WO2023119487A1 - Optical transceiving device, optical communication method, and computer program - Google Patents

Optical transceiving device, optical communication method, and computer program Download PDF

Info

Publication number
WO2023119487A1
WO2023119487A1 PCT/JP2021/047580 JP2021047580W WO2023119487A1 WO 2023119487 A1 WO2023119487 A1 WO 2023119487A1 JP 2021047580 W JP2021047580 W JP 2021047580W WO 2023119487 A1 WO2023119487 A1 WO 2023119487A1
Authority
WO
WIPO (PCT)
Prior art keywords
area
layer
determination
circuit
area selection
Prior art date
Application number
PCT/JP2021/047580
Other languages
French (fr)
Japanese (ja)
Inventor
暁彦 松浦
昭一郎 桑原
由明 木坂
Original Assignee
日本電信電話株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 日本電信電話株式会社 filed Critical 日本電信電話株式会社
Priority to PCT/JP2021/047580 priority Critical patent/WO2023119487A1/en
Publication of WO2023119487A1 publication Critical patent/WO2023119487A1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
    • H04B10/60—Receivers
    • H04B10/61—Coherent receivers

Definitions

  • the present invention relates to the technology of an optical transmitter/receiver, an optical communication method, and a computer program.
  • Non-Patent Document 1 Digital coherent technology is an application of elemental technology of wireless communication to optical communication. Digital coherent technology is affected by various types of dispersion and nonlinear characteristics of optical fibers due to the unique condition of optical communication, which is the use of optical fibers for transmission lines. Therefore, there is a problem that the received waveform is distorted.
  • Non-Patent Document 2 and Non-Patent Document 3 For such problems, methods of applying machine learning to compensate for distortion are being studied (for example, Non-Patent Document 2 and Non-Patent Document 3). Multi-class classification is required for reception determination of multilevel signals such as 16QAM and 64QAM. Therefore, as the number of multilevel values increases, the internal variables of the machine learning circuit increase, and the calculation time tends to increase.
  • Non-Patent Document 2 and Non-Patent Document 3 use a support vector machine (hereinafter referred to as “SVM”) for machine learning to reduce the number of SVM circuits used.
  • SVM support vector machine
  • FIG. 12 is a diagram showing a conventional circuit configuration example using an SVM circuit.
  • FIG. 13 is a diagram showing an example of a signal point array in 16QAM. 12 and 13 are drawings cited from Non-Patent Document 2.
  • Signal points 0 to 15 can be represented by 4 bits of 0000 to 1111 in binary.
  • the upper 2 bits indicate the position in the x-axis direction
  • the lower 2 bits indicate the position in the y-axis direction.
  • an object of the present invention is to provide a technique capable of realizing symbol determination of a multilevel signal for a received optical signal in a shorter calculation time.
  • One aspect of the present invention includes a received symbol determination unit that performs symbol determination of a multilevel signal on a received optical signal and restores data based on the result of the symbol determination, wherein the received symbol determination unit includes a first layer to the M-th layer (M is an integer of 1 or more) and one layer of region determination layers, and the m-th layer (m is an integer of 1 or more and M or less) is provided.
  • a region to which the symbol of the received multilevel signal belongs is selected based on the results of machine learning, and the following is performed according to the selected region: layer, the region selection circuit of the m-th layer performs processing according to the region selected by the region selection circuit of the m-1th layer, and the region determination layer is the M-th layer an area determination circuit corresponding to the total number of outputs of the area selection circuits in the area selection layer of M, wherein the area determination circuit determines the area to which the symbol of the multilevel signal received from the M-th area selection layer belongs
  • This is an optical transmitter/receiver that makes a selection based on a result of machine learning and outputs a decision result indicating the symbol of the multilevel signal according to the selected region.
  • One aspect of the present invention includes a received symbol determination unit that performs symbol determination of a multilevel signal on a received optical signal and restores data based on the result of the symbol determination, wherein the received symbol determination unit includes a first layer to the M-th layer (M is an integer of 1 or more) and one layer of region determination layers, and the m-th layer (m is an integer of 1 or more and M or less) is provided.
  • an m-th layer area selection circuit selects a plurality of areas according to the number of layers (1 to M) for the received multilevel signal. a step of selecting a region to which the symbol of the received multilevel signal belongs from among the above based on the result of machine learning and outputting to the next layer according to the selected region; A region to which the symbol of the multi-level signal received from the region selection layer of the M layer belongs is selected based on the result of machine learning, and a determination result indicating the symbol of the multi-level signal is output according to the selected region. and wherein the area selection circuit of the m-th layer performs processing by an area selection circuit corresponding to the area selected by the area selection circuit of the (m ⁇ 1)-th layer.
  • One aspect of the present invention is a computer program for causing a computer to function as the above optical transceiver.
  • FIG. 1 is a diagram showing a system configuration example of an optical communication system 100 of the present invention
  • FIG. 3 is a diagram showing a detailed configuration example of a transmission signal selection unit 121
  • FIG. 3 is a diagram showing a detailed configuration example of a received symbol determination section 212.
  • FIG. FIG. 11 is a diagram showing a detailed configuration example of a determination unit 2122;
  • FIG. 3 is a diagram showing a specific configuration example of an area selection circuit 41;
  • FIG. FIG. 4 is a diagram showing a configuration example of area determination circuits 51 and 412 using SVM;
  • FIG. 4 is a diagram showing a configuration example of area determination circuits 51 and 412 using NNs;
  • FIG. 3 is a diagram showing a configuration example in which three layers of two-branch area selection circuits 41 are combined to determine a 16QAM signal.
  • FIG. 3 is a diagram showing an outline of processing of a configuration for judging a 16QAM signal by combining three layers of 2-branch area selection circuits 41; It is a figure which shows the principle of SVM. It is a figure which shows a simulation result.
  • 1 is a diagram showing a conventional circuit configuration example using an SVM circuit; FIG. It is a figure which shows the example of a signal point arrangement
  • FIG. 1 is a diagram showing a system configuration example of an optical communication system 100 of the present invention.
  • the optical communication system 100 includes a transmitter 10 , a receiver 20 and a controller 30 .
  • the transmission device 10 includes a multiplexing device 11 and an optical modulation device 12 .
  • the optical modulation device 12 includes a transmission signal selection section 121 and an optical modulation section 122 .
  • Optical data signals transmitted from a plurality of terminal devices are input to the multiplexer 11 of the transmission device 10 .
  • the multiplexer 11 generates an optical multiplexed data signal by multiplexing a plurality of input optical data signals, and outputs the optical multiplexed data signal to the optical modulator 12 .
  • the transmission signal selector 121 of the optical modulator 12 outputs either one of the optical multiplexed data signal and the training signal to the optical modulator 122 under the control of the controller 30 .
  • the optical modulator 122 modulates and transmits the optical signal output from the transmission signal selector 121 .
  • the receiver 20 includes an optical demodulator 21 and a separator 22 .
  • the optical demodulator 21 has a dispersion compensator 211 and a received symbol determiner 212 .
  • the receiver 20 receives the optical signal transmitted from the transmitter 10 .
  • the dispersion compensation unit 211 performs dispersion compensation processing and modulation dimension separation processing such as XY polarization separation on the received optical signal.
  • the received symbol determination unit 212 performs symbol determination on the optical signal output from the dispersion compensation unit 211 and outputs the determination result to the demultiplexer 22 .
  • the received symbol determination unit 212 restores the received symbols to digital data.
  • Received symbol determination section 212 sends the restored data value to demultiplexer 22 paired with multiplexer 11 in transmitting apparatus 10 .
  • Received symbol determination section 212 uses the results of machine learning performed in advance when performing symbol determination.
  • the separating device 22 outputs a signal corresponding to the determination result to a route corresponding to the destination.
  • Control device 30 selects a mode in which received symbol determination section 212 performs machine learning (hereinafter referred to as “training mode”) or a mode in which communication is performed using the results of machine learning (hereinafter referred to as “data determination mode”), control about. For example, when operating the optical communication system 100 in the training mode, the control device 30 outputs a training control signal to the transmission signal selection section 121 and the reception symbol determination section 212 .
  • the training control signal includes information indicating communication in the training mode and information indicating what kind of symbol signal is to be transmitted as the training signal.
  • the control device 30 may be configured as a device different from the transmitting device 10 and the receiving device 20 as shown in FIG. It may be configured as a part.
  • FIG. 2 is a diagram showing a detailed configuration example of the transmission signal selection unit 121.
  • the transmission signal selection section 121 has a training control circuit 1211 and a selection switch 1212 .
  • Training control circuit 1211 receives a training control signal transmitted from control device 30 .
  • Training control circuit 1211 outputs a signal of a symbol indicated by the received training control signal (hereinafter referred to as “training signal”) to selection switch 1212 .
  • training signal a signal of a symbol indicated by the received training control signal
  • the training control circuit 1211 outputs a selection control signal indicating output of the training signal to the selection switch 1212 .
  • the selection switch 1212 inputs either one or both of the training signal and the data signal.
  • the selection switch 1212 has a function of switching between a training signal and a data signal for transmission.
  • the selection switch 1212 outputs either the training signal or the data signal indicated by the selection control signal.
  • FIG. 3 is a diagram showing a detailed configuration example of the received symbol determination unit 212.
  • the received symbol determination section 212 includes a training control circuit 2121 and a determination section 2122 .
  • Training control circuit 2121 receives a training control signal transmitted from control device 30 .
  • Training control circuit 2121 outputs a training control signal including information of the symbol indicated by the received training control signal to determination section 2122 .
  • determination section 2122 uses the data signal output from dispersion compensation section 211 and the information indicating the symbol pattern indicated by the training control signal as a teacher. Machine learning is performed using it as data. That is, the determination section 2122 performs learning processing on the assumption that the pattern of the data signal output from the dispersion compensation section 211 is the pattern indicated by the training signal. The determination unit 2122 learns a compensation parameter for nonlinear distortion in this learning process.
