Symbol determination device, symbol determination method, and program

Abstract

A bit log likelihood ratio generating unit generates a bit log likelihood ratio of a received symbol for each of a plurality of bits. A symbol selecting unit selects a plurality of symbols adjacent to the received symbol. A candidate sequence generating unit generates a plurality of candidate sequences by combining the symbols adjacent to the received symbol in the time sequence. A maximum posterior probability estimating unit uses both an output obtained by whitening the time sequence data of branched received symbols and the candidate sequences reflecting a channel response, to calculate the log likelihood ratio of the adjacent symbols so that the posterior probability is maximized. A likelihood ratio selecting unit selects, on the basis of bits obtained by conversion from the adjacent symbols, a bit log likelihood that is to be corrected. An addition unit performs a weighted addition of the selected bit log likelihood ratio and a correction value that is based on the difference between the bit log likelihood ratio and the log likelihood ratio.

Description

[Technical Field]

[0001] The present invention relates to a symbol determination device, a symbol determination method, and a program. [Background technology]

[0002] In recent years, coherent optical transmission systems have become increasingly multi-level, and the impact of bandwidth narrowing and device nonlinearity on signal quality due to this multi-level approach has been increasing. In a determination method for determining transmitted symbols without using sequence estimation, noise enhancement occurs when the effects of bandwidth narrowing and device nonlinearity are significant. On the other hand, maximum a posteriori (MAP) estimation using sequence estimation is well known as a technique for compensating for bandwidth narrowing and device nonlinearity on the receiving side without noise enhancement. However, MAP estimation requires sequence estimation for all possible candidate sequences (all states). Therefore, the number of states increases exponentially with the multilevel level and candidate symbol length. Thus, MAP estimation has a problem with the computational scale. To address this issue, there is a technique for performing sequence estimation by limiting candidate symbol sequences (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2020-184696 Summary of the Invention [Problem to be solved by the invention]

[0004] The technology of Patent Document 1 described above is based on hard decision, which makes it difficult to support soft decision error correction codes that are generally used in coherent optical transmission.

[0005] In view of the above circumstances, an object of the present invention is to provide a symbol decision device, a symbol decision method, and a program that are capable of performing soft decisions while suppressing the occurrence of noise enhancement and an increase in the amount of calculation. [Means for solving the problem]

[0006] A symbol decision device according to one aspect of the present invention includes a bit log likelihood ratio generation unit that generates a bit log likelihood ratio of a received symbol for each bit, a symbol selection unit that selects a plurality of neighboring symbols of the branched received symbols, a candidate sequence generation unit that generates a plurality of candidate sequences by combining the neighboring symbols selected by the symbol selection unit for each of the time-series received symbols, a channel response reproduction unit that reflects a channel response in the candidate sequences, and a posterior probability maximization unit that uses an output of whitened time-series data of the branched received symbols and the candidate sequences reflecting the channel response. a symbol-to-bit converter that converts the adjacent symbols selected by the symbol selector into bits; a likelihood ratio selector that selects a bit log-likelihood to be corrected from the bit log-likelihood ratios generated by the bit log-likelihood ratio generator based on the bits converted by the symbol-to-bit converter; and an adder that performs weighted addition of the bit log-likelihood ratio selected by the likelihood ratio selector and a correction value based on the difference between the bit log-likelihood ratio and the log-likelihood ratio calculated by the posterior probability maximum estimator.

[0007] A symbol decision method according to one aspect of the present invention includes a bit log likelihood ratio generation step of generating a bit log likelihood ratio of a received symbol for each bit, a symbol selection step of selecting a plurality of adjacent symbols of the branched received symbols, a candidate sequence generation step of generating a plurality of candidate sequences by combining the adjacent symbols selected in the symbol selection step for each of the time-series received symbols, a channel response reproduction step of reflecting a channel response in the candidate sequences, and a step of selecting the adjacent symbols so as to maximize a posteriori probability by using an output of whitened time-series data of the branched received symbols and the candidate sequences reflecting the channel response. a symbol-to-bit conversion step of converting the adjacent symbols selected in the symbol selection step into bits; a likelihood ratio selection step of selecting a bit log-likelihood ratio to be corrected from the bit log-likelihood ratios generated in the bit log-likelihood ratio generation step, based on the bits converted in the symbol-to-bit conversion step; and an addition step of performing weighted addition of the bit log-likelihood ratio selected in the likelihood ratio selection step and a correction value based on the difference between the bit log-likelihood ratio and the log-likelihood ratio calculated in the maximum a posteriori probability estimation step.

