CN1702975A - Binary decoding method for space-time Turbo code - Google Patents

Abstract

This invention relates to binary Turbo coding method, which comprises the following steps: receiving signals for inputting into the coding module, wherein the coding module is to receive signals and the validation information got from coding module to get the logarithm approximate value and outer information sent to the coding module; inputting the interlaced signals into the coding module to get the validation information for coding to get the approximate values sent to the coding module; repeating the above steps for several interlace to make judgment and output the values.

Description

The binary decoding method of Turbo code when empty

Technical field

The invention belongs to the field of information transmission in the multiple antenna communication, the interpretation method of Turbo code when relating generally to sky.

Background technology

The interpretation method of Turbo code during about sky, calendar year 2001 Su has proposed a kind of nonbinary interpretation method, and Kojima in 2002 have improved his method, have proposed the binary decoding method of more complicated.

The nonbinary interpretation method of Su.The decode procedure of Turbo code as shown in Figure 3 during Su empty, the signal r that receiver is accepted enters decoding module 1 (300), the prior information that decoding module 1 (300) then utilizes r and obtained by decoding module 2 (304), obtain after log-likelihood value and the external information through the nonbinary interpretation method, external information is interweaved (302) and it is passed to decoding module 2 (304); Then directly enter decoding module 2 (304) through interleaver (308) received signal afterwards, decoding module 2 (304) utilizes the received signal after interweave (308) and the prior information of coming from decoding module 1 (300) obtains external information and it is sent into decoding module 1 (300); Through after the iteration for several times, the log-likelihood value of decoding module 2 (304) is adjudicated the data bit of (310) and output estimation like this.Here, the nonbinary decoding algorithm of Su is mainly reflected in decoding module 1 (300) and the decoding module 2 (304).

Su has defined

$L^{(a,b)}(c_t^1,c_t^i)=\log \frac{P(c_t^1=a,c_t^i=b)}{P(c_t^1=-1,c_t^i=-1)}---\left(1\right)$

Here, i ∈ 2,3}, and a, b ∈+1 ,-1}.

He has considered maximum a posteriori probability (MAP) and two kinds of decoders of soft output Viterbi algorithm (SOVA).About the MAP decoder, log-likelihood ratio (LLR) expression is

$L^{(a,b)}(c_t^1,c_t^i)=\log \frac{\underset{(c_t^1,c_t^i)=(a,b)}{\underset{(s\text{'}\→s)}{\mathrm{\Σ}}}{\mathrm{\α}}_{t-1}(s\text{'}){\mathrm{\γ}}_t(r_t,s\text{'},s){\mathrm{\β}}_t\left(s\right)}{\underset{(c_t^1,c_t^i)=(-1,-1)}{\underset{(s_i\→s_i+1)}{\mathrm{\Σ}}}{\mathrm{\α}}_{t-1}(s\text{'}){\mathrm{\γ}}_t(r_t,s\text{'},s){\mathrm{\β}}_t\left(s\right)}---\left(2\right)$

Here s and s ' are illustrated in the state in the trrellis diagram, and

${\mathrm{\α}}_t\left(s\right)=\underset{s\text{'}}{\mathrm{\Σ}}{\mathrm{\α}}_{t-1}(s\text{'}){\mathrm{\γ}}_t(r_t,s\text{'},s)---\left(3a\right)$

${\mathrm{\β}}_t\left(s\right)=\underset{s\text{'}}{\mathrm{\Σ}}{\mathrm{\γ}}_{t+1}(r_{t+1},s\text{'},s){\mathrm{\β}}_{t+1}(s\text{'})---\left(3b\right)$

Usually, we use maximum likelihood soft-decision (ML) to decipher in decoder.If we consider first rsc encoder (recursive systematic convolutional code), be RSC1 (200), and think that the trrellis diagram state changes to s from s ', the value of the even number position of the input signal of then corresponding composition decoding module (DEC1) (300) is the input position of coding module RSC1 (200) $c_t^1=a$ With its check digit $c_t^2=b$ Stack.In this case, in the trrellis diagram, for specific a, b has only unique value.Therefore, we can use the information in the trrellis diagram in the maximum likelihood soft-decision.That is,

${\mathrm{\γ}}_t(r_t,s\text{'},s)=P(r_t,s\text{'},s)=P\left(r_t\right|c_t^1=a,c_t^2=b)\·P(c_t^1=a)---\left(4\right)$

