JP2012182559A - Decoder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a decoder in which the operation amount concerning the decoding is reduced when compared with prior art.SOLUTION: In the decoder decoding received data subjected to convolution coding, a Trellis diagram deformed so that the number of bits of encoded data per branch is equal to a common multiple C of logM and the 1/R is used as a Trellis diagram deformed based on a modulation multi-value number M and a code rate R applied to a received signal. Branch metric of each branch of the deformed Trellis diagram is calculated based on the demodulation results of the received signal, and a survival path is specified based on the branch metric thus calculated.

Description

  The present invention relates to a decoding device that decodes convolutionally encoded data in a communication device (receiver).

  The convolutional code is a code that outputs an n-bit code word for k information bits. The convolutional encoder has a delay element, and an n-bit code word is generated under the influence of K × k past information bits. At this time, K is called a constraint length, and k / n is called a coding rate R. In addition, Viterbi decoding is generally used for decoding convolutionally encoded data. In Viterbi decoding, a trellis diagram representing transition of internal states of convolutional codes and input / output at that time is used.

  FIG. 8 shows a configuration example of a convolutional encoder when K = 3 and R = 1/2. FIG. 9 shows a trellis diagram used when Viterbi decoding a convolutional code generated by the encoder of FIG. In the trellis diagram, a square represents a state, and a line connecting the states is called a branch. A route formed by connecting branches is called a path. Each branch has a corresponding “information bit / codeword bit”. In the figure, the branch where the information bit is 0 is indicated by a solid line and the branch where the information bit is 1 is indicated by a dotted line. In general, the information bit of each branch is 1 bit, and the codeword bit is 1 / R bit.

  At the time of Viterbi decoding, first, a branch metric is obtained using a demodulation result of received data. The obtained branch metric is added to the surviving path metric of the state that is the start point of the branch, and in the state that is the end point of the branch, the path with the highest likelihood among the paths that end in that state is the new surviving path. Selected as. This is repeated, and the decoding result of the received data is obtained by the remaining surviving path.

  Here, when obtaining a branch metric, the branch metric is calculated using the likelihood of the received data. When obtaining a branch metric with Viterbi decoding (constraint length K, coding rate R), the likelihood for 1 / R bit codeword bits corresponding to each branch is required.

However, what is obtained by demodulating the received data is M likelihoods corresponding to data of one symbol (that is, log ₂ M bits), where M is the modulation multi-level number of the signal. On the other hand, since Viterbi decoding requires a likelihood of 1 / R bit, the following calculation is required.

(1) A likelihood of one symbol (that is, log ₂ M bits) is obtained by demodulation.
(2) The likelihood per bit is obtained from the likelihood of one symbol.
(3) The likelihood per 1 / R bit is obtained from the likelihood per bit.
For example, Non-Patent Document 1 proposes a calculation method (2) when FSK is used as a modulation method.

Daisuke Nojima, Yuhei Nagao, Masayuki Kurosaki, Hiroshi Ochi, "Soft Decision Viterbi Decoder for FSK Demodulation Under Fast Fading Channel", The Third International Conference on Communications and Electronics (ICCE 2010), Aug., 2010.

  As described above, in order to perform Viterbi decoding, it is necessary to obtain the likelihood per bit from the likelihood of one symbol. However, the calculation for obtaining the likelihood per bit is complicated, and there is a problem that the calculation amount is large.

  Further, in some modulation schemes such as FSK, there is no method for accurately calculating the likelihood of 1 bit from the likelihood for one symbol. In such a case, only an approximate value can be obtained. Therefore, there is an error in the obtained likelihood per bit. For this reason, for example, in the low C / N region, the soft decision gain gradually decreases, and there is a problem that error rate characteristics such as BER characteristics deteriorate.

  The present invention has been made in view of the above, and it is an object of the present invention to obtain a decoding device in which the amount of calculation required for decoding processing is suppressed to be lower than that of the conventional one. Another object is to obtain a decoding device with improved performance.