[0008] A program according to one aspect of the present invention is a program for causing a computer to function as the symbol decision device described above. [Effects of the Invention]

[0009] According to the present invention, it is possible to perform soft decisions while suppressing the occurrence of noise enhancement and an increase in the amount of calculation. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a configuration diagram of an optical transmission system according to an embodiment of the present invention. [Figure 2] FIG. 1 is a diagram illustrating the configuration of a soft decision unit using a conventional technique. [Figure 3] FIG. 1 is a diagram illustrating the configuration of a soft decision unit using a conventional technique. [Figure 4] FIG. 10 is a diagram illustrating the configuration of a symbol determination unit using a conventional technique. [Figure 5] FIG. 2 is a configuration diagram of a soft decision unit according to an embodiment. [Figure 6] FIG. 1 illustrates the selection of candidate symbols according to an embodiment. [Figure 7] FIG. 2 illustrates a trellis diagram according to an embodiment. [Figure 8] FIG. 1 is a diagram illustrating a numerical simulation system. [Figure 9] FIG. 10 is a diagram showing evaluation results by numerical simulation. [Figure 10] FIG. 1 illustrates a hardware configuration of an optical transmission device according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] An embodiment of the present invention will be described in detail below with reference to the drawings. A symbol decision device according to this embodiment combines a soft decision method that does not use sequence estimation with sequence estimation using tentative decision symbols. This reduces the amount of computation and enables soft decision support while reducing noise enhancement due to bandwidth narrowing and device nonlinearity. Specifically, the symbol decision device generates a replica transmission symbol sequence of a candidate signal using nearby symbols of a received signal that has been adaptively equalized. The symbol decision device performs maximum a posteriori estimation using the replica transmission symbol sequence and the received symbol sequence after application of a whitening filter. The symbol decision device combines a soft decision value (log-likelihood ratio) obtained by a soft decision method that does not use sequence estimation with the log-likelihood ratio of bits corresponding to nearby symbols obtained by maximum a posteriori estimation. The symbol decision device limits the number of nearby symbols used to generate the replica transmission symbol sequence, thereby reducing the number of states in maximum a posteriori estimation regardless of the multilevel level. The symbol decision device can also output soft decision values ​​for all bits.

[0012] 1 is a diagram showing an example of the configuration of an optical transmission system 1 according to an embodiment of the present invention. The optical transmission system 1 includes an optical transmission device 2 and an optical transmission device 3. The optical transmission device 2 and the optical transmission device 3 are connected by an optical transmission path 4.

[0013] The optical transmission device 2 includes an optical transmitter 20. The optical transmitter 20 includes an encoding unit 21, a symbol mapping unit 22, a waveform shaping unit 23, a digital-to-analog conversion unit 24, an optical modulation unit 25, and an optical equalization unit 26. The optical transmitter 20 may be an optical transmitter of a general optical transmission device such as a digital coherent optical transmission device or an intensity modulation / direct detection optical transmission device.

[0014] The encoding unit 21 encodes the input bit sequence using an arbitrary error correction code. The symbol mapping unit 22 maps the bit sequence input from the encoding unit 21 to arbitrary symbols using QPSK (Quadrature Phase Shift Keying), QAM (Quadrature Amplitude Modulation), or the like. The waveform shaping unit 23 performs Nyquist filtering or the like on the symbol sequence input from the symbol mapping unit 22. At this time, the waveform shaping unit 23 may also pre-equalize the inverse characteristics of the transfer function of the analog device used in the optical transmitter 20 or the optical transmission line 4.

[0015] The digital-to-analog converter 24 converts the symbol sequence of the digital signal input from the waveform shaping unit 23 into an analog signal. The optical modulator 25 converts the electrical analog signal input from the digital-to-analog converter 24 into an optical signal. The optical equalizer 26 equalizes the transmission signal, which is an optical signal input from the optical modulator 25. The optical equalizer 26 may equalize the transfer functions of the optical transmitter 20, the optical transmission line 4, the optical receiver 30, etc. The optical equalizer 26 outputs the equalized optical signal to the optical transmission line 4. Note that the optical transmitter 20 may be configured without the optical equalizer 26.

[0016] The optical transmission device 3 is an example of a symbol decision device. The optical transmission device 3 includes an optical receiver 30. The optical receiver 30 includes an optical equalizer 31, an optical detector 32, an analog-to-digital converter 33, a chromatic dispersion compensator 34, an adaptive equalizer 35, a soft decision unit 36, and a decoder 37. The optical equalizer 31, the optical detector 32, the analog-to-digital converter 33, the chromatic dispersion compensator 34, the adaptive equalizer 35, and the decoder 37 may be those used in receivers of general optical transmission devices such as digital coherent optical transmission devices or intensity modulation / direct detection optical transmission devices.

[0017] The optical equalizer 31 equalizes the optical signal received from the optical transmitter 20 via the optical transmission line 4. The optical equalizer 31 may equalize the transfer functions of the optical transmitter 20, the optical transmission line 4, the optical receiver 30, etc. The optical receiver 30 may not have the optical equalizer 31. The optical detector 32 converts the signal modulated with the carrier frequency into a baseband analog electrical signal by coherent detection or square-law detection, which causes interference between the optical signal equalized by the optical equalizer 31 and local light. The analog-to-digital converter 33 converts the received signal, which has been converted into an analog electrical signal by the optical detector 32, into a digital signal.

