CN102340320B - Bidirectional and parallel decoding method of convolutional Turbo code - Google Patents

Abstract

The invention provides a decoding method of a convolutional Turbo code for reducing decoding time delay and saving a memory. The decoding method comprises the following steps of: simultaneously carrying out forward recursion and backward recursion in a component decoding process; dividing the forward recursion and the backward recursion into two stages with equivalent computation quantity; and sequentially calculating and obtaining posterior likelihood ratio information at the beginning of the second stage. The time delay from the beginning of recursion operation to the end of the posterior likelihood ratio information operation is shortened once compared with the traditional decoding process. Furthermore, the traditional posterior likelihood ratio operation is serial, while the posteriorlikelihood ratio operation of the invention is carried out bidirectionally and simultaneously in parallel, the required calculation time and the recursive calculation time are overlapped, and it is unnecessary to distribute additional calculation time; in addition, a bidirectional parallel structure can ensure that the memory used for storing state metric is reduced by half. Furthermore, through the calculation of splitting branch metric, redundancy calculation is reduced, and the space for storing the branch metric is reduced by half.

Description

CTC two-way simultaneous interpretation method

Technical field

The invention belongs to the communications field, relate generally to chnnel coding, especially the correlation technique of Turbo code decoding.

Background technology

Since the proposition of iterative decoding concept begins, Turbo code just is widely studied and applied.CTC (CTC) characteristics higher with its code efficiency, that coding rate is faster and free distance is bigger are able to fast-developing in recent years, are elected to be the forward error correction type of physical layer at present by standard 802.16e and 802.16m.

Standard 802.16m selects for use double-binary convolutional Turbo code (DB-CTC) as one of channel coding schemes.Systematic code DB-CTC is each input 2 bit information that walk abreast constantly when compiling, and exports 6 bits.Since adopted the scheme of secondary coding, the forward and backward state consistency of encoding, thus need not the bit that finishes up.Yet these characteristics of DB-CTC also make it decipher more complicated.

The decoder of DB-CTC adopts the structure of two same components decoder parallel iterations, data to be decoded (signal code resulting soft value of information (the soft output likelihood ratio of channel information) after soft demodulation of CTC after the signal transmission) are imported these two component decoders (component decoder 1, component decoder 2) respectively, component decoder 1 output posteriority likelihood ratio information, and with this posteriority likelihood ratio information as external information again through being returned to component decoder 2 after interweaving as the priori likelihood ratio information of deciphering; Component decoder 2 is exported posteriority likelihood ratio information after deciphering, and becomes the priori likelihood ratio information of component decoder 1 as external information after deinterleaving, finishes iteration thus one time.When reaching default maximum iteration time, the posteriority likelihood ratio information of component decoder 2 outputs obtains final decode results through deinterleaving and hard decision.The iterative decoding of CTC makes the decode results convergence gradually by the mode of transmitting external information between two component decoders mutually just, and improves decoding performance thus.

Because contain a large amount of multiplyings and intermediate variable in the implementation procedure of MAP algorithm, so the decoding complexity height, and decoding delay is long.So modified model Log-MAP algorithm or its reduced form Max-Log-MAP decoding algorithm of MAP algorithm commonly used in simulation study and engineering.Though reduced multiplying, adopted the component decoder of Log-MAP algorithm or Max-Log-MAP algorithm that long decoding delay is still arranged.

