JPH07254861A - Viterbi decoding method and convolutional code encoding and transmission method - Google Patents

Abstract

PURPOSE:To correctly obtain reliability information on each decoded symbol. CONSTITUTION:On the transmission side, an information sequence 21 is divided, and dummy bits 22 of a fixed pattern are inserted to transmit it; and on the reception side, a code trellis is terminated by the dummy bit 22, (always $1 in this example) in the end or each frame, and thereafter, the Viterbi algorithm is executed to determine the only surviving path 23. A difference DELTAk between both path metrics of a counter path 24 merged at each point (k) of time on the path 23 is calculated, and reliability information corresponding to the information symbol on the path 23 at the point of time when mutual information symbols are different before merging of the path 24 with the path 23 after diverging of the path 24 from the path 23 is updated.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a decoding method which, when decoding an error correction code using a Viterbi algorithm, simultaneously outputs a decoding result for each bit and reliability information corresponding thereto. A concrete application is a concatenated code.
In the error correction by, the error rate characteristic is improved by soft-decision decoding the outer code by using the reliability information output in the decoding process of the inner code, or the selection control in the selection diversity is referred to the reliability information. Do, and so on.

[0002]

2. Description of the Related Art For example, in the system configuration shown in FIG. 5A, an information sequence 2 to be transmitted by a transmitting side 1 is encoded (convolutional encoding) by a primary error correction encoder 3, and its encoded sequence 4 (outside) is encoded. Code) is re-encoded (convolutional encoding) by the secondary error correction encoder 5 and its encoded sequence 6 (inner code)
Then, the receiving side 7 decodes the inner code 6 by the Viterbi decoder 8, and then decodes the decoded outer code and the outer code 9 by the Viterbi decoder 10, and outputs a decoding result 11. Consider decoding of such a concatenated code. For simplicity of description, the coded sequence is a binary sequence. Generally, when decoding with the Viterbi algorithm,
The method of calculating the branch metric by the Euclidean distance between the received symbol and the code symbol and performing maximum likelihood decoding (soft-decision decoding) is better than the method of performing the decoding after the received sequence is identified (hard-decision decoding). Error rate characteristic is 2
It is known that it is improved from dB to 3 dB [S. L
in and D.D. J. Costello, Jr. ，
"Error Control Coding: Fun
dentals and Application
s "(Prentice-Hall, pp. 322-3.
28, 1983)]]. However, on the reception side 7 in FIG. 5A, the Viterbi decoder 8 with the inner code 6 outputs only a binary decoded sequence. For this reason, the Viterbi decoder 10 of the outer code 9 that receives this will perform hard-decision decoding, and the gain due to soft-decision decoding cannot be obtained.

On the other hand, J. Hagenauer and P.M. When decoding the inner code, Hoehher calculates the reliability information for each symbol corresponding to the decoded sequence using the posterior probability of the received symbol, and performs soft-decision decoding of the outer code using this and the decoded sequence. I have proposed that [J. Hag
enauer and P.M. Hoeher, “A Vi
terbi Algorithm with Soft
-Decision Outputs and its
Applications ”, IEEE GCOM,
No. 47.1.1, 1989 (Reference 1)]. As shown in FIG. 5B, the configuration of the decoding apparatus outputs the reliability information 13 corresponding to the binary decoding sequence 9 from the Viterbi decoder 12 that decodes the inner code 6. In addition to the conventional Viterbi decoding algorithm, the reliability information 13 is generated by the following iterative algorithm.

The state Sk (0
≦ Sk ≦ 2 ^p− 1, p: constraint length of code), i) store difference Δ between both path metrics of two paths to be merged into Sk ii) initial value of sufficiently large value for reliability information Lk (Sk) Substitute values iii) Time points je (e = 1,2) between j = k−p and j = k−δ (δ: path memory length) of two paths (surviving path and opposing path) to be merged in Sk. , ...) are different from each other, the reliability information Lj corresponding to the information symbol on the surviving path is represented by Lj = min (Lj, αΔ), α: a constant (maximum distance between codes and S of the transmission path). / N)) (C = min (a, b) is the smaller of a and b is C) iv) Decoded sequence and corresponding reliability information Lj are output from the path memory Code trellis Is shown in FIG. 5C. In FIG. 5B, the outer code Viterbi decoder 14 calculates the symbol metric (branch metric) by multiplying the product of the decoded sequence 9 and the corresponding reliability information 13 by the code symbol on the code trellis, and Decoding by the Viterbi algorithm is performed and the decoding result 15 is output. For example, in FIG. 5C, the solid lines 16 and 17 are merged with the state Sk, and the two paths 16 and 17 are diverged in the state Sk-5 at the time point k-5. 5 and k-3, the information symbols are different from each other,
At these time points k-5 and k-3, the reliability information Lj corresponding to the information symbol on the surviving path (in this example, path 16) is updated as described above. The "+1" and "-1" added along each path indicate the information symbol at that time.