[0018] The chromatic dispersion compensator 34 equalizes the received signal converted into a digital signal. Specifically, the chromatic dispersion compensator 34 equalizes the chromatic dispersion occurring in the optical transmission line 4 by digital signal processing such as an FIR (Finite Impulse Response) filter or frequency domain equalization. At this time, the chromatic dispersion compensator 34 may simultaneously perform waveform shaping, such as compensation for the transfer function of analog devices in the optical receiver 30. The adaptive equalizer 35 adaptively equalizes the received signal equalized by the chromatic dispersion compensator 34. That is, the adaptive equalizer 35 dynamically estimates and compensates for the dynamically fluctuating polarization state, laser frequency offset, phase noise, clock phase, and other factors occurring in the optical transmission line 4 by digital signal processing such as an FIR filter or frequency domain equalization. The adaptive equalizer 35 also operates matched filters according to the noise added in the optical transmitter 20, the optical transmission line 4, and the optical receiver 30.

[0019] The soft decision unit 36 ​​calculates a log-likelihood ratio, which is a soft decision value for each received bit, from the received symbol sequence of the received signal adaptively equalized by the adaptive equalization unit 35, and outputs the calculated log-likelihood ratio. A specific configuration of the soft decision unit 36 ​​will be described later. The decoding unit 37 performs error correction on the bit sequence or bit likelihood sequence input from the soft decision unit 36.

[0020] The optical transmission path 4 transmits the optical signal output by the optical transmitter 20 to the optical receiver 30. The optical transmission path 4 has an optical fiber 41 and an optical amplifier 42. The optical fiber 41 connects the optical transmitter 20 or the optical amplifier 42 at the upstream stage to the optical amplifier 42 or the optical receiver 30 at the downstream stage. The optical amplifier 42 amplifies the optical signal transmitted through the optical fiber 41 on the input side, and inputs the amplified optical signal to the optical fiber 41 on the output side. The number of optical amplifiers 42 is arbitrary.

[0021] The optical transmission device 2 may further include an optical receiver 30, and the optical transmission device 3 may further include an optical transmitter 20.

[0022] Next, a symbol decision unit that performs soft decision using conventional technology will be described, followed by a description of the soft decision unit 36 ​​according to this embodiment.

[0023] FIG. 2 shows the configuration of the soft decision unit 71. The soft decision unit 71 generates a bit log likelihood ratio using only each symbol at a certain time without using sequence estimation. Although the processing is simple, noise enhancement occurs in a transmission / reception system where the influence of band narrowing is significant. The soft decision unit 71 has a bit log likelihood ratio generation unit 72 and a multiplication unit 73.

[0024] The bit log likelihood ratio generator 72 receives, for example, a received symbol r similar to the output from the adaptive equalizer 35 shown in Fig. 1. The bit log likelihood ratio generator 72 generates a bit log likelihood ratio λ for each bit i from the received symbol r. i Generate each log-likelihood ratio λ i The multiplication unit 73 multiplies the bit log likelihood ratio λ output from the bit log likelihood ratio generation unit 72 by i , the scale parameter β i (βi is a real number greater than or equal to 0). i is a parameter for adjusting the error correction number in the subsequent stage to be minimized. i The bit log-likelihood ratio λ multiplied by i The decoding unit 37 outputs the result to the subsequent decoding unit. The subsequent decoding unit performs the same process as the decoding unit 37 shown in FIG.

[0025] The bit log-likelihood ratio generator 72 calculates the log-likelihood ratio λ between the received symbol r and each bit i using the following equation (1): i The equation (1) is a calculation formula for bit-wise LLR (Log-Likelihood Ratio) to obtain an exact solution.

[0026]

number

[0027] When the received signal is a quaternary signal, the bit log-likelihood ratio generator 72 calculates the log-likelihood ratio λ between the received symbol r and each bit i using the following equations (2) and (3): i The equations (2) and (3) are equations for calculating the approximate bit-wise LLR in a quaternary signal.

[0028]

number

[0029]

number

[0030] Note that MSB is the most significant bit and LSB is the least significant bit. MSB is the bit log-likelihood ratio of the MSB, and λ LSB is the bit log-likelihood ratio of the LSB.

[0031] Alternatively, the bit log-likelihood ratio generating unit 72 may obtain the bit-wise LLR as an approximation using a lookup table.

[0032] 3 is a diagram showing the configuration of a soft decision unit 81 that uses sequence estimation. When sequence estimation is used, calculations are required for all candidate sequences. The soft decision unit 81 has a whitening filter unit 82, a candidate sequence generator 83, a channel response reproducing unit 84, a posterior probability maximum estimator 85, and a multiplier 86.

[0033] The whitening filter unit 82 receives received symbols similar to those output from the adaptive equalizer 35 shown in Fig. 1. The whitening filter unit 82 applies a filter to the received symbols to whiten the noise spectrum contained in the received signals. This filter is implemented using an FIR filter or the like.

