR (the R determined_i) non-uniform quantizing value, and variable r_i ⁽ⁿ⁾(the Q of (j)_r,Q_fr) uniform quantization parameter, carry out performance simulation, with This analogizes, and the non-uniform quantizing parameter obtained is carried out performance simulation, obtain the intermediate value of smallest quantization interval after every time merging As?Individual quantized value；

2.3) use with step 2.1) same procedure determine external information r_i ⁽ⁿ⁾The of (j)Individual quantized value, with being somebody's turn to do Non-uniform quantizing value, and does not merges quantized interval former uniform quantization value, and LLR (R_i) use step 2.1) determine non-homogeneous Quantized value,Use step 2.2) the non-uniform quantizing value that determines, carry out performance simulation, minimum quantization district when obtaining merging Between intermediate value as r_i ⁽ⁿ⁾The of (j)Individual quantized value；

2.4) to fixed three variables L LR (R_i)、r_i ⁽ⁿ⁾(j)Individual non-uniform quantizing value is big Little carry out 1/Q respectively_fL、1/Q_fqAnd 1/Q_frThe non-uniform quantizing value being combined into three variablees after being increased or decreased of multiple is many Individual combination, carries out performance simulation, finally, is contrasted by all simulation performances each combination, selects three during optimal performance Individual variable non-uniform quantizing value is as newly determined LLR (R_i)、r_i ⁽ⁿ⁾(j)Individual non-uniform quantizing value；

2.5) step 2.1 is repeated)-2.4), determine LLR (R_i)、And r_i ⁽ⁿ⁾The of (j)..., 1 non-uniform quantizing value；

3) LLR (R after determining_i)、r_i ⁽ⁿ⁾(j)Individual non-uniform quantizing be worth to three kinds non-all Even quantized value set

The non-uniform quantizing value of three kinds of set is emulated, if decoding performance meets the requirements, as final non-homogeneous amount Change value, if decoding performance is undesirable, returns step 2) redefine the non-uniform quantizing value of ternary.

The feature of the present invention and effect:

During the method for the present invention is based on low density parity check code and long-pending decoding operation, Soft Inform ation and bit node are defeated Go out external information and check-node exports external information and is in the feature of different numerical range, use the uniform quantization represented with many bits Based on parameter, gradually reduce Soft Inform ation and the method for external information quantized value number successively, these three variable is respectively adopted not Same quantized value and quantized interval carry out quantification treatment, finally give the non-uniform quantizing parameter that lowest bit number represents.This Non-uniform quantizing method is for the different spy of quantized interval residing for two kinds of external informations during LDPC code code element Soft Inform ation and decoding Point, merges uniform quantization interval low for three kinds of variable sensitivitys respectively or deletes, can reduce quantized value to a greater extent Number, thus reduce the quantizing bit number being used for representing quantized value, it is possible to utilize each quantization bit to reach close to greatest extent The decoding performance of the uniform quantization parameter that former many bits represent.On the uniform quantization parameter basis that initial many bits represent, often One takes turns search the most successively by Soft Inform ation, bit node output external information, the maximum several uniform quantizations of check-node output external information Interval merging, removes the quantized value corresponding to the quantized interval that performance impact is insensitive, until obtaining Q_nonBit represent non-all Even quantization parameter.

The present invention can reach to be better than the even amount that many bits represent by the non-uniform quantizing parameter that less bit represents Change the decoding performance of parameter, decoding performance and complexity reach compromise, hence it is evident that reduce hardware resource, there is stronger application Prospect.

Accompanying drawing explanation

Fig. 1 is LDPC code two-way Tanner expression figure.

Fig. 2 is the flow chart of the non-uniform quantizing method realizing the present invention.

Detailed description of the invention

The non-uniform quantizing method based on low density parity check code and long-pending decoding operation numerical value that the present invention proposes combines Embodiment describes in detail as follows:

Non-uniform quantizing method such as Fig. 2 based on low density parity check code and long-pending decoding operation numerical value that the present invention proposes Shown in, the method specifically includes the external information, defeated from bit node to code element Soft Inform ation, exporting bit node from check-node The external information these three variable gone out to check-node is respectively adopted different non-uniform quantizing parameters；The choosing of non-uniform quantizing parameter Select based on the uniform quantization parameter that many bits represent, gradually reduce the quantized value number of described three kinds of variablees successively, finally Obtaining the quantized value of three kinds of variable desirable number, the collection of these quantized values is combined into non-uniform quantizing parameter, use different non-all Three kinds of variate-values are quantified by even quantization parameter respectively.