With

$\log {\mathrm{\γ}}_t(r_t,s\text{'},s)$

$=\mathrm{\&Gamma;}(L_0^{(a,+1)}(c_t^1,c_t^3)-\frac{1}{{\mathrm{<figure-callout\; id="0"\; label="N"\; filenames="A20041004272800143.png"\; state="\{\{state\}\}">N</figure-callout>}}_0}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{|r_t^j-\sqrt{E_s}({\mathrm{\&alpha;}}_{1,j}\left(t\right)\&CenterDot;a+{\mathrm{\&alpha;}}_{2,j}\left(t\right)\&CenterDot;b)|}^2,$

$L_0^{(a,-1)}(c_t^1,c_t^3)-\frac{1}{{\mathrm{<figure-callout\; id="0"\; label="N"\; filenames="A20041004272800143.png"\; state="\{\{state\}\}">N</figure-callout>}}_0}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{|r_t^j-\sqrt{E_s}({\mathrm{\&alpha;}}_{1,j}\left(t\right)\&CenterDot;a+{\mathrm{\&alpha;}}_{2,j}\left(t\right)\&CenterDot;b)|}^2+K$

$=-\frac{1}{{\mathrm{<figure-callout\; id="0"\; label="N"\; filenames="A20041004272800143.png"\; state="\{\{state\}\}">N</figure-callout>}}_0}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{|r_t^j-\sqrt{E_s}({\mathrm{\&alpha;}}_{1,j}\left(t\right)\&CenterDot;a+{\mathrm{\&alpha;}}_{2,j}\left(t\right)\&CenterDot;b)|}^2+\mathrm{\&Gamma;}(L_0^{(a,+1)}(c_t^1,c_t^3),L_0^{(a,-1)}(c_t^1,c_t^3))+K---\left(5\right)$

And the value of the odd positions of the input signal of corresponding decoding module (DEC1) (300) is the input position of coding module RSC1 (200) $c_t^1=a$ Check digit with coding module RSC2 (204) $c_t^3=c$ Stack, therefore, for RSC1 (200), $c_t^3=c$ Be not $c_t^1=a$ Check digit, and the check digit of RSC1 (200) has been perforated.In this case, we can't use the information in RSC1 (200) trrellis diagram in the maximum likelihood soft-decision, also just say bit c _t ³Do not concern one to one with a.Promptly

${\mathrm{\&gamma;}}_t(r_t,s\text{'},s)=P(r_t,s\text{'},s)$

$=[P\left(r_t\right|c_t^1=a,c_t^3=+1)\&CenterDot;P(c_t^3=+1|c_t^1=a)$

$+P\left(r_t\right|c_t^1=a,c_t^3=-1)\&CenterDot;P(c_t^3=-1|c_t^1=a)]\&CenterDot;P(c_t^1=a)---\left(6\right)$

$\log {\mathrm{\&gamma;}}_t(r_t,s\text{'},s)$

$=\mathrm{\&Gamma;}(L_0^{(a,+1)}(c_t^1,c_t^3)-\frac{1}{{\mathrm{<figure-callout\; id="0"\; label="N"\; filenames="A20041004272800143.png"\; state="\{\{state\}\}">N</figure-callout>}}_0}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{|r_t^j-\sqrt{E_s}({\mathrm{\&alpha;}}_{1,j}\left(t\right)\&CenterDot;a+{\mathrm{\&alpha;}}_{3,j}\left(t\right)\&CenterDot;1)|}^2,$

$L_0^{(a,-1)}(c_t^1,c_t^3)-\frac{1}{{\mathrm{<figure-callout\; id="0"\; label="N"\; filenames="A20041004272800143.png"\; state="\{\{state\}\}">N</figure-callout>}}_0}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{|r_t^j-\sqrt{E_s}({\mathrm{\&alpha;}}_{1,j}\left(t\right)\&CenterDot;a+{\mathrm{\&alpha;}}_{3,j}\left(t\right)\&CenterDot;(-1))|}^2)+K---\left(7\right)$