[0005]

However, in the above algorithm, since Δ is calculated in each state Sk to update the reliability information Lj, for example, in FIG.
The path metrics of the two paths 18 and 19 indicated by the dotted line to be merged with k are both smaller than the path metrics of the paths 16 and 17 to be merged in the other state S′k at the same time point k (the path metric is small). This means that the path is unlikely to be the correct path on the code trellis), and if the difference Δ ′ is small, then it is on the path with the high path metric (on path 16 in FIG. 5C). However, the reliability information corresponding to the information symbol (underline) which is supposed to be originally correct is updated by the uncertain Δ ′, and there is a possibility that the actual reliability may not be reflected.

The concatenated code has been described above as an example.
The reliability information obtained in the decoding process of the inner code is not limited to the decoding of the outer code. For example, in mobile communication, when selective diversity reception is performed, a mode of use may be considered in which reliability information Lj is referred to for selective control of reception and decoding of each of the plurality of propagation paths. In this case, it is important that the reliability information correctly reflects the reliability of the decoding result. As described above, the conventional method has a drawback that the reliability of the decoding result may not be reflected correctly.

An object of the present invention is to obtain a reliability information corresponding to a decoding result in the process of decoding a convolutional code by the Viterbi algorithm. Two paths are merged into a certain state on the code trellis. Solves the problem that the reliability information corresponding to the information symbol is updated by Δ calculated by the paths that have a small difference and a small difference, and output the reliability information that correctly reflects the decoded sequence and its reliability. A decryption method is provided.

[0008]

According to the present invention, the reliability information is updated and the information symbol and its reliability information are output only for the only surviving path that terminates the code trellis. To do. According to the invention of claim 2, the reliability information is updated and the information symbol and its reliability information are output only for the surviving path having the largest path metric among the surviving paths at each time point.

According to the third aspect of the invention, the transmitting side sorts the information sequence into n bits and inserts m bits (m <n, m: path memory length of the receiving side or more) of a fixed pattern dummy bit. , The information sequence in which the dummy bits are inserted is convolutionally encoded and transmitted, and the receiving side decodes according to the invention of claim 1.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the invention of claim 1 will be described with reference to FIG. In the present invention, the reliability information is updated only for the only surviving path that terminates the code trellis. In order to surely obtain the only surviving path terminated in this way, as shown in FIG. 1A, a dummy bit of a fixed pattern (for example, all -1) of a length m at appropriate intervals with respect to the transmission information sequence 21 on the transmission side. Insert 22. Dummy bit 2
The code length m of 2 is equal to or longer than the length of the path memory in the decoder on the receiving side. Adjacent dummy bit 2
The length n of the information sequence 21 between 2 is greater than m,
Since decoding is delayed by m + n, if it is too large, it may cause a problem, while if n is small, transmission efficiency becomes poor.

On the receiving side, the survivor path in which the code trellis is terminated by the dummy bit 22 is used. However, even if the terminated state is different from that of the dummy bit 22 due to many transmission errors, the state to be terminated should be used. Is known in advance, the path with the larger path metric of the path that has reached that state is used as the only surviving path, and the reliability information is updated for that path.

For example, on the code trellis, as shown in FIG. 1B, the code trellis is terminated by a dummy bit 22 at the rear end of each frame consisting of each n-bit information bit 21 and dummy bits 22 before and after it. Then, by executing the Viterbi algorithm on this, the only surviving path is determined. Next, a difference Δk between the path metric of the opposing path 24 to be merged at each time point k on the surviving path 23 and the path metric of the path 23 at the time point k is calculated. Then, between the time when the path 24 diverges from the path 23 and the time when the information symbols on the path branches are different from each other je (e = 1,
2, ...) The reliability information Lj corresponding to the information symbol on the path 23 is updated from an appropriate initial value Lj by Lj = min (Lj, αΔk), α: constant. This iteration is performed for all k in the frame, and the final Lj is output as the reliability information of the information symbol corresponding to it. For example, when decoding the outer code of the concatenated code, soft-decision decoding can be performed using the decoding result and the reliability information corresponding thereto.