The method specifically includes following steps:

1) LDPC and the uniform quantization parameter of long-pending decoding operation numerical value are selected:

Determined the lowest evenness quantizing bit number meeting decoding performance requirement by emulation, obtain the uniform quantization of Soft Inform ation Parameter (Q_L,Q_fL), the uniform quantization parameter (Q of bit node output external information_q,Q_fq), check-node exports the uniform of external information Quantization parameter (Q_r,Q_fr), wherein Q_L,Q_q,Q_rRepresent the quantizing bit number of three kinds of information, Q respectively_fL,Q_fq,Q_frRepresent three kinds respectively The quantizing bit number of the fractional part of information；It is implemented as:

Sequence of real numbers R=[the R that receiving terminal receives₀,R₁,R₂,…,R_N-1], its i-th, i=0,1 ... i ..., N-1 code Unit Soft Inform ation LLR (R_i) quantized value be:

During decoding, nth iteration bit node i is exported the external information of check-node jQuantized value be:

During decoding, nth iteration check-node j is exported external information r of bit node i_i ⁽ⁿ⁾The quantized value of (j) For:

Wherein [] represents the maximum integer less than its value；

2) based on step 1) the uniform quantization parameter (Q that obtains_L,Q_fL),(Q_q,Q_fq) and (Q_r,Q_fr), the most progressively subtract Few LLR (R_i)、And r_i ⁽ⁿ⁾J the quantized value number of () is extremelyIndividual, use Q_nonRepresent non-uniform quantizing value bit number, Q_nonValue is no more than Q_L,Q_q,Q_rThe positive integer of minima；(owing to forward and negative sense quantized value are full symmetric, theIndividual quantized value is zero, thus for each variable non-uniform quantizing parameter it needs to be determined thatIndividual quantized value), pass through Performance simulation starts to determine successively the of each variable from maximum quantization value..., 1 non-homogeneous amount Change value, specifically comprises the following steps that

2.1) determine and receive each code element Soft Inform ation LLR (R_i)Individual non-uniform quantizing value；The soft letter of code element Breath uniform quantization parameter (Q_L,Q_fLWill in)The arithmetic number of individual forward quantizations value is divided intoIndividual forward quantized interval,The negative realpartition of individual negative quantified value becomesIndividual negative sense quantized interval, positive smallest quantization interval and negative maximum quantization district Between between, the quantized value in region is zero；Merge interval for maximum quantization with time big quantized interval first, the quantized interval that will merge The intermediate value of middle smallest quantization interval is asIndividual non-uniform quantizing value；By this non-uniform quantizing value, with the amount of merging Change interval former uniform quantization value, and variableAnd r_i ⁽ⁿ⁾(the Q of (j)_q,Q_fq) and (Q_r,Q_fr) uniform quantization parameter, enter Row performance simulation, if decoding performance and step 1) in ternary when all using uniform quantization parameter the bit error rate at the same order of magnitude, Then cancel current quantized interval to merge；Secondly maximum three quantized intervals are merged the non-uniform quantizing value obtained, re-start Performance simulation, if close when decoding performance all uses uniform quantization parameter with ternary, then cancel current quantized interval and merges, with This analogizes, and is gradually increased the number merging quantized interval, until the N-1 time, after maximum N number of quantized interval merges, simulation performance When all using uniform quantization parameter than ternary, the bit error rate increases more than an order of magnitude, then cancel current quantized interval and merge, When merging the N-2 time, the intermediate value of smallest quantization interval is as the of Soft Inform ationIndividual quantized value, N is positive integer；