  • determination section 2122 performs symbol determination by performing nonlinear distortion compensation on the data signal output from dispersion compensation section 211 based on the results of machine learning. conduct. A determination unit 2122 outputs a restored data value based on the determination result.
  • FIG. 4 is a diagram showing a detailed configuration example of the determination unit 2122.
  • transmission signal selection section 121 notifies receiving apparatus 20 of the fact that a training signal is being transmitted and the pattern of the training signal being transmitted as a training control signal.
  • a training control circuit 2121 in the receiving device 20 generates a correct label according to the pattern of the training signal.
  • the determination unit 2122 includes a plurality of area selection layers 40, an area determination layer 50, and a multiplexing circuit 60.
  • N layers of the region selection layer 40 are provided.
  • N is an integer value of 2 or more.
  • Each area selection layer 40 comprises one or more area selection circuits 41 .
  • the area selection layer 40_1 of the first layer includes one area selection circuit 41 .
  • Each area selection layer 40 of the second to N-th layers has an area selection circuit 41 corresponding to the total number of outputs in the area selection circuit 41 of the area selection layer 40 in the preceding stage.
  • the area determination layer 50 includes a plurality of area determination circuits 51 .
  • a correct label is input to each area selection circuit 41 and each area determination circuit 51 as a training control signal.
  • the area selection circuit 41 sets internal variables of the machine learning circuit so as to classify the received symbols according to the correct labels.
  • the area selection circuit 41 outputs the received symbol to the corresponding area selection circuit 41 of the next area selection layer 40 according to the correct label.
  • the area selection circuit 41 sets the internal variables of the machine learning circuit and outputs the received symbols to the corresponding area selection circuit 41 of the next area selection layer.
  • the number of layers of the determination unit 2122 and the number of branches of each area selection circuit 41 are set in advance so that the determination result of each area determination circuit 51 in the area determination layer 50, which is the final layer, includes one type of symbol in each area. be done. By sequentially inputting data signals for training and learning all paths in this way, the internal node states of the machine learning circuits included in all the area selection circuits 41 and the area determination circuits 51 are learned and determined. do.
  • the reception symbol is received by the area selection circuit 41 of the first area selection layer 40_1, and area selection is performed according to the machine learning result.
  • the area selection circuit 41 outputs the received symbol to the corresponding area selection circuit 41 of the next area selection layer 40 (40_2).
  • the area selection circuit 41 outputs the received symbol to the corresponding area selection circuit 41 of the next area selection layer 40 (40_3) according to the machine learning result.
  • One type of symbol is determined in the output of the area determination layer 50, which is the final layer.
  • the determination result is input to the multiplexing circuit 60 only from one of the area determination circuits 51 of the area determination layer 50 via a connection.
  • the connection number indicates the region where the determination was made last.
  • the input value from that connection indicates the final decision result.
  • a data value represented by a received symbol is uniquely determined based on the connection number and the determination result.
  • Multiplexing circuit 60 outputs the data value as a recovered data value.
  • the area selection layer 40_1 of the first layer includes one area selection circuit 41 .
  • the area selection circuit 41 classifies the received symbols into one of L(1,1) areas by using machine learning results.
  • the area selection circuit 41 outputs the received symbol as it is only to the next layer (second layer) area selection circuit 41 in charge of the corresponding area.
  • L(m, n) is the number of output connections of the n-th area selection circuit forming the m-th layer. Further, when the total number of output connections from the m-th layer is LS(m), the following formula 1 holds.
  • the area selection layer 40_2 of the second layer includes L (1, 1) area selection circuits 41 .
  • L (1, 1) area selection circuits 41 only the area selection circuit 41 that receives the input of the received symbol from the first layer operates. For example, when there is an input only to the second layer n-th region selection circuit 41, that region selection circuit 41 selects one of L (2, n) regions by using machine learning results. classified as one.
  • the area selection circuit 41 outputs the received symbol as it is only to the next layer (third layer) area selection circuit 41 in charge of the corresponding area.
  • region selection proceeds while sequentially sending received symbols to the next layer, and the following equation 2 holds for the M-th layer.
  • the output of the M-th region selection layer 40_M is input to the region determination layer 50 .
  • the area determination layer 50 includes LS(M) area determination circuits 51 . Only the area determination circuit 51 that receives the input of the received symbol from the area selection layer 40_M of the M-th layer operates. When only the n-th region determination circuit 51 receives an input, the region determination circuit 51 classifies the received symbol into one of L(x, n) regions by using machine learning results, Only the determination result is output to the subsequent multiplexing circuit 60 .
  • the symbol representing the region determination layer 50 is “x”.
  • the multiplexing circuit 60 outputs a restored data value that is uniquely determined from the position (number) of the input region determination circuit 51 and its value.
  • the region selection circuit 41 and the region determination circuit 51 of each layer described above may be configured to classify received symbols into any one of L (m, n) regions by machine learning using a neural network. good.
  • the region selection circuit 41 and the region determination circuit 51 of each layer described above classify received symbols into one of L (m, n) regions by one-to-many machine learning using a support vector machine. It may be configured as Both m and n are integers of 1 or more.
  • FIG. 5 is a diagram showing a specific configuration example of the area selection circuit 41.
  • the area selection circuit 41 includes a 2-branching circuit 411 , an area determination circuit 412 and a 1 ⁇ ns switch 413 .
  • the two-branch circuit 411 splits the training signal or data signal into two, and outputs one to the area determination circuit 412 and the other to the 1 ⁇ ns switch 413 .
  • Area determination circuit 412 identifies whether the input signal is a training signal or a data signal based on the training control signal. If the input signal is a training signal, the correct label is also notified based on the training control signal.
  • the area determination circuit 412 performs learning by machine learning using a neural network (hereinafter referred to as "NN"), SVM, or the like. Further, the area determination circuit 412 determines which area selection circuit 41 or area determination circuit 51 in the subsequent layer (the area selection layer 40 or the area determination layer 50) to output the input signal to, according to the training control signal. Area determination circuit 412 outputs a control signal to 1 ⁇ ns switch 413 according to the determination.
  • NN neural network
  • the area determination circuit 412 outputs the input signal to the area selection circuit 41 or the area determination circuit 51 of the subsequent layer (the area selection layer 40 or the area determination layer 50) based on the learning result of machine learning performed in advance. determine whether Area determination circuit 412 outputs a control signal to 1 ⁇ ns switch 413 according to the determination. A 1 ⁇ ns switch 413 switches the output destination of the received symbol to one of 1 to ns according to the output of the area determination circuit 412 .
  • FIG. 6 is a diagram showing a configuration example of the area determination circuit 51 using SVM.
  • the area determination circuit 51 forms part of the area selection circuit 41 (area determination circuit 412 ) and is also used in the area determination layer 50 .
  • a training signal or a data signal is input to the area determination circuit 51 .
  • the area determination circuit 51 has an SVM circuit 511 .
  • the area determination circuit 51 may be configured as one SVM circuit 511 itself.
  • the area determination circuit 51 may be configured as a round-robin determination circuit of a plurality of SVM circuits 511 .
  • the training control signal specifies whether the input signal is a training signal or a data signal. If the input signal is a training signal, its correct label is also notified at the same time. In this case, the SVM circuit 511 uses the training signal and the correct label to perform machine learning by SVM. When the input signal is a data signal, a determination is made using the result of machine learning, and the determination result is output. If the area determination circuit 51 is part of the area selection circuit 41 (area determination circuit 412 ), the determination result is used as a control signal for the 1 ⁇ ns switch 413 .
  • FIG. 7 is a diagram showing a configuration example of the area determination circuit 51 using NN.
  • the area determination circuit 51 forms part of the area selection circuit 41 (area determination circuit 412 ) and is also used in the area determination layer 50 .
  • a training signal or a data signal is input to the area determination circuit 51 .
  • the area determination circuit 51 has an NN circuit 512 .
  • the area determination circuit 51 may be configured as one NN circuit 512 itself.
  • the training control signal specifies whether the input signal is a training signal or a data signal. If the input signal is a training signal, its correct label is also notified at the same time. In this case, the NN circuit 512 performs NN machine learning using the training signal and the correct label. When the input signal is a data signal, a determination is made using the result of machine learning, and the determination result is output. If the area determination circuit 51 is part of the area selection circuit 41 (area determination circuit 412 ), the determination result is used as a control signal for the 1 ⁇ ns switch 413 .
  • FIG. 8 is a diagram showing a configuration example in which three layers of two-branch area selection circuits 41 are combined to determine a 16QAM signal.
  • FIG. 9 is a diagram showing an outline of processing of a configuration for determining a 16QAM signal by combining three layers of two-branch area selection circuits 41.
  • the 16QAM constellation is divided into left and right regions, and machine learning training is performed.
  • the required number of machine learning circuits (the number of the area determination circuits 412 and the area determination circuits 51) is 15, which is more than the four required in the conventional method.
  • the overall number of internal variables is comparable.
  • the training (machine learning process) is completed and symbol determination is performed, for example, if the received symbol is determined to belong to the right region in the first layer (region selection layer 40_1), the subsequent calculation does not operate on the left region. Therefore, it is possible to reduce the amount of calculation and the calculation execution time as compared with the conventional method.
  • it is assumed that the division into two regions is repeated, but multiple divisions of two or more divisions may be used, and the number of divisions may differ for each layer. Different areas within the same hierarchy may have different numbers of divisions.
  • FIG. 10 is a diagram showing the principle of SVM. It is desired to draw a boundary line so that the received symbols indicated by ⁇ and the received symbols indicated by ⁇ can be discriminated with as low an apology rate as possible.