[0034] A candidate sequence generator 83 generates candidate sequences, which are replicas of the transmitted signal sequence, for all candidate combinations. A channel response reproducing unit 84 applies a linear / nonlinear filter to each candidate sequence, simulating the signal spectrum and nonlinearity after output from the whitening filter unit 82. The linear filter is implemented using an FIR filter or the like, and the nonlinear filter is implemented using a Volterra filter or the like.

[0035] The maximum a posteriori probability estimation unit 85 uses the output of the whitening filter unit 82 and the output from the channel response reproduction unit 84 to calculate the log-likelihood ratio λ of the i-th bit of the symbol at the timing of time t by the BCJR algorithm. i The multiplication unit 86 calculates the bit log likelihood ratio λ output from the maximum a posteriori probability estimation unit 85. i , the scale parameter β i (β i is a real number greater than or equal to 0). i is a parameter for adjusting the number of corrections in the subsequent error correction stage to be minimized.

[0036] 4 is a diagram showing the configuration of the symbol decision unit 91. The symbol decision unit 91 applies, for example, the technology of Patent Document 1. The symbol decision unit 91 outputs a symbol likelihood ratio for hard decision. The symbol decision unit 91 has a whitening filter unit 92, a symbol selection unit 93, a candidate sequence generation unit 94, a channel response reproduction unit 95, and a posterior probability maximum estimation unit 96.

[0037] The whitening filter unit 92 receives received symbols similar to those output from the adaptive equalizer 35 shown in Fig. 1. The whitening filter unit 92 applies a filter to the received symbols to whiten the noise spectrum contained in the received signals. This filter is implemented using an FIR filter or the like.

[0038] The symbol selection unit 93 performs a tentative decision on the received signal to select candidate symbols for the received symbol. The candidate symbols are the closest symbol, second closest symbol, second closest symbol, and so on to the received symbol. The candidate sequence generation unit 94 generates a candidate sequence by combining the candidate symbols selected by the symbol selection unit 93 for each symbol at each time in time sequence. The channel response reproduction unit 95 applies a linear / nonlinear filter that simulates the signal spectrum and nonlinearity after output from the whitening filter unit 92. The linear filter is implemented using an FIR filter or the like, and the nonlinear filter is implemented using a Volterra filter or the like.

[0039] The maximum a posteriori probability estimation unit 96 uses the output of the whitening filter unit 92 and the output from the channel response reproduction unit 95 to calculate the symbol likelihood ratio between the candidate symbols at time t using the BCJR algorithm.

[0040] 5 is a diagram showing the configuration of the soft decision unit 36 ​​according to this embodiment. The soft decision unit 36 ​​includes a branching unit 51, a bit log-likelihood ratio generation unit 52, a whitening filter unit 53, a symbol selection unit 54, a candidate sequence generation unit 55, a channel response reproduction unit 56, a posterior probability maximum estimation unit 57, a symbol-to-bit conversion unit 58, a likelihood ratio selection unit 59, a weighting unit 61, a weighting unit 62, an addition unit 63, and a multiplication unit 64. The bit log-likelihood ratio generation unit 52 can be the bit log-likelihood ratio generation unit 72 shown in FIG. 2. The whitening filter unit 53, the symbol selection unit 54, the candidate sequence generation unit 55, the channel response reproduction unit 56, and the posterior probability maximum estimation unit 57 can be the whitening filter unit 92, the symbol selection unit 93, the candidate sequence generation unit 94, the channel response reproduction unit 95, and the posterior probability maximum estimation unit 96 shown in FIG. 4.

[0041] The branching unit 51 receives the received symbols of the received signal output by the adaptive equalization unit 35 in time series and branches them into three. The branching unit 51 outputs the three branched received symbols to a bit log-likelihood ratio generation unit 52, a whitening filter unit 53, and a symbol selection unit 54, respectively.

[0042] The bit log likelihood ratio generator 52 calculates the bit log likelihood ratio λ of each bit i from the received symbol without using sequence estimation. i If the number of bits of the value represented by one symbol is N, then i = 1 to N. The bit log likelihood ratio λ i Any conventional technique can be used to calculate .

[0043] The whitening filter unit 53 applies a filter that whitens the frequency spectrum of noise to the received symbols in time series, and outputs the symbols after the filter application.

[0044] The symbol selection unit 54 selects a (a is an integer equal to or greater than 2) candidate symbols for the received symbol through tentative decision. The following description will be given taking the case where a=2 as an example. The symbol selection unit 54 selects the nearest symbol to the received symbol r(t) and the second nearest symbol. The symbol selection unit 54 outputs the selected candidate symbols to the candidate sequence generation unit 55 and the symbol-to-bit conversion unit 58.