2.2) at Soft Inform ation LLR (R_i) use step 2.1) determine non-uniform quantizing value, external information r_i ⁽ⁿ⁾J () uses (Q_r,Q_fr) on uniform quantization parameter basis, use and step 2.1) and same procedure determine external information?Individual quantized value, particularly as follows: merge interval for maximum quantization with time big quantized interval first, the quantized interval that will merge The intermediate value of middle smallest quantization interval is asIndividual non-uniform quantizing value, by this non-uniform quantizing value, with the amount of merging Change interval former uniform quantization value, step 2.1) LLR (R that determines_i) non-uniform quantizing value, and variable r_i ⁽ⁿ⁾(the Q of (j)_r,Q_fr) Uniform quantization parameter, carries out performance simulation, if same decoding performance and step 1) in ternary when all using uniform quantization parameter The bit error rate at the same order of magnitude, is then cancelled current quantized interval and is merged；Secondly three quantized intervals of maximum are merged, re-start Performance simulation, if the bit error rate is at the same order of magnitude when decoding performance and ternary all use uniform quantization parameter, then cancels current Quantized interval merges, by that analogy, until the N-1 time, after maximum N number of quantized interval merges, simulation performance is all adopted than ternary Increase more than an order of magnitude by bit error rate during uniform quantization parameter, then cancel current quantized interval and merge, the N-2 time is merged Time smallest quantization interval intermediate value conduct?Individual quantized value；

2.3) use with step 2.1) same procedure determine external information r_i ⁽ⁿ⁾The of (j)Individual quantized value, specifically For: merge interval for maximum quantization with time big quantized interval first, the centre of smallest quantization interval in the quantized interval that will merge It is worth as theIndividual non-uniform quantizing value, by this non-uniform quantizing value, and does not merges quantized interval former uniform quantization value, And LLR (R_i) use step 2.1) the non-uniform quantizing value that determines,Use step 2.2) non-uniform quantizing that determines Value, carries out performance simulation, if decoding performance and step 1) in ternary when all using uniform quantization parameter the bit error rate at same number Magnitude, then cancel current quantized interval and merge, and secondly being merged by three quantized intervals of maximum, re-starting performance simulation, if translating When code performance and ternary all use uniform quantization parameter, the bit error rate is at the same order of magnitude, then cancel current quantized interval and merge, By that analogy, until the N-1 time, after maximum N number of quantized interval merges, simulation performance all uses uniform quantization parameter than ternary Time the bit error rate increase more than an order of magnitude, then cancel current quantized interval merge, by the N-2 time merging time smallest quantization interval Intermediate value as r_i ⁽ⁿ⁾The of (j)Individual quantized value；

24) three variables L LR (R are being determined_i)、r_i ⁽ⁿ⁾(j)After individual non-uniform quantizing value, to this The of three variableesThe size of individual non-uniform quantizing value carries out 1/Q respectively_fL、1/Q_fqAnd 1/Q_frThe increase of multiple or It is combined into multiple combinations of the non-uniform quantizing value of three variablees after minimizing, each combination is carried out performance simulation, finally, by institute There is simulation performance to contrast, select three variable non-uniform quantizing values during optimal performance as the LLR (R finally determined_i)、r_i ⁽ⁿ⁾(j)Individual non-uniform quantizing value；

Detailed description of the invention can be: first the first two variable (for example, LLR (R_i)、) theIndividual non-homogeneous Quantized value keeps constant, the 3rd variable (for example, variable r_i ⁽ⁿ⁾(j)) theIndividual non-uniform quantizing value increases 1/Q_fr's 1-4 times of size, carries out performance simulation；Secondly the first two variable theIndividual non-uniform quantizing value keeps constant, by the 3rd VariableIndividual non-uniform quantizing value reduces 1/Q_fr1-4 times of size, then carry out performance simulation；Again, first change Flow controlIndividual non-uniform quantizing value keeps constant, by second variable theIndividual non-uniform quantizing value increases by 1/ Q_fq1-4 times of size, the 3rd variableIndividual non-uniform quantizing value is increased or decreased 1/Q_fr1-4 times of size, enter Row performance simulation；Again, first variableIndividual non-uniform quantizing value keeps constant, by second variable theIndividual non-uniform quantizing value reduces 1/Q_fq1-4 times of size, the 3rd variableIndividual non-uniform quantizing value increases Add deduct few 1/Q_fr1-4 times of size, carry out performance simulation；Again, first variableIndividual non-uniform quantizing value Increase 1/Q_fL1-4 times of size, by second variableIndividual non-uniform quantizing value is increased or decreased 1/Q_fq1-4 times Size, the 3rd variableIndividual non-uniform quantizing value is increased or decreased 1/Q_fr1-4 times of size, carry out performance imitate Very；Again, first variableIndividual non-uniform quantizing value reduces 1/Q_fL1-4 times of size, by second variableIndividual non-uniform quantizing value is increased or decreased 1/Q_fq1-4 times of size, the 3rd variableIndividual non-homogeneous Quantized value is increased or decreased 1/Q_fr1-4 times of size, carry out performance simulation；Finally, all simulation performances are contrasted, select Ternary non-uniform quantizing value during optimal performance is as the LLR (R after updating_i)、r_i ⁽ⁿ⁾(j)Individual non- Uniform quantization value.Put again or under contracting uniform quantization parameter, the 1-4 times of size combining of quantized interval is determined, select performance Excellent parameter group cooperation is the LLR (R finally determined_i)、r_i ⁽ⁿ⁾(j)Individual non-uniform quantizing value.