The soft value of information of channel that the component decoder is used for receiving is calculated output posteriority likelihood ratio information by the MAP algorithm.Adopted the decode procedure of the component decoder of Log-MAP algorithm or Max-Log-MAP algorithm to comprise: the calculating of the calculating of branch metric, forward recursive, backward recursive and posteriority likelihood ratio information.I posteriority likelihood ratio information L constantly _i(i=0 ..., N-1) being obtained to state measurement and i branch metric calculation constantly by i forward state metric, the back of the i+1 moment constantly, N is two binary bits group length of importing in the CTC cataloged procedure.Usually, the component decoder is used behind soft value of information calculating 0 to the N-1 branch metric constantly of the soft demodulation of channel, need to calculate successively from the forward state metric of 0 to n-hour by forward recursive earlier, and store, calculate constantly back to state measurement from N to 0 successively by backward recursive more afterwards, and storage.At last, carry out the calculating of posteriority likelihood ratio information.0 to N-1 constantly the back of forward state metric, 1 to n-hour is used for 0 to N-1 constantly the direct calculating of posteriority likelihood ratio information to state measurement, constantly back to the state measurement of state measurement as the initialization next iteration of the forward state metric of n-hour, the 0th.

Summary of the invention

Technical problem to be solved by this invention is that the decoding architecture of employing two-way simultaneous provides a kind of interpretation method that reduces decoding delay, saves the CTC of memory.

The present invention solves the problems of the technologies described above the technological means that adopts to be, CTC two-way simultaneous interpretation method comprises:

The iterative decoder of two component decoders parallel connection is adopted in data input to be decoded;

When not reaching default maximum iteration time, output posteriority likelihood ratio information is converted into external information after the component decoder for decoding, again through interweave or deinterleaving after, input to described another component decoder as priori likelihood ratio information;

When reaching default maximum iteration time, the posteriority likelihood ratio information that the component decoder of last work is exported after deciphering through deinterleaving and hard decision, obtains decode results;

It is characterized in that the decode procedure of component decoder specifically may further comprise the steps:

The initialization step of branch metric calculation and forward-backward recutrnce: the data to be decoded of utilizing input and priori likelihood ratio information calculations and storage from 0 to (N/2)-1 constantly preceding N/2 branch metric, the branch metric in N/2 to N-1 back N/2 the moment constantly constantly, the forward state metric in the 0th moment of initialization and n-hour back to state measurement; Described N is two binary bits group length;

The phase I step: the forward state metric with initialized the 0th moment is starting point, participates in forward recursive with preceding N/2 branch metric constantly and calculates, and obtains the forward state metric in 0 to the N/2 moment successively, and stores; Simultaneously, be starting point with the back of n-hour to state measurement, use N/2 the branch metric constantly in back to participate in backward recursive and calculate, obtain the back to state measurement of N to the N/2 moment successively, and storage;

Second stage step: be starting point to state measurement with N/2 constantly back, participating in backward recursive with preceding N/2 branch metric tolerance constantly calculates, obtain constantly back to state measurement from N/2 to 0 successively, and participate in the posteriority likelihood ratio and calculate with (N/2)-1 of phase I storage to 0 constantly forward state metric and preceding N/2 branch metric constantly successively, obtain from (N/2)-1 to 0 constantly the posteriority likelihood ratio information in preceding N/2 the moment; Simultaneously, forward metrics with the N/2 moment is starting point, participate in forward recursive calculating with N/2 the branch metric constantly in back and obtain the forward state metric of N/2 to n-hour successively, and participate in the posteriority likelihood ratio with (N/2)+1 of phase I storage to n-hour back to N/2 branch metric constantly of state measurement and back successively and calculate, obtain the posteriority likelihood ratio information in back N/2 the moment constantly from N/2 to N-1.

After comparing existing component decoder and needing forward recursive to finish, just backward recursive begins, and posteriority likelihood ratio information could begin to calculate then; The present invention carries out forward recursive and backward recursive simultaneously, and forward direction/backward recursive is divided into two suitable stages of operand, and posteriority likelihood ratio information just can calculate successively in the beginning of second stage.Time-delay when namely beginning to finish to the computing of posteriority likelihood ratio information from the beginning recursive operation, the present invention compares existing decode procedure and has shortened one times.And, the computing of existing posteriority likelihood ratio is serial, posteriority likelihood ratio computing of the present invention adopts two-way simultaneous to carry out simultaneously, overlap with the time of recursive calculation required computing time, then do not need the other Distribution Calculation time, in addition, the structure of two-way simultaneous can reduce by half so that be used for the memory of store status tolerance.