[0045] The candidate sequence generator 55 generates multiple candidate sequences in which the candidate symbols are arranged in time series by combining the candidate symbols selected by the symbol selector 54. The channel response reproducing unit 56 applies a channel response reproduced by a linear filter, a nonlinear filter, or the like to each candidate sequence generated by the candidate sequence generator 55.

[0046] The posterior probability maximum estimator 57 calculates the posterior probability taking into account the neighboring sequences between the nearest symbol and the second nearest symbol, using the BCJR algorithm or the like, based on the output of the whitening filter 53 and the output from the channel response reproducing unit 56. The posterior probability maximum estimator 57 calculates the symbol log-likelihood ratio λ obtained by the logarithm of the ratio between the posterior probability of the nearest symbol and the posterior probability of the second nearest symbol. MAP Output.

[0047] The symbol-to-bit converter 58 converts the nearest neighbor symbol and the second nearest neighbor symbol selected by the symbol selector 54 into bit strings of the values ​​represented by those symbols. The symbol-to-bit converter 58 converts each bit b i and each bit c converted from the next nearest symbol i is output to the likelihood ratio selection unit 59.

[0048] The likelihood ratio selector 59 selects the bit log likelihood ratio λ output from the bit log likelihood ratio generator 52. i and the symbol log-likelihood ratio λ output by the maximum a posteriori probability estimation unit 57. MAP and the bit b of the nearest symbol output by the symbol-to-bit converter 58. i and bit c of the next closest symbol iUsing the above, the correction value Δλ is calculated using the following equation (4): i Calculate.

[0049] Δλ i =(b i -c i )(λ i -λ MAP ) …(4)

[0050] Bit b of the nearest symbol i and the bit c of the next closest symbol i If and are the same, (b i -c i ) is 0. In this case, the correction value Δλ i is 0, the bit log likelihood ratio λ output from the bit log likelihood ratio generator 52 is i This means that bit i is not used to correct the bit log-likelihood ratio λ i This corresponds to the fact that the bit b of the nearest symbol is not selected as a target for correction. i and the bit c of the next closest symbol i If different from (b i -c i ) is -1 or 1. In this case, the correction value Δλ i is the bit log likelihood ratio λ of bit i output from the bit log likelihood ratio generator 52. i This means that bit i has a bit log-likelihood ratio λ i The likelihood ratio selection unit 59 selects the calculated correction value Δλ as a correction target. i Output.

[0051] The weighting unit 61 calculates the log likelihood ratio λ output from the bit log likelihood ratio generation unit 52. i Alpha i (α i is a real number between 0 and 1). i to (1-α i The adder 63 multiplies the weighted signal α i The bit log-likelihood ratio λ weighted by i and the weighting unit 62 calculates (1-α i) weighted correction value Δλ i The adder 63 outputs the bit log likelihood ratio of the addition result to the multiplier 64. The multiplier 64 multiplies the bit log likelihood ratio output by the adder 63 by a scale parameter β i (β i is a real number greater than or equal to 0). i is a parameter for adjusting the number of corrections in the subsequent error correction so as to be minimized. The multiplication unit 64 outputs the multiplication result to the decoding unit 37.

[0052] In addition, α i When α = 1, the operation is the same as soft decision when sequence estimation is not used. i is adjusted in accordance with the optical transmission line 4 so that the amount of information to the decoding unit 37 is maximized.

[0053] A specific example of the operation of the soft decision unit 36 ​​will be described. The branching unit 51 branches the received symbol r output by the adaptive equalization unit 35 into three. The received symbol r is divided into two N Hereinafter, the received symbol r received at time t will be referred to as received symbol r(t). The branching unit 51 outputs the three branched received symbols r(t) to a bit log-likelihood ratio generation unit 52, a whitening filter unit 53, and a symbol selection unit 54, respectively.

[0054] The bit log likelihood ratio generator 52 calculates the bit log likelihood ratio λ of each bit i (i is an integer between 1 and N) when the value represented by the received symbol r(t) is converted into an N-bit long bit string. i Calculate.

[0055] The tap length of the filter in the whitening filter unit 53 is assumed to be k. In this case, the whitening filter unit 53 outputs a symbol after filtering using k received symbols r in time series. Here, the tap length k=3. The whitening filter unit 53 outputs a symbol w(t) after filtering using received symbols r(t-1), r(t), and r(t+1).

[0056] 6 is a diagram showing the selection of candidate symbols in the symbol selector 54. The symbol selector 54 selects the nearest symbol s'(t) of the received symbol r(t). 1st and the next nearest symbol s'(t) 2nd The symbol selector 54 outputs the selected candidate symbols to the candidate sequence generator 55 and the symbol-to-bit converter 58.