2.5) step 2.1 is repeated)-2.4), determine LLR (R_i)、And r_i ⁽ⁿ⁾The of (j)..., 1 non-uniform quantizing value；(when determining each non-uniform quantizing value of ternary, will Performance simulation is carried out, when proceeding to determine Soft Inform ation LLR (R on previous step fixed non-uniform quantizing value basis_i), outer InformationAnd external information r_i ⁽ⁿ⁾During the 1st quantized value of (j), determine that the 1st quantifies ginseng merging maximum several quantized intervals Number time, the quantized value not merging quantized interval takes 0, using in most preferably merging method merge smallest quantization interval intermediate value as 1st non-uniform quantizing value of ternary.)

3) at LLR (R_i)、r_i ⁽ⁿ⁾(j)Individual non-uniform quantizing value just obtained after all determining non-all Even quantization parameter (Q_nonun-L,Q_nonun-q,Q_nonun-r), wherein

Represent LLR (R_i) non-uniform quantizing Value set,

RepresentNon-uniform quantizing Value set,

Represent r_i ⁽ⁿ⁾The non-uniform quantizing of (j) Value set,

LLR(R_i) non-uniform quantizing value be:

Non-uniform quantizing value be:

r_i ⁽ⁿ⁾J the non-uniform quantizing value of () is:

Emulate for this non-uniform quantizing value, if decoding performance meets the requirements, as final non-uniform quantizing value, If decoding performance is undesirable, return step 2) redefine the non-uniform quantizing value of ternary.

The present invention proposes the enforcement of a kind of non-uniform quantizing method of low density parity check code and long-pending decoding operation numerical value Example is described as follows:

The LDPC code word parameter of the present embodiment is: code length 15330, information bit length 12775, and code check is 5/6, the present embodiment Select the non-uniform quantizing parameter that three bits for above-mentioned code word and long-pending decoding operation numerical value represent, be embodied as step As follows:

1) LDPC and the uniform quantization parameter of long-pending decoding operation numerical value are selected:

Determined the lowest evenness quantizing bit number meeting decoding performance requirement by emulation, obtain the uniform quantization of Soft Inform ation Parameter (Q_L,Q_fL) it is (6,1), the uniform quantization parameter (Q of bit node output external information_q,Q_fq) it is (6,2), check-node is defeated Go out the uniform quantization parameter (Q of external information_r,Q_fr) it is (6,5), wherein Q_L=6, Q_q=6, Q_r=6 amounts representing three kinds of information respectively Change bit number, Q_fL=1, Q_fq=2, Q_frThe quantizing bit number of=5 fractional parts representing three kinds of information respectively.It is implemented as Sequence of real numbers R=[the R that receiving terminal is received₀,R₁,R₂,…,R₁₅₃₂₉], its i-th, i=0,1 ..., i ..., 15329 codes Soft Inform ation LLR (the R of unit_i), bit node i exported during decoding the external information of check-node jTo verification joint Point j exports external information r of bit node i_i ⁽ⁿ⁾J () is quantified as:

Wherein [] represents the maximum integer less than its value, for the quantization x=LLR (R of Soft Inform ation_i), Q_f=1, for The quantization x=q of variable node output external information_j(i), Q_f=2, for the quantization x=r of check-node output external information_i(j), Q_f =5, n represent iterations.