Concrete, being calculated as of branch metric:

γ _k(s’,s)=L ^a(u _k)+1/2×v _kar _ka+1/2×v _kbr _kb+1/2×v _kyr _ky+1/2×v _kwr _kw；

Wherein, k represents current time, and s ' is the possible state of current time, and s is next possible state constantly, γ _k(s ', s) transfer to the k+1 branch metric of state s constantly, u for k moment state s ' _k=(u _Ka, u _Kb) be k two binary bits information of input coding device constantly, L ^a(u _k) be k priori likelihood ratio information constantly, v _k=(v _Ka, v _Kb, v _Ky, v _Kw) for sending bit, r _k=(r _Ka, r _Kb, r _Ky, r _Kw) the k data to be decoded (the soft value of information after the soft demodulation) constantly that receive of expression channel.

Further, in order to reduce the amount of calculation of branch metric, when Branch Computed is measured, branch metric is split as two factor p _kAnd q _k:

p _k=L ^a(u _k)+1/2×v _kar _ka+1/2×v _kbr _kb；

q _k=1/2×v _kyr _ky+1/2×v _kwr _kw。

According to v _k=(v _Ka, v _Kb, v _Ky, v _Kw) value, p then _kTotal p _{K, 00}, p _{K, 01}, p _{K, 10}, p _{K, 11}Four kinds of values, q _kTotal q _{K, 00}, q _{K, 01}, q _{K, 10}, q _{K, 11}Four kinds of values are so 32 branch metrics between per two moment (16 different values) can all construct with 8 factors.Classical relatively Log-MAP algorithm and Max-Log-MAP algorithm each constantly Branch Computed tolerance need carry out 80 multiplication and 64 sub-additions, the branch metric after splitting only needs to do 24 multiplication and 28 sub-additions (comprising the addition that splits tolerance with factor structure); Each moment Branch Computed degree of classical relatively Log-MAP algorithm and Max-Log-MAP algorithm need be stored 16 different values, and the branch metric after splitting only need be stored 8 different factors.

The invention has the beneficial effects as follows, in the decode procedure of branch's decoder, adopt the mode computing mode tolerance and posteriority likelihood ratio information of two-way simultaneous, to carry out the operation of forward direction, backward recursive carries out simultaneously, time reduces by half, and be dissolved into the computing time of posteriority likelihood ratio in the time of recursive operation, reduced the memory space of decoding delay and state measurement thus greatly; Further, by splitting the calculating of branch metric, the space of reducing redundant computation and stores branch tolerance reduces by half.

Description of drawings

Fig. 1 is coded in any two grid charts between constantly for DB-CTC;

Fig. 2 is inner decoding schematic diagram in the component decoder.

Embodiment

The present invention is directed to that the decode procedure in the component decoder improves in the convolution Turbo decoding, other processing procedures are constant, and the convolution Turbo decode procedure comprises:

The iterative decoder of data to be decoded, the parallel connection of two component decoders of priori likelihood ratio information input;

When not reaching default maximum iteration time, output posteriority likelihood ratio information is converted into external information after the component decoder for decoding, again through interweave or deinterleaving after, input to described another component decoder as priori likelihood ratio information;

When reaching default maximum iteration time, the component decoder for decoding output posteriority likelihood ratio information of working at last through deinterleaving and hard decision, obtains decode results.

Present embodiment improves component decoder Log-MAP interpretation method under the condition that does not change original error performance: comprise fractionation and the two-way simultaneous computing of branch metric.

Branch metric is the critical quantity of contact component decoder input and output, all can use branch metric in recursive calculation forward state metric and back during to state measurement and when calculating posteriority likelihood ratio information, and the final purpose of component decoder is output posteriority likelihood ratio information (calculating of posteriority likelihood ratio information needs forward state metric, back to state measurement and branch metric).