Here, L ₀ ^{(a ,+1)}(c _t ¹, c _t ³), L ₀ ^{(a ,-1)}(c _t ¹, c _t ³) can be from L _e ^{(a, c)}(c _t ¹, c _t ⁱ) obtain in (i=3), and Г (x, y)=log (e ^x+ e ^y)=max (x, y)+log (1+e ^{-| y-x|}), K is a constant, can calculate L ^{(a, b)}(c _t ¹, c _t ⁱ) time, pruned.We just can obtain external information like this, promptly

$L_e^{(a,b)}(c_t^1,c_t^2)=L^{(a,b)}(c_t^1,c_t^2)-L_0^a\left(c_t^1\right).---\left(8\right)$

Fig. 3 is the decoder architecture figure in the background document 1, external information is imported into next decoding module (DEC2) (304), and external information is deciphered and obtained to decoding module 2 (304) in the same way, again external information is passed to decoding module 1 (300), like this iteration several times after, from decoding module 2 (304) output log-likelihood soft-decision-value, and carry out conclusive judgement.

The improved binary decoding method of Kojima.Kojima on the interpretation method basis of Su, has improved the decoding algorithm of decoding module in the above.He supposes $P(c_t^3=+1|c_t^1=a)=P(c_t^3=-1|c_t^1=a),$ Then (6) and (7) have just become (9) and (10).

${\mathrm{\&gamma;}}_t(r_t,s\text{'},s)=P(r_t,s\text{'},s)$

$=\frac{1}{2}[P\left(r_t\right|c_t^1=a,c_t^3=+1)\&CenterDot;P(c_t^1=a)+P\left(r_t\right|c_t^1=a,c_t^3=-1)\&CenterDot;P(c_t^1=a)]---\left(9\right)$

$\log {\mathrm{\&gamma;}}_t(r_t,s\text{'},s)$

$=\mathrm{\&Gamma;}(L_0^{\left(a\right)}\left(c_t^1\right)-\frac{1}{{\mathrm{<figure-callout\; id="0"\; label="N"\; filenames="A20041004272800143.png"\; state="\{\{state\}\}">N</figure-callout>}}_0}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{|r_t^j-\sqrt{E_s}({\mathrm{\&alpha;}}_{1,j}\left(t\right)\&CenterDot;a+{\mathrm{\&alpha;}}_{3,j}\left(t\right)\&CenterDot;1)|}^2,$

$L_0^{\left(a\right)}\left(c_t^1\right)-\frac{1}{{\mathrm{<figure-callout\; id="0"\; label="N"\; filenames="A20041004272800143.png"\; state="\{\{state\}\}">N</figure-callout>}}_0}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{|r_t^j-\sqrt{E_s}({\mathrm{\&alpha;}}_{1,j}\left(t\right)\&CenterDot;a+{\mathrm{\&alpha;}}_{3,j}\left(t\right)\&CenterDot;(-1))|}^2)+K^{\&prime;}---\left(10\right)$

Therefore, nonbinary decoding has just become binary decoding, and Kojima has isolated the estimated channel value from (8) formula

And external information has just become

$L_e^a\left(c_t^1\right)=L^a\left(c_t^1\right)-L_0^a\left(c_t^1\right)-\tilde{L}\left(c_t^1\right).---\left(11\right)$

The external information of the binary decoding of this Kojima and the decoding of the nonbinary of Su are than more accurate, its decoder architecture as shown in Figure 5, the external information that draws from a decoding module (DEC1) (504) is imported into next decoding module (DEC2) (500), and external information is deciphered and obtained to decoding module 2 (500) in the same way, again external information is passed to decoding module 1 (504), like this iteration several times after, from decoding module 2 (500) output log-likelihood soft-decision-value, and carry out conclusive judgement.

In the binary decoding method of Su, because the check digit of RSC1 (200) has been perforated the bitc that receives _t ³Do not concern one to one in the trrellis diagram of RSC1 (200) with a, in the maximum likelihood soft-decision, can't use the information in RSC1 (200) trrellis diagram, so just produced certain error.

The improved binary decoding method of Kojima, utilize the method for separating the delivery channel value of estimating, improved the decoding performance of Turbo code when empty to a certain extent, but in order to estimate that channel value has increased a functional module, thereby certain decoding complexity increased.The original complexity of the decoder of Turbo code is just very big when empty, and the improvement of Kojima has increased this complexity again, and also can increase along with the increase of code length.Therefore, these two kinds of interpretation methods all have its limitation.