2) based on step 1) the uniform quantization parameter (6,1), (6,2) and (6,5) that obtains, gradually reduce LLR the most respectively (R_i)、 And r_i ⁽ⁿ⁾J the quantized value number of (), to 7, uses Q_non=3 represent non-uniform quantizing value bit number；First from Big quantized value starts and passes through performance simulation and determines the 3rd of each variable the, 2,1 non-uniform quantizing value successively；It is embodied as step Rapid as follows:

2.1) determine and receive each code element Soft Inform ation LLR (R_i) the 3rd non-uniform quantizing value；Code element Soft Inform ation is uniform The arithmetic number of 31 forward quantizations values is divided into by quantization parameter (6,1) 32 forward quantized intervals, bearing of 31 negative quantified values Realpartition becomes 32 negative sense quantized intervals, and between positive smallest quantization interval and negative maximum quantization interval, the quantized value in region is Zero；First the 32nd, 31 quantized intervals are merged, using the intermediate value 15 of the 31st quantized interval as the 3rd non-uniform quantizing Value；With this non-uniform quantizing be worth 15, with do not merge quantized interval former uniform quantization value [0,0.5,1.0,1.5 ..., 14.5], And variableAnd r_i ⁽ⁿ⁾J (6,2) and (6,5) the uniform quantization parameter of (), carries out performance simulation, decoding performance and step 1) when in, ternary all uses uniform quantization parameter, the bit error rate is at the same order of magnitude, cancels current quantized interval and merges；Secondly will 32nd, 31,30 quantized intervals are merged, using the intermediate value 14.5 of the 30th quantized interval as the 3rd non-uniform quantizing value； With this non-uniform quantizing be worth 14.5, with do not merge quantized interval former uniform quantization value [0,0.5,1.0,1.5 ..., 14], with And variableAnd r_i ⁽ⁿ⁾J (6,2) and (6,5) the uniform quantization parameter of (), re-starts performance simulation, decoding performance and three When variable all uses uniform quantization parameter, the bit error rate is at the same order of magnitude, cancels current quantized interval and merges, by that analogy, gradually Increase the number merging quantized interval, until the 21st time, after maximum 22 quantized intervals merge, simulation performance is all adopted than ternary Increase more than an order of magnitude by bit error rate during uniform quantization parameter, then cancel current quantized interval and merge, the 20th time is merged Time the 12nd quantized interval intermediate value 5.5 as the 3rd quantized value of Soft Inform ation；

2.2) at Soft Inform ation LLR (R_i) use step 2.1) determine non-uniform quantizing value, external information r_i ⁽ⁿ⁾(j) employing (6, 5) on uniform quantization parameter basis, use with step 2.1) same procedure determine external informationThe 3rd quantized value, tool Body is: merged by the 32nd, 31 quantized intervals first, using the intermediate value 7.5 of the 31st quantized interval as the 3rd non-homogeneous amount Change value, by this non-uniform quantizing value, with do not merge quantized interval former uniform quantization value [0,0.25,0.5,0.75 ..., 7.25], Step 2.1) LLR (R that determines_i) non-uniform quantizing value, and variable r_i ⁽ⁿ⁾J (6,5) the uniform quantization parameter of (), carries out performance Emulation, decoding performance and step 1) in ternary when all using uniform quantization parameter the bit error rate at the same order of magnitude, cancel current Quantized interval merges；Secondly the 32nd, 31,30 quantized intervals are merged, using the intermediate value 7.25 of the 30th quantized interval as 3rd non-uniform quantizing value, by this non-uniform quantizing value, with do not merge quantized interval former uniform quantization value [0,0.25,0.5, 0.75 ..., 7], step 2.1) and the LLR (R that determines_i) non-uniform quantizing value, and variable r_i ⁽ⁿ⁾(6,5) uniform quantization ginseng of (j) Number, re-starts performance simulation, and when decoding performance and ternary all use uniform quantization parameter, the bit error rate is at the same order of magnitude, takes The current quantized interval that disappears merges, by that analogy, until the 12nd time, after maximum 13 quantized intervals merge, simulation performance becomes than three When amount all uses uniform quantization parameter, the bit error rate increases more than an order of magnitude, then cancel current quantized interval and merge, by the 11st Intermediate value 5.0 conduct of the 21st quantized interval during secondary mergingThe 3rd quantized value；