As shown in Figure 1, any two have 32 branch metrics constantly, totally 16 different values.The computing formula of branch metric is:

γ _k(s’,s)=L ^a(u _k)+1/2×v _kar _ka+1/2×v _kbr _kb+1/2×v _kyr _ky+1/2×v _kwr _kw，

Wherein, s ' is the possible state of current time, and s is next possible state constantly, γ _k(s ', s) transfer to the k+1 branch metric of state s constantly, u for k moment state s ' _k=(u _Ka, u _Kb) be k two binary bits information of input coding device constantly, L ^a(u _k) be k priori likelihood ratio information constantly, v _k=(v _Ka, v _Kb, v _Ky, v _Kw) for sending bit, r _k=(r _Ka, r _Kb, r _Ky, r _Kw) the k soft value constantly that receives of expression channel.

Branch metric is split as two factor p _kAnd q _k, wherein:

p _k=L ^a(u _k)+1/2×v _kar _ka+1/2×v _kbr _kb

q _k=1/2×v _kyr _ky+1/2×v _kwr _kw

According to v _k=(v _Ka, v _Kb, v _Ky, v _Kw) value, p then _kTotal p _{K, 00}, p _{K, 01}, p _{K, 10}, p _{K, 11}Four kinds of values, q _kTotal q _{K, 00}, q _{K, 01}, q _{K, 10}, q _{K, 11}Four kinds of values are so 16 branch metrics between per two moment can all construct with 8 factors.Calculate redundancy so the factor that Branch Computed tolerance splits into can reduce, save memory.

After introducing the fractionation of branch metric, ask k+1 forward state metric α constantly _K+1(k=0 ..., forward recursive computing N) is: k forward state metric matrix A constantly _kWith k forward recursive Matrix C constantly _kAddition behind the right index is taken from summation, each capable element of matrix i of obtaining of summation, takes from right logarithm after the summation again, is exactly k+1 state constantly

Corresponding forward state metric, i=1,2 ... 8.

Wherein, k forward state metric matrix A constantly _kWith k forward recursive Matrix C constantly _kAddition is expressed as:

$A_k+C_k=\left(\begin{array}{cccc}{\mathrm{\α}}_k\left(s_0\right)& {\mathrm{\α}}_k\left(s_1\right)& {\mathrm{\α}}_k\left(s_6\right)& {\mathrm{\α}}_k\left(s_7\right)\\ {\mathrm{\α}}_k\left(s_2\right)& {\mathrm{\α}}_k\left(s_4\right)& {\mathrm{\α}}_k\left(s_3\right)& {\mathrm{\α}}_k\left(s_5\right)\\ {\mathrm{\α}}_k\left(s_5\right)& {\mathrm{\α}}_k\left(s_3\right)& {\mathrm{\α}}_k\left(s_4\right)& {\mathrm{\α}}_k\left(s_2\right)\\ {\mathrm{\α}}_k\left(s_7\right)& {\mathrm{\α}}_k\left(s_1\right)& {\mathrm{\α}}_k\left(s_6\right)& {\mathrm{\α}}_k\left(s_0\right)\\ {\mathrm{\α}}_k\left(s_1\right)& {\mathrm{\α}}_k\left(s_7\right)& {\mathrm{\α}}_k\left(s_0\right)& {\mathrm{\α}}_k\left(s_6\right)\\ {\mathrm{\α}}_k\left(s_3\right)& {\mathrm{\α}}_k\left(s_5\right)& {\mathrm{\α}}_k\left(s_2\right)& {\mathrm{\α}}_k\left(s_4\right)\\ {\mathrm{\α}}_k\left(s_4\right)& {\mathrm{\α}}_k\left(s_2\right)& {\mathrm{\α}}_k\left(s_5\right)& {\mathrm{\α}}_k\left(s_3\right)\\ {\mathrm{\α}}_k\left(s_6\right)& {\mathrm{\α}}_k\left(s_0\right)& {\mathrm{\α}}_k\left(s_7\right)& {\mathrm{\α}}_k\left(s_1\right)\end{array}\right)+\left(\begin{array}{cccc}{p}_{k,00+}{q}_{k,00}& p_{k,01+}{q}_{k,10}& p_{k,10+}{q}_{k,11}& p_{k,11+}{q}_{k,01}\\ p_{k,00+}{q}_{k,10}& p_{k,01+}{q}_{k,00}& p_{k,10+}{q}_{k,01}& p_{k,11+}{q}_{k11}\\ p_{k,00+}{q}_{k,11}& p_{k,01+}{q}_{k,01}& p_{k,10+}{q}_{k,00}& p_{k,11+}{q}_{k,10}\\ p_{k,00+}{q}_{k,01}& p_{k,01+}{q}_{k,11}& p_{k,10+}{q}_{k,10}& p_{k,11+}{q}_{k,00}\\ p_{k,00+}{q}_{k,00}& p_{k,01+}{q}_{k,10}& p_{k,10+}{q}_{k,11}& p_{k,11+}{q}_{k,01}\\ p_{k,00+}{q}_{k,10}& p_{k,01+}{q}_{k,00}& p_{k,10+}{q}_{k,01}& p_{k,11+}{q}_{k,11}\\ p_{k,00+}{q}_{k,11}& p_{k,01+}{q}_{k,01}& p_{k,10+}{q}_{k,00}& p_{k,11+}{q}_{k,10}\\ p_{k,00+}{q}_{k,01}& p_{k,01+}{q}_{k,11}& p_{k,10+}{q}_{k,10}& p_{k,11+}{q}_{k,00}\end{array}\right)$

a _k(s _i) (i=1,2 ... 8) expression k moment state s _iForward state metric.From forward direction recursion matrix C _kIn find out that branch metric is by p _kWith q _kTwo kinds of factors constitute.Matrix C _kRow 1 and 5,2 and 6,3 and 7,4 and 8 equate respectively, namely only need calculate C _kPreceding four lines just can express C _kArbitrary factor p _{K, ij}Or q _{K, ij}In Matrix C _kIn the number of times that occurs be 8, only need calculate once and 4 times of non-classical Log-MAP algorithm these factors, so just eliminated the redundant computation in original algorithm.

Similarly, ask the back to state measurement β of the k moment _kThe backward recursive computing be: the back to the state measurement matrix B _K+1With the backward recursive matrix D _kSummation, each element of every row of the matrix that summation obtains are sued for peace respectively after taking from right index, take from right logarithm after the summation and are the corresponding backward recursive result of this row.

Wherein, the k+1 moment is back to the state measurement matrix B _K+1With k backward recursive matrix D constantly _kAddition is expressed as:

$B_{k+1}+D_k=\left(\begin{array}{cccc}{\mathrm{\β}}_{k+1}\left(s_0\right)& {\mathrm{\β}}_{k+1}\left(s_3\right)& {\mathrm{\β}}_{k+1}\left(s_4\right)& {\mathrm{\β}}_{k+1}\left(s_7\right)\\ {\mathrm{\β}}_{k+1}\left(s_4\right)& {\mathrm{\β}}_{k+1}\left(s_7\right)& {\mathrm{\β}}_{k+1}\left(s_0\right)& {\mathrm{\β}}_{k+1}\left(s_3\right)\\ {\mathrm{\β}}_{k+1}\left(s_1\right)& {\mathrm{\β}}_{k+1}\left(s_2\right)& {\mathrm{\β}}_{k+1}\left(s_5\right)& {\mathrm{\β}}_{k+1}\left(s_6\right)\\ {\mathrm{\β}}_{k+1}\left(s_5\right)& {\mathrm{\β}}_{k+1}\left(s_6\right)& {\mathrm{\β}}_{k+1}\left(s_1\right)& {\mathrm{\β}}_{k+1}\left(s_2\right)\\ {\mathrm{\β}}_{k+1}\left(s_1\right)& {\mathrm{\β}}_{k+1}\left(s_2\right)& {\mathrm{\β}}_{k+1}\left(s_5\right)& {\mathrm{\β}}_{k+1}\left(s_6\right)\\ {\mathrm{\β}}_{k+1}\left(s_5\right)& {\mathrm{\β}}_{k+1}\left(s_6\right)& {\mathrm{\β}}_{k+1}\left(s_1\right)& {\mathrm{\β}}_{k+1}\left(s_2\right)\\ {\mathrm{\β}}_{k+1}\left(s_0\right)& {\mathrm{\β}}_{k+1}\left(s_3\right)& {\mathrm{\β}}_{k+1}\left(s_4\right)& {\mathrm{\β}}_{k+1}\left(s_7\right)\\ {\mathrm{\β}}_{k+1}\left(s_4\right)& {\mathrm{\β}}_{k+1}\left(s_7\right)& {\mathrm{\β}}_{k+1}\left(s_0\right)& {\mathrm{\β}}_{k+1}\left(s_3\right)\end{array}\right)+\left(\begin{array}{cccc}{p}_{k,00+}{q}_{k,00}& p_{k,11+}{q}_{k,00}& p_{k,10+}{q}_{k,11}& p_{k,01+}{q}_{k,11}\\ p_{k,00+}{q}_{k,00}& p_{k,11+}{q}_{k,00}& p_{k,10+}{q}_{k,11}& p_{k,01+}{q}_{k11}\\ p_{k,00+}{q}_{k,10}& p_{k,11+}{q}_{k,10}& p_{k,10+}{q}_{k,01}& p_{k,01+}{q}_{k,01}\\ p_{k,00+}{q}_{k,10}& p_{k,11+}{q}_{k,10}& p_{k,10+}{q}_{k,01}& p_{k,01+}{q}_{k,01}\\ p_{k,01+}{q}_{k,00}& p_{k,10+}{q}_{k,00}& p_{k,11+}{q}_{k,11}& p_{k,00+}{q}_{k,11}\\ p_{k,01+}{q}_{k,00}& p_{k,10+}{q}_{k,00}& p_{k,11+}{q}_{k,11}& p_{k,00+}{q}_{k,11}\\ p_{k,01+}{q}_{k,10}& p_{k,10+}{q}_{k,10}& p_{k,11+}{q}_{k,01}& p_{k,00+}{q}_{k,01}\\ p_{k,01+}{q}_{k,10}& p_{k,10+}{q}_{k,10}& p_{k,11+}{q}_{k,01}& p_{k,00+}{q}_{k,01}\end{array}\right)$

β _K+1(s _i) (i=1,2 ... 8) expression k+1 moment state s _iBack to state measurement.

Because the relative independentability of forward recursive and backward recursive computing and symmetry, can realize the two-way simultaneous recursive operation at one time, and the calculating of likelihood ratio information parallel finishing too.

In the component decoder two-way simultaneous computing as shown in Figure 2, wherein dotted line is divided into two stages with the two-way simultaneous structure, empty arrow is represented the direction of computing.The solid line grid that comprises state measurement is the unit memory, and the state measurement in the dotted line grid is the state measurement of interim storage (memory of also namely saving that is used for store status tolerance).α _k, β _kRepresent respectively k constantly whole forward state metric and k whole backs constantly to state measurement.The soft information that the component decoder receives participates in calculating the branch metric factor, and deposits memory in, as P _k={ p _{K, 00}, p _{K, 01}, p _{K, 10}, p _{K, 11}, Q _k={ q _{K, 00}, q _{K, 01}, q _{K, 10}, q _{K, 11}Be k 8 factors constantly.α ₀And β _NBe used for the initialization recursive operation, α _N/2And β _N/2Be phase I recurrence gained, be used as the starting point of second stage state measurement recurrence.α _NAnd β ₀Be used for the corresponding state tolerance of initialization next iteration.Posteriority likelihood ratio information is the output of component decoder, in Fig. 2, comprises posteriority likelihood ratio L _k(k=0,1,2 ..., module N-1) is responsible for calculating posteriority likelihood ratio information.