  • a margin area is set inside each of the circle and triangle areas from the boundary line. Then, only received symbols outside the margin area when viewed from each area are used for decision making. These received symbols are hereinafter referred to as support vectors. Received symbols inside the margin region are not used for decision making. Therefore, if the number of received symbols outside the margin area is reduced during training, the amount of computation during data determination can be reduced.
  • the number of matrix coefficients that can be ignored in determination increases, so the number of parameters decreases, and the amount of computation during data determination can be reduced.
  • the amount of computation during data determination can be reduced by deliberately deleting those internal variables that are close to 0.
  • FIG. 11 is a diagram showing simulation results.
  • the existing scheme in the figure is non-linear compensation according to the prior art.
  • Proposed method 1 performs nonlinear compensation by dividing each layer into two by the region dividing procedure described with reference to FIG.
  • Proposed method 2 performs non-linear compensation by four divisions per layer by a round-robin method using SVM.
  • Three patterns of 16QAM, 64QAM, and 256QAM were simulated, the total number of SVs used for calculation was counted until the determination of one symbol was completed, and the average value was calculated.
  • the Q value was also confirmed.
  • An RBF kernel is used as the SVM kernel function, and hyperparameters are set to optimal values that minimize the number of SVs for each simulation condition.
  • Nonlinear distortion is simulated by rotating each symbol point of the QAM signal by an average of 10 degrees in proportion to its strength. A decrease in the number of SVs and an increase in the Q value are observed under any conditions, and the effect of this proposal can be confirmed
  • Each circuit in this embodiment may be configured using a processor such as a CPU (Central Processing Unit) and a memory. All or part of each circuit in this embodiment may be implemented using hardware such as ASIC (Application Specific Integrated Circuit), PLD (Programmable Logic Device), FPGA (Field Programmable Gate Array), and the like.
  • the above program may be recorded on a computer-readable recording medium.
  • Computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, semiconductor storage devices (such as SSD: Solid State Drives), hard disks and semiconductor storage built into computer systems. It is a storage device such as a device.
  • the above program may be transmitted via telecommunication lines.
  • the present invention is applicable to optical communication.
  • DESCRIPTION OF SYMBOLS 100 Optical communication system 10... Transmission apparatus 11... Multiplex apparatus 12... Optical modulation apparatus 121... Transmission signal selection part 122... Optical modulation part 20... Reception apparatus 21... Optical demodulation apparatus 22... Separation apparatus , 211... dispersion compensation unit, 212... received symbol determination unit, 30... control device, 1211... training control circuit, 1212... selection switch, 2121... training control circuit, 2122... determination unit, 40... area selection layer, 41... area Selection circuit 50 Area determination layer 51 Area determination circuit 60 Multiplexing circuit 411 Two-branch circuit 412 Area determination circuit 413 1xns switch 511 SVM circuit 512 NN circuit

Landscapes

  • Physics & Mathematics (AREA)
  • Electromagnetism (AREA)
  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Digital Transmission Methods That Use Modulated Carrier Waves (AREA)

Abstract

In this optical transceiving device: a reception symbol assessment unit comprises 1st through Mth area selection layers and one area assessment layer; the mth area selection layer comprises the same number of area selection circuits as an area division number in the m-1st area selection layer; an mth area selection circuit, with regard to a received multivalue signal, selects an area to which a symbol of the received multivalue signal belongs from among a plurality of areas in accordance with the stage number of the layer and on the basis of the result of machine learning and outputs the received multivalue signal to the next layer in accordance with the selected area; in the mth area selection circuits, the area selection circuit corresponding to the area selected by the m-1st area selection circuit performs a process; the area assessment layer comprises area assessment circuits corresponding to the total output number of area selection circuits in the Mth area selection layer; and an area assessment circuit selects an area to which the symbol of the multivalue signal received from the Mth area selection layer belongs on the basis of the result of machine learning and outputs an assessment result indicating the symbol of the multivalue signal in accordance with the selected area.

Description

光送受信装置、光通信方法及びコンピュータープログラムOptical transceiver, optical communication method and computer program
　本発明は、光送受信装置、光通信方法及びコンピュータープログラムの技術に関する。 The present invention relates to the technology of an optical transmitter/receiver, an optical communication method, and a computer program.
　光通信分野において、デジタルコヒーレント技術を用いた大容量多値光通信が広く導入されつつある（例えば非特許文献１）。デジタルコヒーレント技術は、無線通信の要素技術を光通信に適用したものである。デジタルコヒーレント技術では、伝送路に光ファイバを使用するという光通信特有の条件により、光ファイバの持つ各種分散や非線形特性の影響を受ける。そのため、受信波形が歪むという問題がある。 In the field of optical communication, large-capacity multilevel optical communication using digital coherent technology is being widely introduced (for example, Non-Patent Document 1). Digital coherent technology is an application of elemental technology of wireless communication to optical communication. Digital coherent technology is affected by various types of dispersion and nonlinear characteristics of optical fibers due to the unique condition of optical communication, which is the use of optical fibers for transmission lines. Therefore, there is a problem that the received waveform is distorted.
　このような問題に対して、機械学習を適用し歪みを補償する方法が検討されている（例えば非特許文献２、非特許文献３）。１６ＱＡＭや６４ＱＡＭなどの多値信号を受信判定するには、多クラス分類が必要となる。そのため、多値数が増えるにしたがって機械学習回路の内部変数が増大し計算時間が長くなりやすい。非特許文献２及び非特許文献３では、機械学習にサポートベクトルマシン（以下「ＳＶＭ」という。）を用い、使用されるＳＶＭ回路の個数を少なくする方式が使用されている。 For such problems, methods of applying machine learning to compensate for distortion are being studied (for example, Non-Patent Document 2 and Non-Patent Document 3). Multi-class classification is required for reception determination of multilevel signals such as 16QAM and 64QAM. Therefore, as the number of multilevel values increases, the internal variables of the machine learning circuit increase, and the calculation time tends to increase. Non-Patent Document 2 and Non-Patent Document 3 use a support vector machine (hereinafter referred to as “SVM”) for machine learning to reduce the number of SVM circuits used.
　図１２は、ＳＶＭ回路を用いた従来の回路構成例を示す図である。図１３は、１６ＱＡＭにおける信号点配列の例を示す図である。なお、図１２及び図１３は、非特許文献２から引用した図面である。以下の説明では、信号点０～１５から成る１６ＱＡＭ信号を判定する構成を例に説明する。信号点０～１５を２進数で表現すると００００～１１１１の４ビットで表現できる。図１３の信号点配列の場合、上位２ビットがｘ軸方向、下位２ビットがｙ軸方向の位置を示す。 FIG. 12 is a diagram showing a conventional circuit configuration example using an SVM circuit. FIG. 13 is a diagram showing an example of a signal point array in 16QAM. 12 and 13 are drawings cited from Non-Patent Document 2. FIG. In the following description, a configuration for determining a 16QAM signal consisting of signal points 0 to 15 will be described as an example. Signal points 0 to 15 can be represented by 4 bits of 0000 to 1111 in binary. In the case of the signal point array of FIG. 13, the upper 2 bits indicate the position in the x-axis direction, and the lower 2 bits indicate the position in the y-axis direction.
　まず、上位２ビットについて考える。上位２ビットの中で、上位ビットが“０”か“１”かにより信号点を分けると、二つの領域に分けられる。そこで、これをＳＶＭで２分類する。次に、上位２ビットの中で下位ビットが“０”か“１”かにより信号点を分けると、同様に二つの領域に分けられる。そこで、これをＳＶＭで２分類する。ｙ軸についても同様の処理を行う。その結果、ｘ及びｙの両軸で合計４つのＳＶＭ分類器を使えば１６信号点が判定できる。 First, consider the upper 2 bits. If a signal point is divided according to whether the upper bit is "0" or "1" in the upper two bits, it is divided into two areas. Therefore, this is classified into two by SVM. Next, if the signal points are divided according to whether the lower bit in the upper two bits is "0" or "1", they are similarly divided into two areas. Therefore, this is classified into two by SVM. Similar processing is performed for the y-axis. As a result, 16 signal points can be determined using a total of 4 SVM classifiers in both the x and y axes.
　しかしながら、機械学習回路の数を減らすことができたとしても、個々の機械学習回路における計算量については考慮されていなかった。そのため、計算実行時間が増大するという問題があった。 However, even if the number of machine learning circuits could be reduced, the computational complexity of each individual machine learning circuit was not considered. Therefore, there is a problem that the calculation execution time increases.
　上記事情に鑑み、本発明は、受信された光信号について多値信号のシンボル判定をより少ない計算時間で実現することができる技術の提供を目的としている。 In view of the above circumstances, an object of the present invention is to provide a technique capable of realizing symbol determination of a multilevel signal for a received optical signal in a shorter calculation time.