2.3) use with step 2.1) same procedure determine external information r_i ⁽ⁿ⁾3rd quantized value of (j), particularly as follows: first Secondary by the 32nd, 31 quantized intervals merging, using the intermediate value 0.9375 of the 31st quantized interval as the 3rd non-uniform quantizing Value, by this non-uniform quantizing value, with do not merge quantized interval former uniform quantization value [0,0.03125,0.0625,0.09375 ..., , and LLR (R 0.90625]_i) use step 2.1) the non-uniform quantizing value that determines,Use step 2.2) determine non- Uniform quantization value, carries out performance simulation, decoding performance and step 1) in ternary when all using uniform quantization parameter the bit error rate exist The same order of magnitude, cancels current quantized interval and merges, secondly merged by the 32nd, 31,30 quantized intervals, quantify district by the 30th Between intermediate value 0.90625 as the 3rd non-uniform quantizing value, by this non-uniform quantizing value, former all with not merging quantized interval Even quantized value [0,0.03125,0.0625,0.09375 ..., 0.875], and LLR (R_i) use step 2.1) determine non- Uniform quantization value,Use step 2.2) the non-uniform quantizing value that determines, re-start performance simulation, decoding performance and three When variable all uses uniform quantization parameter, the bit error rate is at the same order of magnitude, cancels current quantized interval and merges, by that analogy, until 3rd time, after maximum 4 quantized intervals merge, when simulation performance all uses uniform quantization parameter than ternary, the bit error rate increases one More than the individual order of magnitude, then cancel current quantized interval and merge, the intermediate value of the 30th quantized interval when the 2nd time is merged 0.90625 as r_i ⁽ⁿ⁾3rd quantized value of (j)；

2.4) three variables L LR (R are being determined_i)、r_i ⁽ⁿ⁾After (j) the 3rd non-uniform quantizing value, to these three changes After the size of the 3rd non-uniform quantizing value of amount carries out being increased or decreased of the 1-4 times of quantity of 0.5,0.25 and 0.03125 respectively It is combined into multiple combinations of the non-uniform quantizing value of three variablees, each combination is carried out performance simulation, finally, by all emulation Performance contrasts, and selects three variable non-uniform quantizing values during optimal performance as the LLR (R finally determined_i)、r_i ⁽ⁿ⁾(j) the 3rd non-uniform quantizing value；

Detailed description of the invention can be: first the first two variables L LR (R_i) and3rd non-uniform quantizing value keeps not Become, the 3rd variable r_i ⁽ⁿ⁾J () the 3rd non-uniform quantizing value increases by the 1-4 times of size of 0.03125, carry out performance simulation；Secondly The 3rd non-uniform quantizing value of the first two variable keeps constant, and the 3rd non-uniform quantizing value of the 3rd variable is reduced 0.03125 1-4 times of size, then carry out performance simulation；Again, the 3rd non-uniform quantizing value of first variable keeps constant, by second The 3rd non-uniform quantizing value of variable increases by the 1-4 times of size of 0.25, and the 3rd the non-uniform quantizing value increasing of the 3rd variable adds deduct The 1-4 times of size of few 0.03125, carries out performance simulation；Again, the 3rd non-uniform quantizing value of first variable keeps constant, will The 3rd non-uniform quantizing value of second variable reduces by the 1-4 times of size of 0.25, and the 3rd non-uniform quantizing value of the 3rd variable increases Add deduct few 0.03125 1-4 times of size, carry out performance simulation；Again, the 3rd non-uniform quantizing value of first variable increases The 1-4 times of size of 0.5, is increased or decreased the 1-4 times of size of 0.25 by the 3rd non-uniform quantizing value of second variable, the 3rd The 3rd non-uniform quantizing value of variable is increased or decreased the 1-4 times of size of 0.03125, carries out performance simulation；Again, first change 3 non-uniform quantizing values of flow control reduce by the 1-4 times of size of 0.5, the 3rd non-uniform quantizing value of second variable are increased or decreased The 1-4 times of size of 0.25, the 3rd non-uniform quantizing value of the 3rd variable is increased or decreased the 1-4 times of size of 0.03125, carries out Performance simulation；Finally, all simulation performances are contrasted, select ternary non-uniform quantizing value during optimal performance as more LLR (R after Xin_i)、r_i ⁽ⁿ⁾J () the 3rd non-uniform quantizing value, result is 7.0,6.0 and 0.96875.