[0057] For example, when the constraint length in maximum a posteriori estimation is 3, the candidate sequence generator 55 generates eight candidate sequences by combining candidate sequences for the received symbols r(t-1), r(t), and r(t+1) of three time series. Specifically, the candidate sequence x1(t)=s'(t-1) 1st s'(t) 1st s'(t+1) 1st , candidate sequence x2(t)=s'(t-1) 1st s'(t) 1st s'(t+1) 2nd , candidate sequence x3(t)=s'(t-1) 1st s'(t) 2nd s'(t+1) 1st , candidate sequence x4(t)=s'(t-1) 1st s'(t) 2nd s'(t+1) 2nd ,..., candidate sequence x8(t)=s'(t-1) 2nd s'(t) 2nd s'(t+1) 2nd is generated.

[0058] The channel response reproducing unit 56 applies the channel response to each of the candidate sequences x1(t) to x8(t) to generate candidate sequences y1(t) to y8(t). The candidate sequences y1(t) to y8(t) correspond to the symbol sequences of the replica transmission signals. The channel response reproducing unit 56 outputs the generated candidate sequences y1(t) to y8(t) to the posterior probability maximum estimating unit 57.

[0059] The maximum a posteriori probability estimation unit 57 uses the symbol w(t) output from the whitening filter unit 53 and the candidate sequences y1(t) to y8(t) output from the channel response reproduction unit 56 to calculate the nearest symbol s'(t) by maximum a posteriori probability estimation. 1st The posterior probability P(t) 1st and the next nearest symbol s'(t) 2nd The posterior probability P(t) 2nd Ask for.

[0060] 7 is a diagram showing a trellis diagram in the maximum a posteriori probability estimation performed by the maximum a posteriori probability estimation unit 57. In the maximum a posteriori probability estimation, the maximum a posteriori probability estimation unit 57 calculates a candidate sequence y v (t+1) and symbol w(t+1), the candidate sequence y u (t) to candidate sequence y v (t+1) (u, v are integers between 1 and 8). In addition, the maximum posterior probability estimation unit 57 calculates the forward transition probability of the candidate sequence y u (t+1) to candidate sequence y v Then, the maximum posterior probability estimation unit 57 calculates the backward transition probability of the transition to the nearest symbol s'(t). Furthermore, the maximum posterior probability estimation unit 57 updates the forward transition probability by the backward transition probability. The maximum posterior probability estimation unit 57 calculates the nearest symbol s'(t) by using the forward transition probabilities and backward transition probabilities calculated for the candidate sequences y1(t) to y8(t) and the candidate sequences y1(t+1) to y8(t+1). 1st Likelihood P(t) is the posterior probability of 1st and the next nearest symbol s'(t) 2nd Likelihood P(t) is the posterior probability of 2nd The maximum a posteriori probability estimation unit 57 calculates the symbol log likelihood ratio λ between the nearest symbol and the second nearest symbol. MAP =log(P(t) 1st / P(t) 2nd ) is output. In this way, in this embodiment, the candidate sequences are limited to those in the vicinity of the received signal. Furthermore, the number of states does not increase depending on the multi-level degree of the symbol. Therefore, the amount of calculation for maximum a posteriori probability estimation can be reduced.

[0061] The symbol-to-bit converter 58 converts the nearest symbol s'(t)1st Converted bits from b i and the next nearest symbol s'(t) 2nd Bits converted from c i is output to the likelihood ratio selection unit 59.

[0062] The likelihood ratio selector 59 selects the bit log likelihood ratio λ i and the symbol log-likelihood ratio λ MAP and the bit b of the nearest symbol i and bit c of the next closest symbol i Using equation (4), the correction value Δλ corresponding to the i-th bit of the bit string represented by the received symbol r(t) is calculated. i Calculate.

[0063] The adder 63 calculates α i The bit log-likelihood ratio λ weighted by i and (1-α i ) weighted correction value Δλ i The multiplier 64 adds the scale parameter β to the bit log-likelihood ratio output from the adder 63. i (β i is a real number greater than or equal to 0) and output.

[0064] In the above description, the number a of candidate symbols is 2. When a is 3 or more, the symbol selector 54 selects a number of adjacent symbols, such as the first adjacent symbol, the second adjacent symbol, ..., the a-th adjacent symbol, in order of closestness to the received symbol. In this case, the maximum a-posteriori probability estimator 57 calculates the posterior probability of each of the first to a-th adjacent symbols in the maximum a-posteriori probability estimation. The maximum a-posteriori probability estimator 57 selects two adjacent symbols with the highest posterior probabilities from the first to a-th adjacent symbols. The maximum a-posteriori probability estimator 57 uses the selected two adjacent symbols as the nearest symbol and the second nearest symbol, and performs the same processing as in the above-described embodiment. The maximum a-posteriori probability estimator 57 notifies the symbol-to-bit converter 58 of the selected adjacent symbols as the nearest symbol and the second nearest symbol. The symbol-to-bit converter 58 selects adjacent symbols to be used as the nearest symbol and the next nearest symbol from the first nearest symbol, second nearest symbol, ..., ath nearest symbol received from the symbol selector 54 based on the notification from the maximum a posteriori probability estimator 57.