　本発明の一態様は、受信された光信号について多値信号のシンボル判定を行い、前記シンボル判定の結果に基づいてデータを復元する受信シンボル判定部を備え、前記受信シンボル判定部は、第１層から第Ｍ層（Ｍは１以上の整数）の領域選択層と、１層の領域判定層と、を備え、第ｍ層（ｍは１以上Ｍ以下の整数）の領域選択層は、第ｍ－１層の領域選択層での領域分割数（ｍ＝１の場合は“１”）と同数の領域選択回路を備え、第ｍ層の領域選択回路は、受信された前記多値信号について、層の段数（１からＭ）に応じて複数の領域の中から、受信された前記多値信号のシンボルが属する領域を機械学習の結果に基づいて選択し、選択された領域に応じて次の層へ出力し、前記第ｍ層の領域選択回路は、前記第ｍ－１層の領域選択回路で選択された領域に応じた領域選択回路が処理を行い、前記領域判定層は第Ｍ層の領域選択層での領域選択回路の出力数の総数に応じた領域判定回路を備え、前記領域判定回路は、第Ｍ層の領域選択層から受信された前記多値信号のシンボルが属する領域を機械学習の結果に基づいて選択し、選択された領域に応じて前記多値信号のシンボルを示す判定結果を出力する、光送受信装置である。 One aspect of the present invention includes a received symbol determination unit that performs symbol determination of a multilevel signal on a received optical signal and restores data based on the result of the symbol determination, wherein the received symbol determination unit includes a first layer to the M-th layer (M is an integer of 1 or more) and one layer of region determination layers, and the m-th layer (m is an integer of 1 or more and M or less) is provided. The number of area selection circuits equal to the number of area divisions in the m−1 area selection layers (“1” when m=1) is provided, and the area selection circuit of the m-th layer selects the received multilevel signal. , from among a plurality of regions according to the number of layers (1 to M), a region to which the symbol of the received multilevel signal belongs is selected based on the results of machine learning, and the following is performed according to the selected region: layer, the region selection circuit of the m-th layer performs processing according to the region selected by the region selection circuit of the m-1th layer, and the region determination layer is the M-th layer an area determination circuit corresponding to the total number of outputs of the area selection circuits in the area selection layer of M, wherein the area determination circuit determines the area to which the symbol of the multilevel signal received from the M-th area selection layer belongs This is an optical transmitter/receiver that makes a selection based on a result of machine learning and outputs a decision result indicating the symbol of the multilevel signal according to the selected region.
　本発明の一態様は、受信された光信号について多値信号のシンボル判定を行い、前記シンボル判定の結果に基づいてデータを復元する受信シンボル判定部を備え、前記受信シンボル判定部は、第１層から第Ｍ層（Ｍは１以上の整数）の領域選択層と、１層の領域判定層と、を備え、第ｍ層（ｍは１以上Ｍ以下の整数）の領域選択層は、第ｍ－１層の領域選択層での領域分割数（ｍ＝１の場合は“１”）と同数の領域選択回路を備え、前記領域判定層は第Ｍ層の領域選択層での領域選択回路の出力数の総数に応じた領域判定回路を備える光通信装置において、第ｍ層の領域選択回路が、受信された前記多値信号について、層の段数（１からＭ）に応じて複数の領域の中から、受信された前記多値信号のシンボルが属する領域を機械学習の結果に基づいて選択し、選択された領域に応じて次の層へ出力するステップと、前記領域判定回路が、第Ｍ層の領域選択層から受信された前記多値信号のシンボルが属する領域を機械学習の結果に基づいて選択し、選択された領域に応じて前記多値信号のシンボルを示す判定結果を出力するステップと、を有し、前記第ｍ層の領域選択回路は、前記第ｍ－１層の領域選択回路で選択された領域に応じた領域選択回路が処理を行う、光通信方法である。 One aspect of the present invention includes a received symbol determination unit that performs symbol determination of a multilevel signal on a received optical signal and restores data based on the result of the symbol determination, wherein the received symbol determination unit includes a first layer to the M-th layer (M is an integer of 1 or more) and one layer of region determination layers, and the m-th layer (m is an integer of 1 or more and M or less) is provided. The number of area selection circuits equal to the number of area divisions in the m−1 area selection layer (“1” when m=1) is provided, and the area determination layer is the area selection circuit in the M-th area selection layer. In an optical communication device comprising an area determination circuit corresponding to the total number of outputs, an m-th layer area selection circuit selects a plurality of areas according to the number of layers (1 to M) for the received multilevel signal. a step of selecting a region to which the symbol of the received multilevel signal belongs from among the above based on the result of machine learning and outputting to the next layer according to the selected region; A region to which the symbol of the multi-level signal received from the region selection layer of the M layer belongs is selected based on the result of machine learning, and a determination result indicating the symbol of the multi-level signal is output according to the selected region. and wherein the area selection circuit of the m-th layer performs processing by an area selection circuit corresponding to the area selected by the area selection circuit of the (m−1)-th layer.
　本発明の一態様は、上記の光送受信装置としてコンピューターを機能させるためのコンピュータープログラムである。 One aspect of the present invention is a computer program for causing a computer to function as the above optical transceiver.
　本発明により、受信された光信号について多値信号のシンボル判定をより少ない計算時間で実現することが可能となる。 According to the present invention, it is possible to realize symbol determination of a multilevel signal for a received optical signal in a shorter calculation time.
本発明の光通信システム１００のシステム構成例を示す図である。1 is a diagram showing a system configuration example of an optical communication system 100 of the present invention; FIG. 送信信号選択部１２１の詳細な構成例を示す図である。3 is a diagram showing a detailed configuration example of a transmission signal selection unit 121; FIG. 受信シンボル判定部２１２の詳細な構成例を示す図である。3 is a diagram showing a detailed configuration example of a received symbol determination section 212. FIG. 判定部２１２２の詳細な構成例を示す図である。FIG. 11 is a diagram showing a detailed configuration example of a determination unit 2122; FIG. 領域選択回路４１の具体的な構成例を示す図である。3 is a diagram showing a specific configuration example of an area selection circuit 41; FIG. ＳＶＭを用いた領域判定回路５１および４１２の構成例を示す図である。FIG. 4 is a diagram showing a configuration example of area determination circuits 51 and 412 using SVM; ＮＮを用いた領域判定回路５１および４１２の構成例を示す図である。FIG. 4 is a diagram showing a configuration example of area determination circuits 51 and 412 using NNs; ２分岐の領域選択回路４１を３層組み合わせて１６ＱＡＭ信号を判定する構成例を示す図である。FIG. 3 is a diagram showing a configuration example in which three layers of two-branch area selection circuits 41 are combined to determine a 16QAM signal. ２分岐の領域選択回路４１を３層組み合わせて１６ＱＡＭ信号を判定する構成の処理の概略を示す図である。FIG. 3 is a diagram showing an outline of processing of a configuration for judging a 16QAM signal by combining three layers of 2-branch area selection circuits 41; ＳＶＭの原理を示す図である。It is a figure which shows the principle of SVM. シミュレーション結果を示す図である。It is a figure which shows a simulation result. ＳＶＭ回路を用いた従来の回路構成例を示す図である。1 is a diagram showing a conventional circuit configuration example using an SVM circuit; FIG. １６ＱＡＭにおける信号点配列の例を示す図である。It is a figure which shows the example of a signal point arrangement|sequence in 16QAM.
　本発明の実施形態について、図面を参照して詳細に説明する。
　図１は、本発明の光通信システム１００のシステム構成例を示す図である。光通信システム１００は、送信装置１０、受信装置２０及び制御装置３０を備える。 Embodiments of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a diagram showing a system configuration example of an optical communication system 100 of the present invention. The optical communication system 100 includes a transmitter 10 , a receiver 20 and a controller 30 .
　送信装置１０は、多重装置１１及び光変調装置１２を備える。光変調装置１２は、送信信号選択部１２１及び光変調部１２２を備える。送信装置１０の多重装置１１には、不図示の複数の端末装置から送信された光データ信号が入力される。多重装置１１は、入力された複数の光データ信号を多重することで光多重データ信号を生成し、光変調装置１２に出力する。光変調装置１２の送信信号選択部１２１は、制御装置３０の制御に応じて、光多重データ信号及びトレーニング信号のいずれか一方を光変調部１２２に出力する。光変調部１２２は、送信信号選択部１２１から出力された光信号を変調して送信する。 The transmission device 10 includes a multiplexing device 11 and an optical modulation device 12 . The optical modulation device 12 includes a transmission signal selection section 121 and an optical modulation section 122 . Optical data signals transmitted from a plurality of terminal devices (not shown) are input to the multiplexer 11 of the transmission device 10 . The multiplexer 11 generates an optical multiplexed data signal by multiplexing a plurality of input optical data signals, and outputs the optical multiplexed data signal to the optical modulator 12 . The transmission signal selector 121 of the optical modulator 12 outputs either one of the optical multiplexed data signal and the training signal to the optical modulator 122 under the control of the controller 30 . The optical modulator 122 modulates and transmits the optical signal output from the transmission signal selector 121 .
　受信装置２０は、光復調装置２１及び分離装置２２を備える。光復調装置２１は、分散補償部２１１及び受信シンボル判定部２１２を備える。受信装置２０は、送信装置１０から送信された光信号を受信する。分散補償部２１１は、受信された光信号について分散補償処理やＸ－Ｙ偏波分離などの変調の次元分離処理を行う。受信シンボル判定部２１２は、分散補償部２１１から出力された光信号についてシンボル判定を行い、判定結果を分離装置２２に出力する。受信シンボル判定部２１２は、受信シンボルをデジタルデータに復元する。受信シンボル判定部２１２は、復元データ値を、対向する送信装置１０内の多重装置１１と対をなす分離装置２２に送出する。受信シンボル判定部２１２は、シンボル判定を行うに際して、予め実施された機械学習の結果を用いる。分離装置２２は、判定結果に応じた信号を宛先に応じた方路へ出力する。 The receiver 20 includes an optical demodulator 21 and a separator 22 . The optical demodulator 21 has a dispersion compensator 211 and a received symbol determiner 212 . The receiver 20 receives the optical signal transmitted from the transmitter 10 . The dispersion compensation unit 211 performs dispersion compensation processing and modulation dimension separation processing such as XY polarization separation on the received optical signal. The received symbol determination unit 212 performs symbol determination on the optical signal output from the dispersion compensation unit 211 and outputs the determination result to the demultiplexer 22 . The received symbol determination unit 212 restores the received symbols to digital data. Received symbol determination section 212 sends the restored data value to demultiplexer 22 paired with multiplexer 11 in transmitting apparatus 10 . Received symbol determination section 212 uses the results of machine learning performed in advance when performing symbol determination. The separating device 22 outputs a signal corresponding to the determination result to a route corresponding to the destination.