2.5) step 2.1 is repeated)-2.4), determine LLR (R_i)、And r_i ⁽ⁿ⁾J 2nd the non-uniform quantizing value of () is 4.0,3.25 and 0.25；Repeat step 2.1)-2.4), determine LLR (R_i)、And r_i ⁽ⁿ⁾1st non-uniform quantizing of (j) Value, when merging maximum several quantized intervals and determining the 1st quantization parameter, the quantized value not merging quantized interval takes 0, determines The 1st non-uniform quantizing value of ternary is 2.5,1.75 and 0.03125.

3) at LLR (R_i)、r_i ⁽ⁿ⁾J 3 non-uniform quantizing values of () have just obtained non-uniform quantizing after all determining Parameter (Q_nonun-L,Q_nonun-q,Q_nonun-r), wherein Q_nonun-L=[-7.0 ,-4.0 ,-2.5,0,2.5,4.0,7.0] represent LLR (R_i) non-uniform quantizing value set, Q_nonun-q=[-6.0 ,-3.25 ,-1.75,0,1.75,3.25,6.0] represent's Non-uniform quantizing value set, Q_nonun-r=[-0.96875 ,-0.25 ,-0.03125,0,0.03125,0.25,0.96875] represents r_i ⁽ⁿ⁾The non-uniform quantizing value set of (j), LLR (R_i) non-uniform quantizing value be:

Non-uniform quantizing value be:

r_i ⁽ⁿ⁾J the non-uniform quantizing value of () is:

Emulate, at E for this non-uniform quantizing value_b/N₀9.860 × 10 are reached for bit error rate during 3.7dB^-8, by mistake Code check reaches 1 × 10^-7Time do not occur that error code is flat, decoding performance meets the requirements, as final 3 bit non-uniform quantizing parameters.

Table 1LDPC code difference quantization scheme performance under awgn channel

Table 2LDPC code difference quantization scheme FPGA realizes resource comparison (FPGA model: XC7VX485T)

Claims (3)

1. a low density parity check code and the non-uniform quantizing method of long-pending decoding operation numerical value, it is characterised in that to code element Soft Inform ation, export from check-node bit node external information, export from bit node check-node external information this three Plant variable and be respectively adopted different non-uniform quantizing parameters；First to the Soft Inform ation of LDPC code i-th code element, to bit node i The external information exporting check-node j, the external information that check-node j exports bit node i are selected bit more than a group and are represented Uniform quantization parameter；Again based on this uniform quantization parameter, gradually reduce the quantized value number of described three kinds of variablees successively, Finally give the quantized value of the three kinds of variable desirable number meeting decoding performance requirement, these quantized values form three kinds different non- The set of uniform quantization parameter, uses described different non-uniform quantizing parameter to quantify three kinds of variate-values respectively.

2. low density parity check code as claimed in claim 1 and the non-uniform quantizing method of long-pending decoding operation numerical value, its feature Being, the method specifically includes following steps:

1) determined by emulation and meet the lowest evenness quantizing bit number of decoding performance requirement, obtain the uniform quantization ginseng of Soft Inform ation Number (Q_L,Q_fL), the uniform quantization parameter (Q of bit node output external information_q,Q_fq), the even amount of check-node output external information Change parameter (Q_r,Q_fr), wherein Q_L,Q_q,Q_rRepresent the quantizing bit number of three kinds of information, Q respectively_fL,Q_fq,Q_frRepresent three kinds of letters respectively The quantizing bit number of the fractional part of breath；