In order to solve the above-described problems and achieve the object, the present invention is a decoding apparatus for decoding convolutionally-encoded received data, which is a modulation multilevel number M and a coding rate applied to the received signal. As a modified trellis diagram modified based on R, a trellis diagram modified so that the number of bits of encoded data per branch is log ₂ M and a common multiple C of 1 / R is used. A branch metric of each branch of the post-transform trellis diagram is calculated based on the demodulation result, and a surviving path is specified based on the calculated branch metric.

  According to the present invention, there is no need to obtain the likelihood per bit in the decoding process, and there is an effect that it is possible to realize a decoding apparatus in which the amount of calculation required for the decoding process is kept lower than before. Further, in the case of a modulation method in which the likelihood per bit is not accurately obtained, there is an effect that the error rate characteristic can be improved.

FIG. 1 is a diagram illustrating a configuration example of a decoding device according to the present invention. FIG. 2 is a diagram illustrating an example of a trellis diagram. FIG. 3 is a diagram illustrating an example of a trellis diagram after deformation. FIG. 4 is a diagram illustrating an example of a trellis diagram after deformation. FIG. 5 is a diagram illustrating an example of a path metric value. FIG. 6 is a diagram illustrating an example of the branch metric value. FIG. 7 is a diagram illustrating a branch and a branch metric between states Si and Sj included in the trellis diagram after deformation. FIG. 8 is a diagram illustrating a configuration example of a convolutional encoder. FIG. 9 is a diagram illustrating an example of a trellis diagram.

  Embodiments of a decoding device according to the present invention will be described below in detail with reference to the drawings. Note that the present invention is not limited to the embodiments.

Embodiment.
FIG. 1 is a diagram illustrating a configuration example of a decoding device according to the present invention. This decoding apparatus includes a demodulation unit 11, a demodulation result memory 12, a branch metric calculation unit 13, an ACS calculation unit 14, and a path memory 15, and uses a Viterbi decoding method as a decoding method.

  In the decoding apparatus shown in FIG. 1, the demodulator 11 demodulates the received signal. The demodulation result memory 12 receives and stores the demodulation result from the demodulation unit 11. The branch metric calculator 13 calculates a branch metric based on the demodulation result stored in the demodulation result memory 12. The ACS calculation unit 14 performs ACS calculation (details will be described later). The path memory 15 stores a path metric.

First, the Viterbi decoding method in the present embodiment will be described with an example. In this embodiment, when the modulation multilevel number is M, Viterbi decoding is performed using the likelihood itself for one symbol log ₂ M bits obtained by the demodulation unit without obtaining the likelihood for one bit.

For simplification, the modulation multi-level number M is transmitted with 8 bits, that is, 3 bits per modulation symbol (= log ₂ 8), and is generated as a convolutional code with a coding rate R = 1/2 and a constraint length K = 3. An example in which the polynomial uses codes of G1 (D) = 1 + D ² and G2 (D) = 1 + D + D ² will be described. The encoder of this code has the configuration shown in FIG. A trellis diagram is shown in FIG. As in the trellis diagram of FIG. 9, the square represents a state, and a branch with an information bit of 0 is indicated by a solid line and a branch of 1 by a dotted line.

  Since the coding rate R = 1/2, 2-bit encoded data corresponds to each branch. On the other hand, the likelihood for 3 bits of encoded data is obtained from the demodulation result. Since conventional Viterbi decoding requires a 2-bit likelihood, the likelihood of each bit is obtained from the 3-bit likelihood obtained by demodulation, and the sum of the 2-bit likelihood corresponding to each branch is calculated. Let that be the metric for that branch. Here, if the number of bits of the encoded data corresponding to each branch of the trellis diagram is 3 × A (A: natural number), the sum of the likelihoods of the corresponding A symbols is 3 Since it has a likelihood of xA bits, it can be used as a branch metric. Therefore, it is not necessary to calculate the likelihood for each bit. Therefore, consider transforming the trellis diagram as such.