　制御装置３０は、受信シンボル判定部２１２において機械学習を行うモード（以下「トレーニングモード」という。）か、機械学習の結果を用いて通信を行うモード（以下「データ判定モード」という。）か、について制御する。制御装置３０は、例えばトレーニングモードとして光通信システム１００を動作させる場合には、トレーニング制御信号を送信信号選択部１２１及び受信シンボル判定部２１２に出力する。トレーニング制御信号には、トレーニングモードとして通信することを示す情報と、トレーニング信号としてどのようなシンボルの信号が送信されるかを示す情報と、が含まれる。制御装置３０は、図１に示されるように送信装置１０及び受信装置２０とは異なる装置として構成されてもよいし、送信装置１０の一部として構成されてもよいし、受信装置２０の一部として構成されてもよい。 Control device 30 selects a mode in which received symbol determination section 212 performs machine learning (hereinafter referred to as “training mode”) or a mode in which communication is performed using the results of machine learning (hereinafter referred to as “data determination mode”), control about. For example, when operating the optical communication system 100 in the training mode, the control device 30 outputs a training control signal to the transmission signal selection section 121 and the reception symbol determination section 212 . The training control signal includes information indicating communication in the training mode and information indicating what kind of symbol signal is to be transmitted as the training signal. The control device 30 may be configured as a device different from the transmitting device 10 and the receiving device 20 as shown in FIG. It may be configured as a part.
　図２は、送信信号選択部１２１の詳細な構成例を示す図である。送信信号選択部１２１は、トレーニング制御回路１２１１及び選択スイッチ１２１２を備える。トレーニング制御回路１２１１は、制御装置３０から送信されたトレーニング制御信号を受信する。トレーニング制御回路１２１１は、受信されたトレーニング制御信号が示すシンボルの信号（以下「トレーニング信号」という。）を選択スイッチ１２１２に出力する。また、トレーニング制御回路１２１１は、トレーニング制御信号を受信した際には、トレーニング信号を出力することを示す選択制御信号を選択スイッチ１２１２に出力する。 FIG. 2 is a diagram showing a detailed configuration example of the transmission signal selection unit 121. As shown in FIG. The transmission signal selection section 121 has a training control circuit 1211 and a selection switch 1212 . Training control circuit 1211 receives a training control signal transmitted from control device 30 . Training control circuit 1211 outputs a signal of a symbol indicated by the received training control signal (hereinafter referred to as “training signal”) to selection switch 1212 . Also, upon receiving the training control signal, the training control circuit 1211 outputs a selection control signal indicating output of the training signal to the selection switch 1212 .
　選択スイッチ１２１２は、トレーニング信号及びデータ信号のいずれか一つ又は両方を入力する。選択スイッチ１２１２は、トレーニング信号とデータ信号とを切り替えて送出する機能を持つ。選択スイッチ１２１２は、トレーニング信号とデータ信号とのうち、選択制御信号が示す方の信号を送出する。 The selection switch 1212 inputs either one or both of the training signal and the data signal. The selection switch 1212 has a function of switching between a training signal and a data signal for transmission. The selection switch 1212 outputs either the training signal or the data signal indicated by the selection control signal.
　図３は、受信シンボル判定部２１２の詳細な構成例を示す図である。受信シンボル判定部２１２は、トレーニング制御回路２１２１及び判定部２１２２を備える。トレーニング制御回路２１２１は、制御装置３０から送信されたトレーニング制御信号を受信する。トレーニング制御回路２１２１は、受信されたトレーニング制御信号が示すシンボルの情報を含むトレーニング制御信号を判定部２１２２に出力する。 FIG. 3 is a diagram showing a detailed configuration example of the received symbol determination unit 212. As shown in FIG. The received symbol determination section 212 includes a training control circuit 2121 and a determination section 2122 . Training control circuit 2121 receives a training control signal transmitted from control device 30 . Training control circuit 2121 outputs a training control signal including information of the symbol indicated by the received training control signal to determination section 2122 .
　判定部２１２２は、トレーニング制御信号が受信された場合（トレーニングモードである場合）には、分散補償部２１１から出力されたデータ信号と、トレーニング制御信号が示すシンボルのパターンを示す情報と、を教師データとして用いて機械学習を行う。すなわち、判定部２１２２は、分散補償部２１１から出力されたデータ信号のパターンが、トレーニング信号が示すパターンであるという前提で学習処理を行う。判定部２１２２は、この学習処理において、非線形歪みの補償パラメータを学習する。判定部２１２２は、トレーニング制御信号が受信されない場合（データ判定モードである場合）には、分散補償部２１１から出力されたデータ信号について機械学習の結果に基づいて非線形歪み補償を行ってシンボル判定を行う。判定部２１２２は、判定結果に基づいて復元データ値を出力する。 When the training control signal is received (in training mode), determination section 2122 uses the data signal output from dispersion compensation section 211 and the information indicating the symbol pattern indicated by the training control signal as a teacher. Machine learning is performed using it as data. That is, the determination section 2122 performs learning processing on the assumption that the pattern of the data signal output from the dispersion compensation section 211 is the pattern indicated by the training signal. The determination unit 2122 learns a compensation parameter for nonlinear distortion in this learning process. When the training control signal is not received (in the data determination mode), determination section 2122 performs symbol determination by performing nonlinear distortion compensation on the data signal output from dispersion compensation section 211 based on the results of machine learning. conduct. A determination unit 2122 outputs a restored data value based on the determination result.
　図４は、判定部２１２２の詳細な構成例を示す図である。機械学習のトレーニングモードでは、送信信号選択部１２１からトレーニング信号を送信中であることと、送信中のトレーニング信号のパターンと、が受信装置２０にトレーニング制御信号として通知される。受信装置２０内のトレーニング制御回路２１２１は、トレーニング信号のパターンにしたがって正解ラベルを生成する。 FIG. 4 is a diagram showing a detailed configuration example of the determination unit 2122. As shown in FIG. In the machine learning training mode, transmission signal selection section 121 notifies receiving apparatus 20 of the fact that a training signal is being transmitted and the pattern of the training signal being transmitted as a training control signal. A training control circuit 2121 in the receiving device 20 generates a correct label according to the pattern of the training signal.
　判定部２１２２は、複数の領域選択層４０と、領域判定層５０と、多重化回路６０と、を備える。図４の例では、領域選択層４０はＮ層設けられている。ここで、Ｎは２以上の整数値である。各領域選択層４０は、１又は複数の領域選択回路４１を備える。第１層の領域選択層４０＿１は、１つの領域選択回路４１を備える。第２層～第Ｎ層の各領域選択層４０は、各一つ前段の領域選択層４０の領域選択回路４１における出力数の総数に応じた領域選択回路４１を備える。領域判定層５０は、複数の領域判定回路５１を備える。 The determination unit 2122 includes a plurality of area selection layers 40, an area determination layer 50, and a multiplexing circuit 60. In the example of FIG. 4, N layers of the region selection layer 40 are provided. Here, N is an integer value of 2 or more. Each area selection layer 40 comprises one or more area selection circuits 41 . The area selection layer 40_1 of the first layer includes one area selection circuit 41 . Each area selection layer 40 of the second to N-th layers has an area selection circuit 41 corresponding to the total number of outputs in the area selection circuit 41 of the area selection layer 40 in the preceding stage. The area determination layer 50 includes a plurality of area determination circuits 51 .
　各領域選択回路４１と各領域判定回路５１には、トレーニング制御信号として正解ラベルが入力される。領域選択回路４１は、受信シンボルを正解ラベルに従って分類するように機械学習回路の内部変数を設定する。領域選択回路４１は、正解ラベルに従って次の領域選択層４０の該当する領域選択回路４１に受信シンボルを出力する。次の領域選択層４０においても同様に、領域選択回路４１は、機械学習回路の内部変数を設定し、次の領域選択層の該当する領域選択回路４１に受信シンボルを出力する。 A correct label is input to each area selection circuit 41 and each area determination circuit 51 as a training control signal. The area selection circuit 41 sets internal variables of the machine learning circuit so as to classify the received symbols according to the correct labels. The area selection circuit 41 outputs the received symbol to the corresponding area selection circuit 41 of the next area selection layer 40 according to the correct label. Similarly, in the next area selection layer 40, the area selection circuit 41 sets the internal variables of the machine learning circuit and outputs the received symbols to the corresponding area selection circuit 41 of the next area selection layer.
　判定部２１２２の階層数と各領域選択回路４１の分岐数は、最終層にあたる領域判定層５０において各領域判定回路５１の判定の結果、各領域に含まれるシンボルが１種類となるように予め設定される。このようにトレーニング用のデータ信号を次々と入力して、すべてのパスの学習を行うことで、全ての領域選択回路４１及び領域判定回路５１が備える機械学習回路の内部のノード状態を学習し決定する。 The number of layers of the determination unit 2122 and the number of branches of each area selection circuit 41 are set in advance so that the determination result of each area determination circuit 51 in the area determination layer 50, which is the final layer, includes one type of symbol in each area. be done. By sequentially inputting data signals for training and learning all paths in this way, the internal node states of the machine learning circuits included in all the area selection circuits 41 and the area determination circuits 51 are learned and determined. do.