The decode procedure of component decoder specifically may further comprise the steps:

The calculation procedure of branch metric fractionation: the data to be decoded of utilization input and priori likelihood ratio information calculations and storage are from P, the Q factor, N/2 to N-1 back N/2 whole P, the Q factor constantly constantly in preceding N/2 the moment in 0 to (N/2)-1 moment;

Initialization step: initialization the 0th forward state metric constantly; Initialization n-hour back to state measurement; N is two binary bits group length;

Phase I step: carry out phase I forward recursive and phase I backward recursive simultaneously;

Phase I forward recursive: with the forward state metric α in the 0th moment ₀Be starting point, structure forward state metric matrix A _k, with preceding N/2 branch metric factor P, Q structure forward recursive Matrix C constantly _k(k=0,1,2 ..., (N/2)-1).Calculate (N/2+1) * 8 forward state metric that obtained for 0 to the N/2 moment successively by forward recursive, and storage;

Phase I backward recursive: back to state measurement β with n-hour _NBe starting point, the structure back is to the state measurement matrix B _K+1, with N/2 branch metric factor P, the Q structure backward recursive matrix D constantly in back _k(k=N-1, N-2 ..., N/2), calculate by backward recursive and to obtain successively behind N to N/2 (N/2+1) * 8 constantly to state measurement, and storage;

The second stage step:

Second stage backward recursive and likelihood ratio are calculated: back to state measurement β with N/2 the moment _N/2Be starting point, the structure back is to the state measurement matrix B _K+1, with preceding N/2 branch metric factor P, Q structure backward recursive matrix D constantly _k(k=N/2-1, N/2-2,0), calculate by backward recursive and to obtain successively behind N/2 to 0 (N/2+1) * 8 constantly to state measurement, and calculate to 0 constantly forward state metric and preceding N/2 branch metric factor P, Q constantly with (N/2)-1 of phase I storage successively, obtain from (N/2)-1 to 0 constantly the posteriority likelihood ratio information in preceding N/2 the moment; The 0th moment back that backward recursive calculating obtains at last is to state measurement β ₀Be used for the back to state measurement of initialization next iteration;

Second stage forward recursive and posteriority likelihood ratio information calculations: with N/2 forward state metric α constantly _N/2Be starting point, structure forward state metric matrix A _k, with N/2 branch metric factor P, the Q structure forward recursive Matrix C constantly in back _k(k=N/2, N/2+1, N-1), obtain (N/2+1) * 8 forward state metric of N/2 to n-hour successively by forward recursive, and calculate to N/2 branch metric factor constantly of state measurement and back with back from (N/2)+1 to n-hour of phase I storage successively, obtain the posteriority likelihood ratio information in back N/2 the moment constantly from N/2 to N-1; The n-hour forward state metric α that forward recursive calculates _NBe used for the forward state metric of initialization next iteration.

The two-way simultaneous computing of phase I is part forward state metric and the parallel recurrence of part back to state measurement, and the two-way simultaneous computing of second stage is the parallel recurrence of part forward and backward state measurement and the parallel computation of posteriority likelihood ratio information.Each and then calculating of state measurement constantly of calculating of posteriority likelihood ratio information constantly in second stage, thus be fused to the computing time of posteriority likelihood ratio in the time of forward-backward recutrnce computing, and brought reducing by half of state measurement memory space thus.Brought decoding quicker and more economize the effect of memory space so the fractionation of branch metric and two-way simultaneous structure be the decoding of DB-CTC.