  Since the encoded data of 2 bits corresponds to one state transition in the trellis diagram, the encoded data of 2 × 3 = 6 bits uniquely corresponds to three consecutive state transitions. . Therefore, when the trellis diagram is modified so that the three consecutive state transitions are combined into one state transition, 6-bit encoded data uniquely corresponds to each branch of the trellis diagram after the modification. If the trellis diagram can be modified in this way (the method will be described later), the branch metric can be 6 bits of encoded data, that is, the likelihood of two symbols, so the sum of the likelihood of the two corresponding symbols is taken as the branch metric. do it. That is, the branch metric can be obtained without obtaining the likelihood per bit.

  Hereinafter, a specific example of a method for modifying the trellis diagram of FIG. 2 will be described. When the three consecutive state transitions shown in FIG. 2 are combined into one state transition, a trellis diagram as shown in FIG. 3 is obtained. Note that FIG. 3 shows only eight branches from the state “00” for simplification. Actually, there are 8 branches from other states.

In FIG. 2, for example, there are the following two paths that are shown in bold in the drawing from the state 00 to the state 00 through three state transitions (routes formed by connecting branches).
(Path # 1) State 00 → State 00 → State 00 → State 00 (branches A, B, C)
(Path # 2) State 00 → State 10 → State 01 → State 00 (branches a, b, c)

  Encoded data “00” and information data “0” correspond to the three state transitions of 00 → 00 in pass # 1, respectively. Therefore, the encoded data “000000” and the information data “000” correspond to the state transition by the path # 1 from the state 00 to the state 00. Also, the encoded data “11” and the information data “1” are used for the first state transition of 00 → 10 in pass # 2, and the encoded data “01” and the information data are used for the second state transition of 10 → 01 The encoded data “11” and the information data “0” correspond to “0” and the third state transition from 01 to 00. Therefore, the encoded data “110111” and the information data “100” correspond to the state transition by the path # 2 from the state 00 to the state 00.

  Similarly, encoded data and information data corresponding to all state transitions from each state at the left end to each state at the right end can be determined. Thus, the trellis diagram shown in FIG. 3 can be obtained by combining three consecutive state transitions into one and combining the 6-bit encoded data corresponding to each state transition after the combination.

  In the trellis diagram created in this way, each branch corresponds to encoded data of 6 bits, that is, 2 symbols. The 6-bit likelihood can be obtained by the sum of the likelihoods of two consecutive corresponding symbols, and this can be used as a branch metric as it is. For example, there are two branches (path # 1, path # 2) corresponding to the state transition of 00 → 00 in FIG. 3, and the encoded data corresponding to the first branch (path # 1) is “000000”, the second The encoded data corresponding to the branch (pass # 2) is “110111”. As the metric of the first branch, the sum of the likelihood when the first symbol is 000 and the likelihood when the second symbol is 000 may be used. As the metric of the second branch, the sum of the likelihood when the first symbol is 110 and the likelihood when the second symbol is 111 may be used. The same calculation can be made for other state transition branches.

  The Viterbi decoding using the trellis diagram of FIG. 3 is equivalent to decoding by regarding a code with a coding rate of 1/2 as a code with a coding rate of 3/6.

  This method focuses on 6 (in this case, the least common multiple), which is one of the common multiples of the number of bits 2 of encoded data per branch and the number of bits 3 per symbol in the conventional trellis diagram. This is achieved by putting together a trellis diagram so that the number of bits of encoded data per branch is six.

This process is generalized as follows. Consider a case where the modulation multilevel number is M, that is, log ₂ M bits are transmitted with one modulation symbol, and a convolutional code with a coding rate R is used. If one of the common multiples of log ₂ M and 1 / R is C, the trellis diagram is modified so that the number of bits of encoded data per branch is C. This is realized by combining the continuous C × R state transitions of the conventional trellis diagram into one state transition. Then, decoding is performed using this trellis diagram.