　データ判定モードでは、最初の領域選択層４０＿１の領域選択回路４１において受信シンボルを受信し、機械学習結果に従って領域選択を行う。領域選択回路４１は、次の領域選択層４０（４０＿２）の該当する領域選択回路４１に受信シンボルを出力する。領域選択層４０＿２においても同様に、領域選択回路４１は機械学習結果に従って次の領域選択層４０（４０＿３）の該当する領域選択回路４１に受信シンボルを出力する。最終層にあたる領域判定層５０の出力では、シンボルは１種類に定まる。多重化回路６０には、領域判定層５０のいずれか一つの領域判定回路５１からのみ判定結果が結線を介して入力される。結線の番号は最後に判定を行った領域を示す。その結線からの入力値は最後の判定結果を示す。結線の番号と判定結果とに基づいて、受信シンボルが表すデータ値が一意に定まる。多重化回路６０は、そのデータ値を復元データ値として出力する。 In the data judgment mode, the reception symbol is received by the area selection circuit 41 of the first area selection layer 40_1, and area selection is performed according to the machine learning result. The area selection circuit 41 outputs the received symbol to the corresponding area selection circuit 41 of the next area selection layer 40 (40_2). Similarly, in the area selection layer 40_2, the area selection circuit 41 outputs the received symbol to the corresponding area selection circuit 41 of the next area selection layer 40 (40_3) according to the machine learning result. One type of symbol is determined in the output of the area determination layer 50, which is the final layer. The determination result is input to the multiplexing circuit 60 only from one of the area determination circuits 51 of the area determination layer 50 via a connection. The connection number indicates the region where the determination was made last. The input value from that connection indicates the final decision result. A data value represented by a received symbol is uniquely determined based on the connection number and the determination result. Multiplexing circuit 60 outputs the data value as a recovered data value.
　次に、各層における処理についてさらに説明する。
　第１層の領域選択層４０＿１は、１個の領域選択回路４１を備える。データ判定モードにおいては、領域選択回路４１は、機械学習結果を用いることで受信シンボルをＬ（１，１）個の領域のうちいずれか一つに分類する。領域選択回路４１は、該当する領域を担当する次層（第２層）の領域選択回路４１のみに受信シンボルをそのまま出力する。ここでＬ（ｍ，ｎ）は第ｍ層を構成する第ｎ個目の領域選択回路の出力接続数とする。また第ｍ層からの総出力接続数をＬＳ（ｍ）とすると、以下の式１が成り立つ。 Next, processing in each layer will be further described.
The area selection layer 40_1 of the first layer includes one area selection circuit 41 . In the data determination mode, the area selection circuit 41 classifies the received symbols into one of L(1,1) areas by using machine learning results. The area selection circuit 41 outputs the received symbol as it is only to the next layer (second layer) area selection circuit 41 in charge of the corresponding area. Here, L(m, n) is the number of output connections of the n-th area selection circuit forming the m-th layer. Further, when the total number of output connections from the m-th layer is LS(m), the following formula 1 holds.
Figure JPOXMLDOC01-appb-M000001
Figure JPOXMLDOC01-appb-M000001
　第２層の領域選択層４０＿２は、Ｌ（１，１）個の領域選択回路４１を備える。これらの複数の領域選択回路４１のうち、第１層から受信シンボルの入力があった領域選択回路４１のみが動作する。例えば、第２層ｎ番目の領域選択回路４１のみに入力があった場合、その領域選択回路４１は、機械学習結果を用いることで受信シンボルをＬ（２，ｎ）個の領域のうちいずれか一つに分類する。領域選択回路４１は、該当する領域を担当する次層（第３層）の領域選択回路４１にのみ受信シンボルをそのまま出力する。 The area selection layer 40_2 of the second layer includes L (1, 1) area selection circuits 41 . Among these plurality of area selection circuits 41, only the area selection circuit 41 that receives the input of the received symbol from the first layer operates. For example, when there is an input only to the second layer n-th region selection circuit 41, that region selection circuit 41 selects one of L (2, n) regions by using machine learning results. classified as one. The area selection circuit 41 outputs the received symbol as it is only to the next layer (third layer) area selection circuit 41 in charge of the corresponding area.
　以下同様に、受信シンボルを順次次の層に送りながら領域選択が進められ、第Ｍ層では以下の式２が成り立つ。 In the same way, region selection proceeds while sequentially sending received symbols to the next layer, and the following equation 2 holds for the M-th layer.
Figure JPOXMLDOC01-appb-M000002
Figure JPOXMLDOC01-appb-M000002
　第Ｍ層の領域選択層４０＿Ｍの出力は、領域判定層５０に入力される。領域判定層５０はＬＳ（Ｍ）個の領域判定回路５１を備える。第Ｍ層の領域選択層４０＿Ｍから受信シンボルの入力があった領域判定回路５１のみが動作する。ｎ番目の領域判定回路５１のみに入力があった場合、領域判定回路５１は、機械学習結果を用いることで受信シンボルをＬ（ｘ，ｎ）個の領域のうちいずれか一つに分類し、判定結果のみを後段の多重化回路６０に出力する。ここで領域判定層５０を表す記号を“ｘ”とした。 The output of the M-th region selection layer 40_M is input to the region determination layer 50 . The area determination layer 50 includes LS(M) area determination circuits 51 . Only the area determination circuit 51 that receives the input of the received symbol from the area selection layer 40_M of the M-th layer operates. When only the n-th region determination circuit 51 receives an input, the region determination circuit 51 classifies the received symbol into one of L(x, n) regions by using machine learning results, Only the determination result is output to the subsequent multiplexing circuit 60 . Here, the symbol representing the region determination layer 50 is “x”.
　多重化回路６０は、入力のあった領域判定回路５１の位置（番号）とその値から、一意に定まる復元データ値を出力する。 The multiplexing circuit 60 outputs a restored data value that is uniquely determined from the position (number) of the input region determination circuit 51 and its value.
　上述した各層の領域選択回路４１及び領域判定回路５１は、ニューラルネットワークを用いた機械学習により受信シンボルをＬ（ｍ，ｎ）個の領域のうちいずれか一つに分類するように構成されてもよい。上述した各層の領域選択回路４１及び領域判定回路５１は、サポートベクトルマシンを用いた１対他方式の機械学習により受信シンボルをＬ（ｍ，ｎ）個の領域のうちいずれか一つに分類するように構成されてもよい。なお、ｍ及びｎはいずれも１以上の整数を示す。 The region selection circuit 41 and the region determination circuit 51 of each layer described above may be configured to classify received symbols into any one of L (m, n) regions by machine learning using a neural network. good. The region selection circuit 41 and the region determination circuit 51 of each layer described above classify received symbols into one of L (m, n) regions by one-to-many machine learning using a support vector machine. It may be configured as Both m and n are integers of 1 or more.
　図５は、領域選択回路４１の具体的な構成例を示す図である。領域選択回路４１は、２分岐回路４１１、領域判定回路４１２及び１ｘｎｓスイッチ４１３を備える。２分岐回路４１１は、トレーニング信号又はデータ信号を２分岐し、一方を領域判定回路４１２へ、他方を１ｘｎｓスイッチ４１３へ出力する。領域判定回路４１２は機械学習回路である。ｎｓ＝Ｌ（ｍ，ｎ）である。 FIG. 5 is a diagram showing a specific configuration example of the area selection circuit 41. As shown in FIG. The area selection circuit 41 includes a 2-branching circuit 411 , an area determination circuit 412 and a 1×ns switch 413 . The two-branch circuit 411 splits the training signal or data signal into two, and outputs one to the area determination circuit 412 and the other to the 1×ns switch 413 . The area determination circuit 412 is a machine learning circuit. ns=L(m,n).
　まずトレーニングモードにおける動作について説明する。領域判定回路４１２は、トレーニング制御信号に基づいて、入力信号がトレーニング信号であるかデータ信号であるかを識別する。入力信号がトレーニング信号である場合、トレーニング制御信号に基づいて正解ラベルも通知される。領域判定回路４１２は、ニューラルネットワーク（以下「ＮＮ」という。）やＳＶＭ等を用いた機械学習による学習を行う。また、領域判定回路４１２は、トレーニング制御信号に従って、入力信号を後段の層（領域選択層４０又は領域判定層５０）のどの領域選択回路４１又は領域判定回路５１へ出力するかを決定する。領域判定回路４１２は、決定にしたがって、１ｘｎｓスイッチ４１３へ制御信号を出力する。 First, we will explain the operation in training mode. Area determination circuit 412 identifies whether the input signal is a training signal or a data signal based on the training control signal. If the input signal is a training signal, the correct label is also notified based on the training control signal. The area determination circuit 412 performs learning by machine learning using a neural network (hereinafter referred to as "NN"), SVM, or the like. Further, the area determination circuit 412 determines which area selection circuit 41 or area determination circuit 51 in the subsequent layer (the area selection layer 40 or the area determination layer 50) to output the input signal to, according to the training control signal. Area determination circuit 412 outputs a control signal to 1×ns switch 413 according to the determination.
　次にデータ判定モードにおける動作について説明する。領域判定回路４１２は、予め行われた機械学習の学習結果に基づいて、入力信号を後段の層（領域選択層４０又は領域判定層５０）のどの領域選択回路４１又は領域判定回路５１へ出力するかを決定する。領域判定回路４１２は、決定にしたがって、１ｘｎｓスイッチ４１３へ制御信号を出力する。１ｘｎｓスイッチ４１３は、領域判定回路４１２の出力に応じて、受信シンボルの出力先を１～ｎｓのうちいずれか一つに切り替える。 Next, the operation in the data judgment mode will be explained. The area determination circuit 412 outputs the input signal to the area selection circuit 41 or the area determination circuit 51 of the subsequent layer (the area selection layer 40 or the area determination layer 50) based on the learning result of machine learning performed in advance. determine whether Area determination circuit 412 outputs a control signal to 1×ns switch 413 according to the determination. A 1×ns switch 413 switches the output destination of the received symbol to one of 1 to ns according to the output of the area determination circuit 412 .