Summary of the invention

The purpose of this invention is to provide a kind of binary decoding method of Turbo code when empty, the decoding performance of Turbo code the time, reduce its complexity when being used to improve sky.

For achieving the above object, Turbo code interpretation method when a kind of binary system is empty comprises step:

Received signal is directly inputted into decoding module 1, and decoding module 1 utilizes received signal and carries out soft decoding from the prior information that decoding module 2 obtains and obtain log-likelihood value and external information, and external information is sent into decoding module 2;

Received signal after interweaving inputs to decoding module 2, and decoding module 2 utilizes the received signal after interweaving and carries out soft decoding from the prior information that decoding module 1 obtains and obtain log-likelihood value and external information, and external information is sent into decoding module 1;

By above-mentioned two steps, carry out repeatedly loop iteration, after iteration several times, the log-likelihood value of decoding module 2 outputs is adjudicated, and output.

The present invention is better than the nonbinary interpretation method of Su on decoding performance, approach the binary decoding method of Kojima; And on decoding complexity, have lower complexity, therefore, be more conducive in real system, use than the nonbinary interpretation method of Su and the binary decoding method of Kojima.

Description of drawings

Turbo code decoder architecture figure when Fig. 1 is to use new binary decoding method empty;

Figure a is the decoder architecture figure with respect to encoder in the background document 1 of Su;

Figure b is the decoder architecture figure with respect to encoder in the background document 2 of Kojima;

Turbo code coder structure figure when Fig. 2 is Su empty in background document 1;

Turbo code decoder architecture figure when Fig. 3 is Su empty in background document 1;

Turbo code coder structure figure when Fig. 4 is Kojima empty in background document 2;

Turbo code decoder architecture figure when Fig. 5 is Kojima empty in background document 2;

The performance comparative graph of the binary decoding method of the decoding of the nonbinary of the new binary decoding method of Turbo code and Su, Kojima when Fig. 6 is sky;

Fig. 7 is in the frequency selective fading channels of ofdm system, and the performance of the binary decoding method of the decoding of the nonbinary of the new binary decoding method of Turbo code and Su, Kojima relatively when empty.

Embodiment

New when empty the decode procedure of Turbo code shown in Fig. 1 .a, the signal r that receiver receives enters decoding module 1 (100), the prior information that decoding module 1 (100) then utilizes r and obtained by decoding module 2 (104), obtain after log-likelihood value and the external information through new binary decoding method, external information is interweaved (102) and it is passed to decoding module 2 (104); Then directly enter decoding module 2 (104) through interleaver (106) received signal afterwards, decoding module 2 (104) utilizes the received signal after interweave (106) and the prior information of coming from decoding module 1 (100) obtains external information and it is sent into decoding module 1 (100); Through after the iteration for several times, the data bit of estimation is adjudicated and exported to the log-likelihood value of decoding module 2 (104) like this.

Interpretation method of the present invention:

1. draw data bit c by decoding module 1 (100) _t ¹Log-likelihood value and external information.Improvement of the present invention mainly is to ask log γ _t(γ _t, s ' is in algorithm s).

1) asks log γ _t(γ _t, s ', s)

Usually, we use maximum likelihood soft-decision (ML) to decipher in decoder.If we consider first RSC coding module (recursive system sign indicating number), be RSC1 (200), and think that the trrellis diagram state changes to s from s ', the value of the even number position of the input signal of then corresponding decoding module (DEC1) (100) is the input position of coding module RSC1 (200) $c_t^1=a$ With its check digit $c_t^2=b$ Stack.In this case, in the trrellis diagram, for specific a, b has only unique value.Therefore, we can use the information in the trrellis diagram in the maximum likelihood soft-decision.

Be log γ _t(γ _t, s ' s) uses likelihood probability $P\left(r_t\right|c_t^1=a,c_t^2=b)$ And prior probability $P(c_t^1=a)$ Try to achieve, as following two formulas,

${\mathrm{\&gamma;}}_t(r_t,s\text{'},s)=P(r_t,s\text{'},s)=P\left(r_t\right|c_t^1=a,c_t^2=b)\&CenterDot;P(c_t^1=a)---\left(12\right)$