That is, as shown in FIG. 2, a metric computer 26 in an ordinary Viterbi decoder calculates and outputs a branch metric 27 on the code trellis. Using this branch metric 27 and the path metric 29 stored in the path memory 28, the selection of the surviving path in each state and the output of the information symbol 31 are ACS (A
This is performed by the dd-Compare-Select) circuit 32. This output information symbol 29 enters the path memory (since the code trellis is used for termination, the path memory length is the frame length of the inner code) 28, and the surviving path of each state,
Information symbols, paths to be merged, and respective path metrics are stored. Then, when the end of the frame of the code sequence is reached, the difference Δ of the path metric from the opposing path at each time point on the only surviving path is calculated, and the reliability information is updated and stored. When this repetition is finished, the decoded sequence 33 for one frame and the reliability information 34 corresponding thereto are output. The outer code metric computer 35 calculates a branch metric by using the decoding sequence 33, the reliability information 34 corresponding to the decoding code 33, and the code symbol on the code trellis, and the ACS circuit 36 and the path memory 37 perform decoding by the Viterbi algorithm. To do.

Next, an embodiment of the invention of claim 2 will be described.
The invention of claim 2 does not utilize a terminated code trellis and therefore does not require transmission of the dummy bit 22. According to the present invention, among the surviving paths in the path memory, the surviving path having the largest path metric is selected. For example, as shown in FIG. 3A, at time k, paths 41 and 42 are merged into state Sk, and paths 43 and 44 are merged into state S′k.
However, since the corresponding path metric of the surviving paths 41 and 44 is larger in the path 41, only this path 41 is selected. Next, the difference Δ′k between the path metric of the opposing path 42 to be merged at each time k on the surviving path 41 and the path metric of the path 41 at time k is calculated. Then, from the time when the path 42 is diverged from 41 to the time when the information symbols on the path branch are different from each other je (e = 1, 2, ...)
The reliability information L'j corresponding to the information symbol on the path 41 at is updated from an appropriate initial value L'j by L'j = min (L'j, αΔ'k), α: constant. This repetition is performed for all k in the path memory, and L'j corresponding to the information symbol output from the path memory is output as reliability information corresponding to the information symbol. For example, when decoding the outer code of the concatenated code, soft-decision decoding can be performed using the decoding result and the reliability information corresponding thereto.

That is, for example, as shown in FIG. 3B by attaching the same reference numerals to the portions corresponding to FIG. 2, the branch metric 27 calculated by the metric computer 26 and the path metric 29 from the path memory 40 in the ACS circuit 45. Using and to select the surviving path in each state, calculate the difference Δ in the path metric between the surviving path having the maximum path metric and the opposing path to be merged with it, and the state having the information symbols 31, Δ46 and the maximum path metric. The number 47 is output to the path memory 40. Path memory (The path memory length is long enough for all paths to merge when the surviving paths in each state go back in the direction of time lag, and usually about 5 to 6 times the constraint length. (Used) 40, the survivor path, information symbol, and path metric of each state are stored, and the reliability information is updated and stored using Δ. Then, the decoded sequence 33 and the reliability information 34 corresponding thereto are output. The outer code metric computer 35 calculates a branch metric by using the decoding sequence 33, the reliability information 34 corresponding thereto and the code symbol on the code trellis, and the ACS circuit 36 and the path memory 37 perform decoding by the Viterbi algorithm.

[0016]

As described above, according to the first aspect of the present invention, Δ is calculated and reliability is calculated only for the only surviving path terminated on the code trellis (and merged into a known state). Since the information is updated, there is no possibility that the reliability information is updated by Δ calculated between the paths having a low path metric, that is, the possibility that the path is correct.