2) based on step 1) the uniform quantization parameter (Q that obtains_L,Q_fL),(Q_q,Q_fq) and (Q_r,Q_fr), gradually reduce the most respectively LLR(R_i)、WithQuantized value number to arrivingIndividual, use Q_nonRepresent non-uniform quantizing value bit number, Q_non Value is no more than Q_L,Q_q,Q_rThe positive integer of minima；First start and pass through performance simulation from maximum quantization value to determine successively often The of individual variable..., 1 non-uniform quantizing value；

3) LLR (R after determining_i)、'sIndividual non-uniform quantizing is worth to three kinds of non-homogeneous amounts Change value set, the non-uniform quantizing value of three kinds of set is emulated, if decoding performance meets the requirements, as the most non-homogeneous Quantized value, if decoding performance is undesirable, returns step 2) redefine the non-uniform quantizing value of ternary.

3. low density parity check code as claimed in claim 2 and the non-uniform quantizing method of long-pending decoding operation numerical value, its feature It is, described step 2) to be embodied as step as follows:

2.1) determine and receive each code element Soft Inform ation LLR (R_i)Individual non-uniform quantizing value；Code element Soft Inform ation is equal Even quantization parameter (Q_L,Q_fLWill in)The arithmetic number of individual forward quantizations value is divided intoIndividual forward quantized interval,The negative realpartition of individual negative quantified value becomesIndividual negative sense quantized interval, positive smallest quantization interval and negative maximum quantization district Between between, the quantized value in region is zero；Merge interval for maximum quantization with time big quantized interval first, the quantized interval that will merge The intermediate value of middle smallest quantization interval is asIndividual non-uniform quantizing value；By this non-uniform quantizing value, with the amount of merging Change interval former uniform quantization value, and variableWith(Q_q,Q_fq) and (Q_r,Q_fr) uniform quantization parameter carries out Performance simulation, if decoding performance and step 1) in ternary the bit error rate is at the same order of magnitude when all using uniform quantization parameter, then Cancel current quantized interval to merge；Secondly maximum three quantized intervals are merged the non-uniform quantizing value obtained, re-starting property Can emulate, if the bit error rate is at the same order of magnitude when decoding performance and ternary all use uniform quantization parameter, then cancel this secondary amounts Change interval to merge, by that analogy, be gradually increased the number merging quantized interval, until the N-1 time, maximum N number of quantized interval closes After and, when simulation performance all uses uniform quantization parameter than ternary, the bit error rate increases more than an order of magnitude, then cancel current Quantized interval merges, using the intermediate value of smallest quantization interval during the N-2 time merging as the of Soft Inform ationIndividual quantization Value, N is positive integer；

2.2) at Soft Inform ation LLR (R_i) use step 2.1) determine non-uniform quantizing value, external informationUse (Q_r,Q_fr) On uniform quantization parameter basis, use with step 2.1) same procedure determine external information?Individual non-all Even quantized value, by this non-uniform quantizing value, and does not merges quantized interval former uniform quantization value, step 2.1) LLR (R that determines_i) Non-uniform quantizing value, and variable(Q_r,Q_fr) uniform quantization parameter, carry out performance simulation, by that analogy, to every time The non-uniform quantizing parameter obtained after merging carries out performance simulation, obtains the intermediate value conduct of smallest quantization interval?Individual quantized value；

2.3) use with step 2.1) same procedure determine external information?Individual quantized value, non-homogeneous with this Quantized value, and does not merges quantized interval former uniform quantization value, and LLR (R_i) use step 2.1) non-uniform quantizing that determines Value,Using step 2.2) the non-uniform quantizing value that determines carries out performance simulation, smallest quantization interval when obtaining merging Intermediate value conduct?Individual quantized value；

2.4) to fixed three variables L LR (R_i)、TheIndividual non-uniform quantizing value size is divided Do not carry out 1/Q_fL、1/Q_fqAnd 1/Q_frMultiple groups of non-uniform quantizing value of three variablees it are combined into after being increased or decreased of multiple Close, each combination is carried out performance simulation, finally, all simulation performances is contrasted, select three changes during optimal performance Amount non-uniform quantizing value is as newly determined LLR (R_i)、TheIndividual non-uniform quantizing value；

2.5) step 2.1 is repeated)-2.4), determine LLR (R_i)、And?...、 1 non-uniform quantizing value.