If present embodiment is expanded to Max-Log-MAP, then the computing of branch metric fractionation is identical, when the two-way simultaneous computing, two-way simultaneous structure and calculation step are constant, and what need variation is to introduce the max log approximate processing in the computing of forward direction and backward recursion computing and posteriority likelihood ratio information.

Claims (3)

1. CTC two-way simultaneous interpretation method comprises:

The iterative decoder of two component decoders parallel connection is adopted in data input to be decoded;

When not reaching default maximum iteration time, output posteriority likelihood ratio information is converted into external information after the component decoder for decoding, again through interweave or deinterleaving after, input to described another component decoder as priori likelihood ratio information;

When reaching default maximum iteration time, the posteriority likelihood ratio information that the component decoder of last work is exported after deciphering through deinterleaving and hard decision, obtains decode results;

It is characterized in that the decode procedure of component decoder specifically may further comprise the steps:

The initialization step of branch metric calculation and forward-backward recutrnce: the data to be decoded of utilizing input and priori likelihood ratio information calculations and storage from 0 to (N/2)-1 constantly preceding N/2 branch metric, the branch metric in N/2 to N-1 back N/2 the moment constantly constantly, the forward state metric in the 0th moment of initialization and n-hour back to state measurement; Described N is two binary bits group length;

The phase I step: the forward state metric with initialized the 0th moment is starting point, participates in forward recursive with preceding N/2 branch metric constantly and calculates, and obtains the forward state metric in 0 to the N/2 moment successively, and stores; Simultaneously, be starting point with the back of n-hour to state measurement, use N/2 the branch metric constantly in back to participate in backward recursive and calculate, obtain the back to state measurement of N to the N/2 moment successively, and storage;

Second stage step: be starting point to state measurement with N/2 constantly back, participating in backward recursive with preceding N/2 branch metric tolerance constantly calculates, obtain constantly back to state measurement from N/2 to 0 successively, and participate in the posteriority likelihood ratio and calculate with (N/2)-1 of phase I storage to 0 constantly forward state metric and preceding N/2 branch metric constantly successively, obtain from (N/2)-1 to 0 constantly the posteriority likelihood ratio information in preceding N/2 the moment; Simultaneously, forward metrics with the N/2 moment is starting point, participate in forward recursive calculating with N/2 the branch metric constantly in back and obtain the forward state metric of N/2 to n-hour successively, and participate in the posteriority likelihood ratio with (N/2)+1 of phase I storage to n-hour back to N/2 branch metric constantly of state measurement and back successively and calculate, obtain the posteriority likelihood ratio information in back N/2 the moment constantly from N/2 to N-1.

2. CTC two-way simultaneous interpretation method according to claim 1 is characterized in that, being calculated as of branch metric:

γ _k(s’,s)=L ^a(u _k)+1/2×v _kar _ka+1/2×v _kbr _kb+1/2×v _kyr _ky+1/2×v _kwr _kw；

Wherein, k represents current time, and s ' is the possible state of current time, and s is next possible state constantly, γ _k(s ', s) transfer to the k+1 branch metric of state s constantly, u for k moment state s ' _k=(u _Ka, u _Kb) be k two binary bits information of input coding device constantly, L ^a(u _k) be k priori likelihood ratio information constantly, v _k=(v _Ka, v _Kb, v _Ky, v _Kw) for sending bit, r _k=(r _Ka, r _Kb, r _Ky, r _Kw) the k data to be decoded constantly that receive of expression channel.

3. as CTC two-way simultaneous interpretation method as described in the claim 2, it is characterized in that, in the branch metric calculation step, branch metric is split as two factor p _kAnd q _k:

p _k=L ^a(u _k)+1/2×v _kar _ka+1/2×v _kbr _kb；

q _k=1/2×v _kyr _ky+1/2×v _kwr _kw。

Images