$\log {\mathrm{\&gamma;}}_t(r_t,s\text{'},s)$

$=-\frac{1}{{\mathrm{<figure-callout\; id="0"\; label="N"\; filenames="A20041004272800143.png"\; state="\{\{state\}\}">N</figure-callout>}}_0}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{|r_t^j-\sqrt{E_s}({\mathrm{\&alpha;}}_{1,j}\left(t\right)\&CenterDot;a+{\mathrm{\&alpha;}}_{2,j}\left(t\right)\&CenterDot;b)|}^2+L_0\left(c_t^1\right)+K---\left(13\right)$

And the value of the odd positions of the input signal of corresponding decoding module (DEC1) (100) is the input position of coding module RSC1 (200) $c_t^1=a$ Check digit with coding module RSC2 (204) $c_t^3=c$ Stack, therefore, for RSC1 (200), $c_t^3=c$ Be not $c_t^1=a$ Check digit.In this case, we can't use the information in RSC1 (200) trrellis diagram in the maximum likelihood soft-decision, and c _t ³With a is not to concern one to one.Yet at RSC2 (204), c _t ³Be the check digit of a, and relation is one to one arranged, therefore, can utilize the log-likelihood calculations result of RSC2 (204), i.e. the prior information that draws from decoding module 2 (104) $P(c_t^1=a)$ And L ₀(c _t ¹).We can suppose $P\left(r_t\right|c_t^1=a,c_t^3=c)a,c\&Element;(+1,-1)$ Be equiprobable, and its value is for KK,

Be log γ _t(γ _t, s ' s) can be only by prior probability $P(c_t^1=a)$ Try to achieve,

${\mathrm{\&gamma;}}_t(r_t.s\text{'},s)=P(r_t,s\text{'},s)=P(c_t^1=a)\&CenterDot;\mathrm{KK}---\left(14\right)$

$\log {\mathrm{\&gamma;}}_t(r_t,s\text{'},s)=\log \left(L_0\left(c_t^1\right)\right)+K^{\&prime;\&prime;}.---\left(15\right)$

Here, K " be constant, and will calculate Λ _iThe time, pruned.

2) ask α _t(s) and β _t(s)

α _t(s) and β _t(s) can obtain by following two formulas, promptly

${\mathrm{\&alpha;}}_t\left(s\right)=\underset{s\text{'}}{\mathrm{\&Sigma;}}{\mathrm{\&alpha;}}_{t-1}(s\text{'}){\mathrm{\&gamma;}}_t(r_t,s\text{'},s,)---\left(16a\right)$

${\mathrm{\&beta;}}_t\left(s\right)=\underset{s\text{'}}{\mathrm{\&Sigma;}}{\mathrm{\&gamma;}}_{t+1}(r_{t+1},s\text{'},s){\mathrm{\&beta;}}_{t+1}(s\text{'})---\left(16b\right)$

State in the trrellis diagram of s and s ' expression here.

3) ask data bit c _t ¹The log-likelihood ratio of posterior probability

${\mathrm{\&Lambda;}}_t=\log \frac{\underset{c_t^1=+1}{\underset{(s\text{'},s)}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{t-1}(s\text{'}){\mathrm{\&gamma;}}_t(r_t,s\text{'},s){\mathrm{\&beta;}}_t\left(s\right)}{\underset{c_t^1=-1}{\underset{(s\text{'},s)}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{t-1}(s\text{'}){\mathrm{\&gamma;}}_t(r_t,s\text{'},s){\mathrm{\&beta;}}_t\left(s\right)}---\left(17\right)$

4) external information L _e(c _t ¹)

External information can deduct prior information by log-likelihood ratio and try to achieve, promptly

$L_e\left(c_t^1\right)={\mathrm{\&Lambda;}}_i-L_0\left(c_t^1\right).---\left(18\right)$

5) external information is interweaved (102), and pass to decoding module 2 (104) as prior information;

2. be directly inputted into decoding module 2 (104) through interleaver (106) received signal afterwards, decoding module 2 (104) utilization interweave received signal after (106) and the prior information that obtains from decoding module 1 (100) are with 1 in the top step 1) to 5) method obtain external information and it sent into decoding module 1 (100);

3. carry out step 1 to 2 again, like this loop iteration several times after, the log-likelihood value of decoding module 2 (104) output is adjudicated, and output.

Embodiment

To lift several real example here, illustrates.