　図６は、ＳＶＭを用いた領域判定回路５１の構成例を示す図である。領域判定回路５１は、領域選択回路４１の一部（領域判定回路４１２）を構成するものであるとともに、領域判定層５０でも使用される。領域判定回路５１には、トレーニング信号又はデータ信号が入力される。領域判定回路５１はＳＶＭ回路５１１を備える。領域判定回路５１は、１個のＳＶＭ回路５１１そのものとして構成されてもよい。領域判定回路５１は、複数個のＳＶＭ回路５１１の総当たり判定回路として構成されてもよい。 FIG. 6 is a diagram showing a configuration example of the area determination circuit 51 using SVM. The area determination circuit 51 forms part of the area selection circuit 41 (area determination circuit 412 ) and is also used in the area determination layer 50 . A training signal or a data signal is input to the area determination circuit 51 . The area determination circuit 51 has an SVM circuit 511 . The area determination circuit 51 may be configured as one SVM circuit 511 itself. The area determination circuit 51 may be configured as a round-robin determination circuit of a plurality of SVM circuits 511 .
　トレーニング制御信号により、入力信号がトレーニング信号であるかデータ信号であるかが指定される。入力信号がトレーニング信号である場合には、その正解ラベルも同時に通知される。この場合、ＳＶＭ回路５１１は、トレーニング信号及び正解ラベルを用いて、ＳＶＭによる機械学習を行う。入力信号がデータ信号である場合、機械学習の結果を用いた判定を行い、判定結果を出力する。領域判定回路５１が領域選択回路４１の一部（領域判定回路４１２）である場合は、判定結果は１ｘｎｓスイッチ４１３の制御信号として使用される。 The training control signal specifies whether the input signal is a training signal or a data signal. If the input signal is a training signal, its correct label is also notified at the same time. In this case, the SVM circuit 511 uses the training signal and the correct label to perform machine learning by SVM. When the input signal is a data signal, a determination is made using the result of machine learning, and the determination result is output. If the area determination circuit 51 is part of the area selection circuit 41 (area determination circuit 412 ), the determination result is used as a control signal for the 1×ns switch 413 .
　図７は、ＮＮを用いた領域判定回路５１の構成例を示す図である。領域判定回路５１は、領域選択回路４１の一部（領域判定回路４１２）を構成するものであるとともに、領域判定層５０でも使用される。領域判定回路５１には、トレーニング信号又はデータ信号が入力される。領域判定回路５１はＮＮ回路５１２を備える。領域判定回路５１は、１個のＮＮ回路５１２そのものとして構成されてもよい。 FIG. 7 is a diagram showing a configuration example of the area determination circuit 51 using NN. The area determination circuit 51 forms part of the area selection circuit 41 (area determination circuit 412 ) and is also used in the area determination layer 50 . A training signal or a data signal is input to the area determination circuit 51 . The area determination circuit 51 has an NN circuit 512 . The area determination circuit 51 may be configured as one NN circuit 512 itself.
　トレーニング制御信号により、入力信号がトレーニング信号であるかデータ信号であるかが指定される。入力信号がトレーニング信号である場合には、その正解ラベルも同時に通知される。この場合、ＮＮ回路５１２は、トレーニング信号及び正解ラベルを用いて、ＮＮによる機械学習を行う。入力信号がデータ信号である場合、機械学習の結果を用いた判定を行い、判定結果を出力する。領域判定回路５１が領域選択回路４１の一部（領域判定回路４１２）である場合は、判定結果は１ｘｎｓスイッチ４１３の制御信号として使用される。 The training control signal specifies whether the input signal is a training signal or a data signal. If the input signal is a training signal, its correct label is also notified at the same time. In this case, the NN circuit 512 performs NN machine learning using the training signal and the correct label. When the input signal is a data signal, a determination is made using the result of machine learning, and the determination result is output. If the area determination circuit 51 is part of the area selection circuit 41 (area determination circuit 412 ), the determination result is used as a control signal for the 1×ns switch 413 .
　図８は、２分岐の領域選択回路４１を３層組み合わせて１６ＱＡＭ信号を判定する構成例を示す図である。図９は、２分岐の領域選択回路４１を３層組み合わせて１６ＱＡＭ信号を判定する構成の処理の概略を示す図である。領域選択層４０＿１では、図９の第１層に示されるように、１６ＱＡＭコンスタレーションを左右２つの領域に分け、機械学習のトレーニングが行われる。 FIG. 8 is a diagram showing a configuration example in which three layers of two-branch area selection circuits 41 are combined to determine a 16QAM signal. FIG. 9 is a diagram showing an outline of processing of a configuration for determining a 16QAM signal by combining three layers of two-branch area selection circuits 41. In FIG. In the region selection layer 40_1, as shown in the first layer of FIG. 9, the 16QAM constellation is divided into left and right regions, and machine learning training is performed.
　領域選択層４０＿２では、図９の第２層に示されるように、左右各々の領域を上下に分けて機械学習のトレーニングが行われる。以下同様に、領域が１６分割されるまでこのような領域分割が繰り返される。すなわち、図８の具体例では、全ての領域選択回路４１においてＬ（ｍ，ｎ）＝２と設定され、且つ、全ての領域判定回路５１においてＬ（ｘ，ｎ）＝２と設定されている。また、領域判定回路５１は、サポートベクトルマシンを用いた１対１方式の機械学習により受信シンボルを２個の領域のうちいずれか一つに分類する。 In the region selection layer 40_2, as shown in the second layer of FIG. 9, machine learning training is performed by dividing the left and right regions into upper and lower regions. Similarly, such region division is repeated until the region is divided into 16 regions. That is, in the specific example of FIG. 8, all the area selection circuits 41 are set to L(m, n)=2, and all the area determination circuits 51 are set to L(x, n)=2. . Also, the area determination circuit 51 classifies the received symbols into one of the two areas by one-to-one machine learning using a support vector machine.
　本実施形態では、必要な機械学習回路数（領域判定回路４１２及び領域判定回路５１の数）は１５個となり、従来方式で必要な４個と比べると機械学習回路数は増加する。しかし、個々の機械学習回路の内部変数が減少するため、全体での内部変数の数は同程度となる。一方、トレーニング（機械学習処理）を終了してシンボルの判定を行う際には、例えば第１層（領域選択層４０＿１）で受信シンボルが右側の領域に属すると判定されると、それ以降の計算では左側の領域の演算は行われない。このため、従来方式と比べて演算量及び計算実行時間を減らすことができる。この例では領域２分割を繰り返すように仮定したが、２分割以上の多分割でも良いし、階層ごとに分割数が異なっても良い。同じ階層内の異なる領域において分割数が異なっても良い。 In this embodiment, the required number of machine learning circuits (the number of the area determination circuits 412 and the area determination circuits 51) is 15, which is more than the four required in the conventional method. However, since the internal variables of the individual machine learning circuits are reduced, the overall number of internal variables is comparable. On the other hand, when the training (machine learning process) is completed and symbol determination is performed, for example, if the received symbol is determined to belong to the right region in the first layer (region selection layer 40_1), the subsequent calculation does not operate on the left region. Therefore, it is possible to reduce the amount of calculation and the calculation execution time as compared with the conventional method. In this example, it is assumed that the division into two regions is repeated, but multiple divisions of two or more divisions may be used, and the number of divisions may differ for each layer. Different areas within the same hierarchy may have different numbers of divisions.
　図１０は、ＳＶＭの原理を示す図である。○で示す受信シンボルと△で示す受信シンボルとを、できるだけ低い謝り率で判別できるような境界線を引きたい。ＳＶＭの機械学習のトレーニング時には、境界線から○、△各々の領域の内側にマージン領域が設定される。そして、各々の領域から見てマージン領域の外側にある受信シンボルのみが判定に使用される。以下この受信シンボルをサポートベクトルと呼ぶ。マージン領域の内側にある受信シンボルは、判定には使用されない。このため、トレーニング時において、マージン領域の外側にある受信シンボルの数を減らせば、データ判定時の演算量を減らすことができる。このような処理が行われることで、判定に当たって無視できる行列係数が増加するため、パラメータが減少し、データ判定時の演算量を減らすことができる。ＮＮの場合は、内部変数のうち０に近いものをあえて削除することで、データ判定時の演算量を減らすことができる。 FIG. 10 is a diagram showing the principle of SVM. It is desired to draw a boundary line so that the received symbols indicated by ◯ and the received symbols indicated by Δ can be discriminated with as low an apology rate as possible. During machine learning training of the SVM, a margin area is set inside each of the circle and triangle areas from the boundary line. Then, only received symbols outside the margin area when viewed from each area are used for decision making. These received symbols are hereinafter referred to as support vectors. Received symbols inside the margin region are not used for decision making. Therefore, if the number of received symbols outside the margin area is reduced during training, the amount of computation during data determination can be reduced. By performing such processing, the number of matrix coefficients that can be ignored in determination increases, so the number of parameters decreases, and the amount of computation during data determination can be reduced. In the case of the neural network, the amount of computation during data determination can be reduced by deliberately deleting those internal variables that are close to 0.
　図１１は、シミュレーション結果を示す図である。図中の既存方式は、従来技術による非線形補償である。提案方式１は、図８にて説明した領域分割手順により１層につき２分割ずつの非線形補償を行なったものである。提案方式２はＳＶＭを用いた総当たり方式により１層につき４分割ずつの非線形補償を行なったものである。１６ＱＡＭ、６４ＱＡＭ、２５６ＱＡＭの３つのパターンにてシミュレーションを行ない、１シンボルの判定が完了するまでに演算に使用するＳＶの総数をカウントし、その平均値を算出した。同時にＱ値についても確認した。ＳＶＭのカーネル関数にはＲＢＦカーネルを使用し、ハイパーパラメーターはシミュレーション条件毎にＳＶ数が最小となる最適値に設定している。非線形歪みについてはＱＡＭ信号の各々のシンボル点をその強度に比例して平均１０度回転させることにより模擬している。いずれの条件でもＳＶ数の減少とＱ値の上昇が認められ、本提案の効果が確認できる。 FIG. 11 is a diagram showing simulation results. The existing scheme in the figure is non-linear compensation according to the prior art. Proposed method 1 performs nonlinear compensation by dividing each layer into two by the region dividing procedure described with reference to FIG. Proposed method 2 performs non-linear compensation by four divisions per layer by a round-robin method using SVM. Three patterns of 16QAM, 64QAM, and 256QAM were simulated, the total number of SVs used for calculation was counted until the determination of one symbol was completed, and the average value was calculated. At the same time, the Q value was also confirmed. An RBF kernel is used as the SVM kernel function, and hyperparameters are set to optimal values that minimize the number of SVs for each simulation condition. Nonlinear distortion is simulated by rotating each symbol point of the QAM signal by an average of 10 degrees in proportion to its strength. A decrease in the number of SVs and an increase in the Q value are observed under any conditions, and the effect of this proposal can be confirmed.