In the conventional trellis diagram, the number of bits of information data corresponding to each branch is 1, the number of bits of encoded data is 1 / R, the number of branches input to each state, and the number of branches output from each state is Both are 2. On the other hand, in the trellis diagram (FIG. 3) in which continuous C × R state transitions used in the present embodiment are combined into one state transition, the number of bits of information data corresponding to each branch is C × R, The number of bits of encoded data is C, the number of branches input to each state, and the number of branches output from each state are both 2 ^{C × R.}

At the time of Viterbi decoding, the sum of the likelihoods of consecutive C / log ₂ M symbols is used as a branch metric. C may be anything as long as it is a common multiple of log ₂ M and 1 / R. The error rate characteristics such as the BER characteristics are the same regardless of the value. However, the smaller the C, the smaller the amount of computation required for Viterbi decoding. Therefore, in order to reduce the amount of calculation, it is desirable that C be the least common multiple of log ₂ M and 1 / R.

Based on the above, the flow of processing in the decoding apparatus of the present embodiment will be described with reference to FIG. The demodulator 11 demodulates the received signal to obtain M likelihoods for each symbol. These values are input and stored in the demodulation result memory 12. The branch metric calculator 13 calculates the branch metric in the trellis diagram using the likelihood stored in the demodulation result memory 12. Specifically, the sum of the likelihood of consecutive C / log ₂ M symbols corresponding to each branch is calculated, and this is used as the branch metric.

The ACS calculation unit 14 performs the following ACS calculation for each state in the trellis diagram. Here, the ACS calculation means processing of Add (addition), Compare (comparison), and Select (selection).
<Step # 1>
First, for all 2 ^{C × R} paths to a certain state, the branch metric calculated by the branch metric calculation unit 13 and the path metric of the surviving path to the state that is the starting point of the branch (stored in the path memory 15) Is added) (Add).
<Step # 2>
Next, the path metrics after the addition are compared (Compare), and the path with the highest likelihood is selected from the 2 ^{C × R} paths as the surviving path (Select). In this case, the path with the largest path metric may be the path with the highest likelihood.
<Step # 3>
Next, the selected survivor path and its path metric are stored in the path memory 15.

  The ACS calculation unit 14 performs the processes of steps # 1 to # 3 for each state in the trellis diagram. When the end of the trellis diagram is reached and Viterbi decoding ends, the original information data is obtained by the surviving path stored in the path memory 15. This is output as a decoding result.

Here, the processing of the ACS calculation unit 14 will be described with a specific example. It is assumed that the modulation multilevel number M = 16, that is, 4 bits are transmitted in one symbol. Further, a convolutional code with constraint length K = 3 and coding rate R = 1/2 is used, and the encoder shown in FIG. 8 is used. The conventional trellis diagram at this time is as shown in FIG. The value of C is the least common multiple of 4 with log ₂ M = 4 and 1 / R = 2. At this time, decoding is performed using the trellis diagram shown in FIG. 4 so that the number of bits of encoded data per branch is 4, that is, the two state transitions of the conventional trellis diagram are combined into one. Do. One branch corresponds to 4 bits of encoded data, which is the same as the number of bits of 1 symbol. Therefore, the likelihood of one symbol obtained by demodulation is used as it is as a branch metric.

  Now, it is assumed that the path metrics of the surviving paths up to the respective states stored in the path memory 15 are as shown in FIG. Further, it is assumed that the branch metric (the likelihood of one symbol) obtained by the branch metric calculation unit 13 is as shown in FIG. First, processing is performed for the path input to state 00. By executing the operation of step # 1, the path metrics of the four paths input to the state 00 are obtained as follows.

(a) Metric of surviving path up to state 00 + branch metric corresponding to encoded data 0000 = 30 + 8 = 38
(b) Metric of surviving path up to state 01 + branch metric corresponding to encoded data 1100 = 31 + 18 = 49
(c) Survival path metric to state 10 + branch metric corresponding to encoded data 0111 = 40 + 17 = 57
(d) Metric of surviving path up to state 11 + branch metric corresponding to encoded data 1011 = 49 + 12 = 61

  Next, in the operation of step # 2, the four path metrics obtained as described above are compared, and the maximum (d) is selected. Then, in the operation of step # 3, information on the path that transitions from the state 11 to the state 00 and the corresponding path metric value (= 61) are stored in the path memory 15.