1. the complexity of the present invention and other two kinds of interpretation methods relatively

Background document 1 and 2 has proposed some different coder structure on the structure, as Fig. 2 and shown in Figure 4.In the decoder architecture of Turbo code, new binary decoding method and apparatus can be used when corresponding sky.

When 1) using as shown in Figure 2 coder structure, its decoder architecture is shown in Fig. 1 .a, input signal y is directly inputted to new binary decoding module 1 (100), be input to new binary decoding module 2 (104) after interweaving by interleaver again, decoding module 1 (100) interweaves the external information that obtains and sends into decoding module 2 (104) after (102), and decoding module 2 (104) carries out the external information that obtains to send into decoding module 1 (100) after the deinterleaving (108).After the circulation repeatedly, export decision value through judgement (110) like this from decoding module 2 (104).

When 2) using as shown in Figure 4 coder structure, its decoder architecture is shown in Fig. 1 .b, input signal y is directly inputted to new binary decoding module 2 (112), undertaken being input to new binary decoding module 1 (116) after the deinterleaving by deinterleaver (114) again, decoding module 1 (116) interweaves the external information that obtains and sends into decoding module 2 (112) after (118), and decoding module 2 (112) carries out the external information that obtains to send into decoding module 1 (116) after the deinterleaving (114).

After the circulation repeatedly, export decision value through judgement (122) like this from decoding module 2 (112).These two kinds of decoders structurally do not have essential distinction, and just the position of decoding module 1, decoding module 2 and interleaver, deinterleaver is different, therefore, can think can not exert an influence to performance, and the complexity of decoder also is the same.

Therefore, here we only analyze three kinds of complexity differences between the interpretation method.

We suppose that N is the input information bits number of decoder, and order

P is the operation of real number product;

A is the real add operation;

E is the operation of real number index;

L is the operation of real number logarithm;

C is two compare operations between the real number.

Fig. 1 .a and Fig. 3 have been to use the decoder architecture figure of the nonbinary interpretation method of new binary decoding method and Su respectively.In the drawings as can be seen, except forming the decoding module difference, other module is just the same, and therefore, we are just passable as long as the complexity of decoding module is formed in contrast.The nonbinary decoding module of Su needs (51E+51L+51C+162P+289A) * N, and new binary decoding module needs (30E+30L+30C+108P+197A) * N, and is as shown in table 1.By this comparison, the operand that the binary decoding module that we can make new advances needs is few, and promptly decoding complexity is lower than the nonbinary decoding of Su.

Fig. 1 .b and Fig. 4 have been to use the decoder architecture of the binary decoding method of new binary decoding method and Kojima respectively.In the drawings as can be seen, except forming the decoding module difference, more than the binary decoding method of Kojima a functional module, promptly estimate the module of channel value.Therefore, we need to calculate simultaneously the complexity of forming decoding module and the complexity of estimating the channel value module.Kojima estimates that the module of channel value needs (2E+2L+2C+36P+27A) * N, and the composition module of its binary decoding needs (33E+33L+33C+162P+232A) * N, and new binary decoding module only needs (30E+30L+30C+108P+197A) * N, and is as shown in table 1.By relatively, we as can be seen, the operand that new binary decoding module needs lacks than the binary decoding module of Kojima, and also lacks a functional block, therefore, decoding complexity is lower than the binary decoding of Kojima.

The binary decoding method that table 1. is new and the complexity of other two kinds of interpretation methods are relatively

Decoding algorithm	The nonbinary interpretation method of Su	The binary decoding method of Kojima	New binary decoding method
				Form decoding module	(51E+51L+51C+162P+289 ?A)*N	(33E+33L+33C+162P+232 ?A)*N	(30E+30L+30C+108P+197 ?A)*N
The channel value estimation module	Do not have	(2E+2L+2C+36P+27A)*N	Do not have

2. the performance of the present invention and other two kinds of interpretation methods relatively

Turbo code when we consider two one receive empty, data bit length is 510, and the not punching of tail bit; Used the recursive systematic convolutional code of two (1,5/7), the BPSK modulation; Used " rotation " structure in the background document 1; Also used the odd-even device that satisfies the S=12 of light condition in the background document 2; And iterative times all is 4.We have considered three kinds of Rayleigh fading situations: 1) quasistatic, and promptly decline gain keeps a constant in a frame length, and is mutually independently from this frame to next frame; 2) Fd*Ts=0.001, here, Fd is a maximum doppler frequency, the Ts is-symbol cycle; 3) Fd*Ts=0.01.