[0065] Next, the results of evaluating the generalized information content by numerical simulation using a four-level signal will be described. Fig. 8 is a diagram showing a numerical simulation system. White noise was added to a band-limited four-level signal and input to an adaptive equalization unit. The signal equalized by the adaptive equalization unit was subjected to soft decision by the soft decision unit 36 ​​of this embodiment, the soft decision unit 71 of the prior art shown in Fig. 2, and the soft decision unit 81 of the prior art shown in Fig. 3.

[0066] FIG. 9 is a diagram showing the generalized information amount of the soft decision results obtained by the soft decision unit 36 ​​of this embodiment and the soft decision units 71 and 81 of the prior art. Prior art (1) shows the soft decision unit 71, and prior art (2) shows the soft decision unit 81. As shown in FIG. 9, it can be confirmed that by using the soft decision unit 36 ​​of this embodiment, a higher information amount can be obtained compared to prior art (1) which does not use sequence estimation. It was also confirmed that the soft decision unit 36 ​​of this embodiment achieved information amount performance equivalent to that of prior art (2) which calculates the posterior probability of sequence transitions for all symbols. The number of states when the memory length is 5 is 4 in prior art (2). 5-1 In the soft decision unit 36 ​​of this embodiment, 5-1 is.

[0067] 10 is a diagram illustrating an example of the hardware configuration of the optical transmission device 3. The optical transmission device 3 includes a processor 101, a storage unit 102, a communication interface 103, and a user interface 104.

[0068] The processor 101 is a central processing unit that performs calculations and control. The processor 101 is, for example, a CPU. The processor 101 reads and executes programs from the storage unit 102. The storage unit 102 further has a work area and the like for the processor 101 to execute various programs. The communication interface 103 is connected to other devices so as to be able to communicate with them. The communication interface includes an optical equalization unit 31, an optical detection unit 32, and an analog-to-digital conversion unit 33. The user interface 104 is an input device such as a keyboard, a pointing device (mouse, tablet, etc.), a button, a touch panel, etc., and a display device such as a display. Human operations are input via the user interface 104.

[0069] Some or all of the functions of the chromatic dispersion compensator 34, adaptive equalizer 35, soft decision unit 36, and decoder 37 are realized by the processor 101 reading and executing a program from the storage unit 102. Note that some or all of these functions may be realized using hardware such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array).

[0070] According to the above-described embodiment, the symbol decision device includes a bit log-likelihood ratio generator, a symbol selector, a candidate sequence generator, a channel response reproducing unit, a posterior probability maximum estimator, a symbol-to-bit converter, a likelihood ratio selector, and an adder. The bit log-likelihood ratio generator generates a bit log-likelihood ratio of a received symbol for each bit. For example, the bit log-likelihood ratio generator generates a bit log-likelihood ratio of a received symbol without using sequence estimation. The symbol selector selects multiple adjacent symbols of the branched received symbols. The candidate sequence generator generates multiple candidate sequences by combining the adjacent symbols selected by the symbol selector for each received symbol in the time series. The channel response reproducing unit reflects the channel response in the candidate sequence. The posterior probability maximum estimator calculates the log-likelihood ratio of the adjacent symbols so as to maximize the posterior probability, using an output obtained by whitening the time series data of the branched received symbols and the candidate sequence reflecting the channel response. The symbol-to-bit converter converts the adjacent symbols selected by the symbol selector into bits. The bit log likelihood ratio generation unit selects a bit log likelihood to be corrected from the bit log likelihood ratios generated by the bit log likelihood ratio generation unit based on the bits converted by the symbol-to-bit conversion unit. The adder unit weights and adds the bit log likelihood ratio selected by the likelihood ratio selection unit and a correction value based on the difference between the bit log likelihood ratio and the log likelihood ratio calculated by the maximum a posteriori probability estimation unit.

[0071] The symbol selector may select the nearest symbol and the second nearest symbol of the branched received symbols as the nearest symbols. When a bit obtained by converting the nearest symbol differs from a bit obtained by converting the second nearest symbol, the likelihood ratio selector selects a bit log-likelihood ratio corresponding to the bit as a target for correction.

[0072] The symbol selector may select three or more adjacent symbols from the branched received symbols. The posterior probability maximization estimator calculates the likelihood of each adjacent symbol so as to maximize the posterior probability using an output obtained by whitening the time-series data of the branched received symbols and a candidate sequence to which the channel response has been added by the channel response reproducing unit, and calculates a log-likelihood ratio of the two selected adjacent symbols based on the calculated likelihoods. The symbol-to-bit converter converts the two adjacent symbols selected by the posterior probability maximization estimator into bits.