  The processes in steps # 1 to # 3 are similarly performed for the paths input to the state 01, the state 10, and the state 11.

  The above is the processing of the ACS calculation unit 14. Here, a characteristic point of the trellis diagram used by the decoding apparatus according to the present embodiment is that there may be a plurality of branches that transition from a certain state to a certain state. For example, in the trellis diagram of FIG. 3, there are two branches each transitioning from one state to another state. In this case, the amount of calculation can be reduced by calculating only the path metric of the path that passes through the branch with the highest likelihood among the plurality of branches in the above step # 1.

  The reason why the amount of calculation can be reduced will be described with reference to FIG. In FIG. 7, there are N branches that transition from the state Si to the state Sj, and the branch metrics are bm (0), bm (1),..., Bm (N−1). When the path metric of the surviving path in the state Si stored in the path memory 15 is pm_prev (Si), the path metric calculated in the above step # 1, that is, N branches that transition from the state Si to Sj Each of the path metrics of the N paths that pass is expressed by the following equation (1).

First path metric = pm_prev (Si) + bm (0)
Second path metric = pm_prev (Si) + bm (1)
・
・
・
Nth path metric = pm_prev (Si) + bm (N−1)
... (1)

  In step # 2, the path with the highest likelihood (path metric) is selected from all the paths to the state Sj including the N paths.

  Here, bm (0) to bm (N−1) is the largest one bm (k) (0 ≦ k <N), and among the path metrics shown in the above equation (1) , The (k + 1) th path metric “pm_prev (Si) + bm (k)” becomes the maximum, and the remaining N−1 paths are not selected in step # 2. Therefore, in step # 1, there is no need to calculate the path metrics of N−1 paths not selected in step # 2 from the beginning. That is, the ACS calculation unit 14 of the decoding apparatus according to the present embodiment compares bm (0) to bm (N−1) in step # 1, and selects the one with the largest branch metric from these N values. The path metric of the corresponding path is obtained using the selected branch metric. The same applies to a branch from a state other than Si to Sj. By doing in this way, the number of times of the add operation in the ACS operation can be reduced.

  The above is the Viterbi decoding method in the present embodiment. Note that ASK, PSK, QAM, APSK, FSK, and the like exist as modulation schemes, but the Viterbi decoding method of this embodiment can be applied to any scheme. Also, the present invention can be applied to a method such as M-ary / SS.

  By using the Viterbi decoding method of the present embodiment, an operation for obtaining a 1-bit likelihood becomes unnecessary, and the amount of operation can be reduced. Furthermore, when there are a plurality of branches that change from a certain state to a certain state in the modified trellis diagram, it is possible to reduce the number of addition processes in the ACS calculation. Further, in the case of a modulation scheme such as FSK or M-ary / SS that cannot accurately calculate the likelihood of one bit from the likelihood of one symbol, the Viterbi decoding method of this embodiment can be used to generate an error. The rate characteristics can be improved.

  Next, as an example, a method for obtaining M likelihoods corresponding to one symbol in the case of using the M-value FSK method and the case of using the M-ary / SS method will be described.

  First, a case where M value FSK is used will be described. In this case, one of the M carriers is selected from the transmitter and transmitted. At the time of demodulation at the receiver, synchronous detection or envelope detection is performed on each of the M carriers, and as a result, M values corresponding to the M carriers are obtained. Then, it is considered that the carrier wave corresponding to the maximum value among the M values is transmitted from the transmission side. The M values at this time may be used as M likelihoods corresponding to one symbol.

  Further, as a special method for generating the M-value FSK signal in the transmitter, a method using inverse discrete Fourier transform can be considered. Assuming that the number of points in the inverse discrete Fourier transform is N, when modulation is performed using M-value FSK, a modulation signal is generated using M points out of N points. That is, according to the data to be transmitted, any one of the M points is set to a non-zero value, and the remaining N-1 points are all set to 0 to perform an inverse discrete Fourier transform to generate an M-value FSK signal. .