Similarly, according to the second aspect of the present invention, since reliability information is updated only for the surviving path having the maximum path metric among the paths to be merged at each time point, it is calculated between paths having low path metrics. There is no possibility that the reliability information will be updated by Δ. The error rate characteristics after decoding the concatenated code by simulation using the methods of the inventions of claims 1 and 2 are shown by curves 51 and 52 in FIG. 4, respectively. Curves 53 and 54 are shown with reference to the error rate characteristic after hard-decision decoding and the error rate characteristic after decoding by the method shown in the above-mentioned Document 1, respectively. The horizontal axis and the vertical axis are the error rate after the inner code decoding and the outer code decoding, respectively. The coding rates of the inner code and the outer code are both 1/2 and the constraint length is 7. Interleaving was used to randomize the errors in the output sequence of the inner code decoder. Its size was 40 × 32. From this simulation result,
By performing the decoding according to the inventions of claims 1 and 2, the error rate characteristic is greatly improved as compared with the case where the hard decision decoding is performed (the error rate is about two digits in the decoding of claim 1, and the decoding of claim 2 is performed. It is understood that it is about 1 digit). On the other hand, even if the decoding by the method of Hagenauer et al. Is performed, the improvement on the hard-decision decoding is negligible.

The present invention is not limited to the case where the outer code is soft-decision-decoded by using the reliability information output in the decoding process of the inner code in the error correction by the concatenated code, but the error-correction decoding of the reception decoding of a plurality of propagation paths is performed. However, the present invention can also be applied to diversity reception in which the decoding result is selected using reliability information, and so on. Although the information sequence is divided in FIG. 1A, dummy bits may be added only at the end of the transmission information sequence.

[Brief description of drawings]

1A is a diagram showing an example of a transmission frame suitable for a decoding method according to the invention of claim 1, and FIG. 1B is a diagram showing an example of a code trellis when performing decoding according to the invention of claim 1;

FIG. 2 is a block diagram showing a decoding device to which the invention of claim 1 is applied.

3A is a diagram showing an example of a code trellis when performing decoding according to the invention of claim 2, and FIG. 3B is a block diagram showing a decoding device to which the invention of claim 2 is applied.

FIG. 4 is a diagram showing simulation results of error rate characteristics after concatenated code decoding using the method of the present invention and the conventional method.

FIG. 5A is a block diagram showing coded transmission using a concatenated code and reception and decoding thereof, and B is a decoding of the concatenated code.
FIG. 3 is a block diagram showing an apparatus for outputting Viterbi decoding and reliability information together with a result to perform soft decision decoding of an outer code, and FIG. C is a diagram showing an example of a code trellis when decoding is performed by a conventional method.

Claims (3)

[Claims] 1. A difference in path metric between a pair of paths that decodes an input code sequence by a Viterbi algorithm on a code trellis, and merges from each state at a certain time on the code trellis at each time k on a surviving path. Δk is calculated, and a large value (probably) in the path metric of the pair of paths at a time point j when the information symbols on the path branches are different between the diverging and merging of the pair of paths. The reliability information Lj corresponding to the information symbol on the other path is updated from the initial value Lj by Lj = min (Lj, αΔk), α: a constant, and the final Lj corresponds to the information symbol. In the Viterbi decoding method that outputs the reliability information together with the information symbol, the only surviving path that terminates the code trellis is For, Viterbi decoding method and executes updating the reliability information Lj, the information symbols and the output of the reliability information. 2. A difference in path metric between a pair of paths, which decodes an input code sequence by a Viterbi algorithm on a code trellis and merges from each state at a certain time on the code trellis at each time k on a surviving path. Δk is calculated, and a large value (probably) in the path metric of the pair of paths at a time point j when the information symbols on the path branches are different between the diverging and merging of the pair of paths. The reliability information Lj corresponding to the information symbol on the other path is updated from the initial value Lj by Lj = min (Lj, αΔk), α: a constant, and the final Lj corresponds to the information symbol. In the Viterbi decoding method for outputting the reliability information together with the information symbol, the path metric of each surviving path at each time point k is used. The largest for the survivor path only, update of the reliability information Lj, Viterbi decoding method characterized by executing the output of the reliability information and the information symbols. 3. A transmission side divides an information sequence into n bits, and a dummy bit of a fixed pattern of m bits (m <n, m: path memory length of the reception side or more) is inserted, and the dummy bits are The inserted information sequence is convolutionally encoded and transmitted, and the reception sequence is decoded by the Viterbi algorithm on the trellis that terminates the code trellis at the receiving side, and at the time of decoding, the only surviving path on the above code trellis. At each time point k, the difference Δk in the path metric with the opposing path to be merged is calculated, and between the surviving path and the opposing path diverging and merging, information symbols on the path branch are From the initial value Lj, the reliability information Lj corresponding to the information symbol on the surviving path at different time points j is Lj = min (Lj, αΔk), α: constant And the final Lj is output as reliability information corresponding to the information symbol together with the information symbol.