Under Doppler's decline situation, we have used the block of channels interleaver of a 35*15.

Fig. 6 is exactly the error rate (BER) curve under three kinds of channel fading situations.What wherein small circle was represented is the performance curve of the nonbinary interpretation method of Su; What starlet was represented is the performance curve of the nonbinary interpretation method of Kojima; And square is represented is the performance curve of new binary decoding method.As can be seen from the figure under all three kinds of channel statuss, new binary decoding method BER performance is than the good 0.2-0.3dB of nonbinary interpretation method of Su, and approaches the binary decoding method of Kojima.

3.0FDM the performance of the present invention and other two kinds of interpretation methods relatively in the system

Consistent among Turbo code and the embodiment 2 when we consider two one receive empty, promptly data bit length is 510, and the tail bit is not perforated; Used the recursive systematic convolutional code of two (1,5/7), the BPSK modulation; Used " rotation " structure in the background document 1; Also used the odd-even device that satisfies the S=12 of light condition in the background document 2; And iterative times all is 4; Under Doppler's decline situation, used the block of channels interleaver of a 35*15.

We suppose that carrier frequency is 5GHz, and character rate is 20MHz, and the 20MHz bandwidth is divided into 512 subcarriers, and promptly the FFT size is 512.We use the M.1225 channel model A of vehicle environmental, and maximum rate travel is made as 250km/h.

Fig. 7 is exactly in ofdm system, the performance simulation curve of three kinds of interpretation methods in frequency selective fading channels.In Fig. 7, the performance that we can observe new binary decoding method equally is better than the nonbinary interpretation method of Su, and and the binary decoding method of Kojima good equally.

Claims (6)

1. Turbo code interpretation method when a binary system is empty comprises step:

Received signal is directly inputted into decoding module (1), and decoding module (1) utilizes received signal and carries out soft decoding from the prior information that decoding module (2) obtains and obtain log-likelihood value and external information, and external information is sent into decoding module (2);

Received signal after interweaving inputs to decoding module (2), received signal after decoding module (2) utilizes and to interweave and carry out soft decoding from the prior information that decoding module (1) obtains and obtain log-likelihood value and external information, and external information is sent into decoding module (1);

By above-mentioned two steps, carry out repeatedly loop iteration, after iteration several times, the log-likelihood value of decoding module (2) output is adjudicated and exported.

2. by the described method of claim 1, it is characterized in that obtaining log-likelihood value and external information and comprise step:

Ask log γ _t(r _t, s ', s);

Ask α _t(s) and β _t(s);

Ask data bit c _t ¹The log-likelihood ratio of posterior probability;

Ask external information L _e(c _t ¹);

3. by the described method of claim 2, it is characterized in that the described log of asking γ _t(r _t, s ', s) adopt following formula:

${\mathrm{\&Lambda;}}_i=\log \frac{\underset{c_t^1=+1}{\underset{(s^{\&prime;},s)}{\mathrm{\&Sigma;}}}{{\mathrm{\&alpha;}}_{t-1}\left(s^{\&prime;}\right)r}_t(r_t,s^{\&prime;},s){\mathrm{\&beta;}}_t\left(s\right)}{\underset{c_t^1=-1}{\underset{(s^{\&prime;},s)}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{t-1}\left(s^{\&prime;}\right)r_t(r_t,s^{\&prime;},s){\mathrm{\&beta;}}_t\left(s\right)}.$

4. by the described γ of claim 1 _t(r _t, s ', method s) is characterized in that what current decoding module calculated $P\left(r_t\right|c_t^1=a,c_t^3=c)a,c\&Element;(+1,-1)$ Be equiprobable, and only utilize current prior information L ₀(c _t ¹) calculate γ _t(r _t, s ', s).

5. in accordance with the method for claim 4, it is characterized in that described prior information L ₀(c _t ¹) external information L by the output of previous decoding module _e(c _t ¹) obtain.

6. the external information L that it is characterized in that in accordance with the method for claim 3, described previous decoding module output _e(c _t ¹) the log-likelihood ratio Λ that calculates by previous decoding module _iDeduct its prior information and obtain, promptly $L_e\left(C_t^1\right)={\mathrm{\&Lambda;}}_i-L_0\left(c_t^1\right).$

Images