  In such a case, the demodulator of the receiver performs demodulation by subjecting the received signal to discrete Fourier transform (number of points N). Then, among the obtained N values, M point values used when generating the FSK signal at the transmitter are selected (these are complex numbers). Then, the demodulation process is performed by selecting the one having the maximum absolute value from among the M values. The absolute values of the M point values at this time may be used as M likelihoods corresponding to one symbol (corresponding to envelope detection). Alternatively, the demodulation process may be performed by selecting the largest real part value from among the M values. In this case, the value of the real part of the values of M points may be used as M likelihoods corresponding to one symbol (corresponding to synchronous detection).

Next, a case where the M-ary / SS system is used will be described. In this case, the transmitter prepares M sequences such as a Walsh sequence of sequence length M, and transmits any one of these M sequences to transmit log ₂ M bits with one symbol. On the other hand, in the receiver, the demodulating section correlates the received data with the M sequences to obtain M correlation values. Then, by selecting the maximum of the M pieces, it is determined which of the M series has been transmitted. The M correlation values at this time may be used as M likelihoods corresponding to one symbol.

Thus, in the decoding apparatus according to the present embodiment, the likelihood per bit is not calculated, and the Viterbi decoding is performed using the likelihood per symbol. For example, the modulation multi-level number is M (that is, log ₂ M bits are transmitted with one modulation symbol), and a signal transmitted using a convolutional code with a coding rate R is used for a convolutional code. A trellis diagram modified based on C, which is one of the common multiples of the reciprocal 1 / R of the coding rate and the number of bits log ₂ M per FSK symbol, specifically, the encoded data per branch Decoding is performed using a trellis diagram modified so that the number of bits becomes C. Thereby, it is not necessary to obtain the likelihood per bit, and the amount of calculation can be reduced. Further, when there are a plurality of branches that transition from a certain state to a certain state in the trellis diagram after deformation, the number of addition processes in the ACS calculation can be reduced. Also, in the case of a modulation scheme in which the likelihood of 1 bit cannot be accurately calculated from the likelihood of 1 modulation symbol, it is not affected by errors that occur when obtaining the likelihood of 1 bit, and the error rate such as the BER characteristics. The characteristics can be improved.

  As described above, the decoding apparatus according to the present invention is useful as a decoding apparatus that decodes data in a receiver of convolutionally encoded data.

DESCRIPTION OF SYMBOLS 11 Demodulation part 12 Demodulation result memory 13 Branch metric calculation part 14 ACS calculation part 15 Path memory

Claims (5)

A decoding device for decoding convolutionally encoded received data,
As a modified trellis diagram modified based on the modulation level M and the coding rate R applied to the received signal, the number of bits of the encoded data per branch is log ₂ M and a common multiple C of 1 / R. The branch metric of each branch of the trellis diagram after modification is calculated based on the demodulation result of the received signal, and the surviving path is identified based on the calculated branch metric. A decoding device characterized by the above.   The post-change trellis diagram is created by carrying out a transformation in which the continuous C × R state transitions of the trellis diagram before the transformation are combined into one state transition. The decoding device according to claim 1. Branch metric calculating means for calculating, as the branch metric, a sum of likelihoods of symbols respectively corresponding to the C × R state transitions;
Path specifying means for specifying a surviving path from a predetermined first state to a second state based on the branch metric calculated by the branch metric calculating means and calculating a path metric of the surviving path;
The decoding apparatus according to claim 2, further comprising: In the post-deformation trellis diagram, when there are a plurality of branches that transition from the state immediately before the second state to the second state,
The path specifying means selects a branch having the largest branch metric from among the plurality of existing branches, treats a path passing through the branch as a surviving path to the second state, and sets the first state. The path metric of the surviving path to the second state is calculated by adding the branch metric of the selected branch to the path metric of the surviving path to the state immediately before the second state from The decoding device according to claim 3. The decoding apparatus according to claim 1, wherein C is a least common multiple of log ₂ M and 1 / R.

Images