[0073] Although an embodiment of the present invention has been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs within the scope of the gist of the present invention. [Explanation of symbols]

[0074] 1...optical transmission system, 2...optical transmission equipment, 3...optical transmission equipment, 4...optical transmission path, 20...optical transmitter, 21...encoding unit, 22...symbol mapping unit, 23...waveform shaping unit, 24...digital-analog conversion unit, 25...optical modulation unit, 26...optical equalization unit, 30...optical receiver, 31...optical equalization unit, 32...optical detection unit, 33...analog-to-digital conversion unit, 34...chromatic dispersion compensation unit, 35...adaptive equalization unit, 36...soft decision unit, 37...decoding unit, 41...optical fiber, 42...optical amplifier, 51...branching unit, 52...bit log-likelihood ratio generation unit, 53...whitening filter unit, 54...symbol selection unit, 55...candidate sequence generation unit, 56...channel response reproduction unit, 57...posteriori maximum probability Estimation unit, 58...symbol-to-bit conversion unit, 59...likelihood ratio selection unit, 61...weighting unit, 62...weighting unit, 63...addition unit, 64...multiplication unit, 71...soft decision unit, 72...bit log-likelihood ratio generation unit, 73...multiplication unit, 81...soft decision unit, 82...whitening filter unit, 83...candidate sequence generation unit, 84...channel response reproduction unit, 85...maximum a posteriori probability estimation unit, 86...multiplication unit, 91...symbol decision unit, 92...whitening filter unit, 93...symbol selection unit, 94...candidate sequence generation unit, 95...channel response reproduction unit, 96...maximum a posteriori probability estimation unit, 101...processor, 102...storage unit, 103...communication interface, 104...user interface

Claims

1. a bit log likelihood ratio generator that generates a bit log likelihood ratio of a received symbol for each bit; a symbol selection unit that selects a plurality of adjacent symbols of the branched received symbols; a candidate sequence generation unit that generates a plurality of candidate sequences by combining the adjacent symbols selected by the symbol selection unit for each of the received symbols in the time series; a channel response reproducing unit that reflects a channel response in the candidate sequence; a maximum a posteriori probability estimation unit that calculates the log-likelihood ratio of the adjacent symbols so as to maximize the a posteriori probability, using an output obtained by whitening the time-series data of the branched received symbols and the candidate sequence reflecting the channel response; a symbol-to-bit converter that converts the adjacent symbols selected by the symbol selector into bits; a likelihood ratio selection unit that selects a bit log likelihood to be corrected from the bit log likelihood ratios generated by the bit log likelihood ratio generation unit based on the bits converted by the symbol-to-bit conversion unit; an adder that performs weighted addition of the bit log-likelihood ratio selected by the likelihood ratio selector and a correction value based on a difference between the bit log-likelihood ratio and the log-likelihood ratio calculated by the maximum a posteriori probability estimator; A symbol decision device comprising:

2. the bit log likelihood ratio generator generates a bit log likelihood ratio of the received symbol without using sequence estimation. The symbol determination device according to claim 1 .

3. the symbol selection unit selects the nearest symbol and the second nearest symbol of the branched received symbol as the nearest symbols; the likelihood ratio selection unit, when the bit obtained by converting the nearest-neighbor symbol and the bit obtained by converting the second nearest-neighbor symbol are different, selects the bit log-likelihood ratio corresponding to the bit as a target for correction.

3. The symbol determination device according to claim 1.

4. the symbol selection unit selects three or more of the adjacent symbols of the branched received symbols; the a posteriori probability maximum estimation unit calculates likelihoods of the adjacent symbols so as to maximize the a posteriori probability, using an output obtained by whitening the time-series data of the branched received symbols and the candidate sequence reflecting a channel response, and calculates a log-likelihood ratio of the two adjacent symbols selected based on the calculated likelihoods; the symbol-to-bit converter converts the two adjacent symbols selected by the maximum a posteriori probability estimator into bits.

3. The symbol determination device according to claim 1.

5. a bit log likelihood ratio generating step of generating a bit log likelihood ratio of a received symbol for each bit; a symbol selection step of selecting a plurality of adjacent symbols of the branched received symbols; a candidate sequence generation step of generating a plurality of candidate sequences by combining the adjacent symbols selected in the symbol selection step for each of the received symbols in the time series; a channel response reproduction step of reflecting a channel response in the candidate sequence; a posterior probability maximum estimation step of calculating a log-likelihood ratio of the adjacent symbols so as to maximize the posterior probability, using an output obtained by whitening the time series data of the branched received symbols and the candidate sequence reflecting the channel response; a symbol-to-bit conversion step of converting the adjacent symbols selected in the symbol selection step into bits; a likelihood ratio selection step of selecting a bit log likelihood to be corrected from the bit log likelihood ratios generated in the bit log likelihood ratio generation step, based on the bits converted in the symbol-to-bit conversion step; an addition step of weighting and adding the bit log-likelihood ratio selected in the likelihood ratio selection step and a correction value based on a difference between the bit log-likelihood ratio and the log-likelihood ratio calculated in the maximum a posteriori probability estimation step; A symbol determination method having the following structure:

6. Computer, A program for causing the symbol determination device according to any one of claims 1 to 4 to function.

Images