　本実施形態における各回路は、ＣＰＵ（Central Processing Unit）等のプロセッサーとメモリーとを用いて構成されてもよい。本実施形態における各回路の全て又は一部は、ＡＳＩＣ（Application Specific Integrated Circuit）やＰＬＤ（Programmable Logic Device）やＦＰＧＡ（Field Programmable Gate Array）等のハードウェアを用いて実現されても良い。上記のプログラムは、コンピューター読み取り可能な記録媒体に記録されても良い。コンピューター読み取り可能な記録媒体とは、例えばフレキシブルディスク、光磁気ディスク、ＲＯＭ、ＣＤ－ＲＯＭ、半導体記憶装置（例えばＳＳＤ：Solid State Drive）等の可搬媒体、コンピューターシステムに内蔵されるハードディスクや半導体記憶装置等の記憶装置である。上記のプログラムは、電気通信回線を介して送信されてもよい。 Each circuit in this embodiment may be configured using a processor such as a CPU (Central Processing Unit) and a memory. All or part of each circuit in this embodiment may be implemented using hardware such as ASIC (Application Specific Integrated Circuit), PLD (Programmable Logic Device), FPGA (Field Programmable Gate Array), and the like. The above program may be recorded on a computer-readable recording medium. Computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, semiconductor storage devices (such as SSD: Solid State Drives), hard disks and semiconductor storage built into computer systems. It is a storage device such as a device. The above program may be transmitted via telecommunication lines.
　以上、この発明の実施形態について図面を参照して詳述してきたが、具体的な構成はこの実施形態に限られるものではなく、この発明の要旨を逸脱しない範囲の設計等も含まれる。 Although the embodiment of the present invention has been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes design within the scope of the gist of the present invention.
　本発明は、光通信に適用可能である。 The present invention is applicable to optical communication.
１００…光通信システム、１０…送信装置、１１…多重装置、１２…光変調装置、１２１…送信信号選択部、１２２…光変調部、２０…受信装置、２１…光復調装置、２２…分離装置、２１１…分散補償部、２１２…受信シンボル判定部、３０…制御装置、１２１１…トレーニング制御回路、１２１２…選択スイッチ、２１２１…トレーニング制御回路、２１２２…判定部、４０…領域選択層、４１…領域選択回路、５０…領域判定層、５１…領域判定回路、６０…多重化回路、４１１…２分岐回路、４１２…領域判定回路、４１３…１ｘｎｓスイッチ、５１１…ＳＶＭ回路、５１２…ＮＮ回路 DESCRIPTION OF SYMBOLS 100... Optical communication system 10... Transmission apparatus 11... Multiplex apparatus 12... Optical modulation apparatus 121... Transmission signal selection part 122... Optical modulation part 20... Reception apparatus 21... Optical demodulation apparatus 22... Separation apparatus , 211... dispersion compensation unit, 212... received symbol determination unit, 30... control device, 1211... training control circuit, 1212... selection switch, 2121... training control circuit, 2122... determination unit, 40... area selection layer, 41... area Selection circuit 50 Area determination layer 51 Area determination circuit 60 Multiplexing circuit 411 Two-branch circuit 412 Area determination circuit 413 1xns switch 511 SVM circuit 512 NN circuit

Claims (6)

  1. 　受信された光信号について多値信号のシンボル判定を行い、前記シンボル判定の結果に基づいてデータを復元する受信シンボル判定部を備え、
    　前記受信シンボル判定部は、第１層から第Ｍ層（Ｍは１以上の整数）の領域選択層と、１層の領域判定層と、を備え、
    　第ｍ層（ｍは１以上Ｍ以下の整数）の領域選択層は、第ｍ－１層の領域選択層での領域分割数（ｍ＝１の場合は“１”）と同数の領域選択回路を備え、
    　第ｍ層の領域選択回路は、受信された前記多値信号について、層の段数（１からＭ）に応じて複数の領域の中から、受信された前記多値信号のシンボルが属する領域を機械学習の結果に基づいて選択し、選択された領域に応じて次の層へ出力し、
    　前記第ｍ層の領域選択回路は、前記第ｍ－１層の領域選択回路で選択された領域に応じた領域選択回路が処理を行い、
    　前記領域判定層は第Ｍ層の領域選択層での領域選択回路の出力数の総数に応じた領域判定回路を備え、
    　前記領域判定回路は、第Ｍ層の領域選択層から受信された前記多値信号のシンボルが属する領域を機械学習の結果に基づいて選択し、選択された領域に応じて前記多値信号のシンボルを示す判定結果を出力する、光送受信装置。     a received symbol determination unit that performs symbol determination of a multilevel signal on a received optical signal and restores data based on the result of the symbol determination;
    The received symbol determination unit includes a first layer to an Mth layer (M is an integer equal to or greater than 1) region selection layers, and a first layer region determination layer,
    The area selection layer of the m-th layer (m is an integer of 1 or more and M or less) has the same number of area selection circuits as the number of area divisions in the m-1-th area selection layer ("1" when m = 1) with
    An m-th layer region selection circuit selects a region to which a symbol of the received multi-level signal belongs from among a plurality of regions according to the number of layers (1 to M) for the received multi-level signal. Select based on the result of learning, output to the next layer according to the selected area,
    The area selection circuit of the m-th layer performs processing by an area selection circuit corresponding to the area selected by the area selection circuit of the (m−1)-th layer,
    The area determination layer includes an area determination circuit corresponding to the total number of outputs of the area selection circuits in the M-th area selection layer,
    The area determination circuit selects an area to which the symbol of the multi-level signal received from the M-th area selection layer belongs based on the result of machine learning, and selects the symbol of the multi-level signal according to the selected area. An optical transmitting/receiving device that outputs a determination result indicating
  2. 　前記領域選択回路及び前記領域判定回路は、トレーニングモードで動作する場合には、受信された前記多値信号と、前記多値信号が有するシンボルを示す正解ラベルと、を用いて機械学習を行うことで、受信された多値信号に応じて前記複数の領域からどの領域を選択するか学習し、
    　前記領域選択回路及び前記領域判定回路は、データ判定モードで動作する場合には、受信された前記多値信号について、前記トレーニングモードで得られた学習結果に基づいて前記領域を選択する、請求項１に記載の光送受信装置。     When operating in a training mode, the area selection circuit and the area determination circuit perform machine learning using the received multilevel signal and a correct label indicating a symbol included in the multilevel signal. learning which region to select from the plurality of regions according to the received multilevel signal,
    3. The area selection circuit and the area determination circuit, when operating in the data determination mode, select the area for the received multilevel signal based on learning results obtained in the training mode. 2. The optical transmitter-receiver according to 1.
  3. 　前記領域選択回路及び前記領域判定回路における前記機械学習は、ニューラルネットワークを用いた機械学習であり、前記領域選択回路及び前記領域判定回路は、前記シンボルを複数の領域のうちいずれか一つに分類する、請求項１又は２に記載の光送受信装置。 The machine learning in the area selection circuit and the area determination circuit is machine learning using a neural network, and the area selection circuit and the area determination circuit classify the symbol into one of a plurality of areas. 3. The optical transmitting/receiving device according to claim 1 or 2.
  4. 　前記領域選択回路及び前記領域判定回路における前記機械学習は、サポートベクトルマシンを用いた機械学習であり、１対他方式の機械学習により、前記領域選択回路及び前記領域判定回路は、前記シンボルを複数の領域のうちいずれか一つに分類する、請求項１又は２に記載の光送受信装置。 The machine learning in the area selection circuit and the area determination circuit is machine learning using a support vector machine, and by one-to-many machine learning, the area selection circuit and the area determination circuit select a plurality of symbols. 3. The optical transmitting/receiving device according to claim 1, which is classified into any one of the regions of .
  5. 　受信された光信号について多値信号のシンボル判定を行い、前記シンボル判定の結果に基づいてデータを復元する受信シンボル判定部を備え、
    　前記受信シンボル判定部は、第１層から第Ｍ層（Ｍは１以上の整数）の領域選択層と、１層の領域判定層と、を備え、
    　第ｍ層（ｍは１以上Ｍ以下の整数）の領域選択層は、第ｍ－１層の領域選択層での領域分割数（ｍ＝１の場合は“１”）と同数の領域選択回路を備え、
    　前記領域判定層は第Ｍ層の領域選択層での領域選択回路の出力数の総数に応じた領域判定回路を備える光通信装置において、
    　第ｍ層の領域選択回路が、受信された前記多値信号について、層の段数（１からＭ）に応じて複数の領域の中から、受信された前記多値信号のシンボルが属する領域を機械学習の結果に基づいて選択し、選択された領域に応じて次の層へ出力するステップと、
    　前記領域判定回路が、第Ｍ層の領域選択層から受信された前記多値信号のシンボルが属する領域を機械学習の結果に基づいて選択し、選択された領域に応じて前記多値信号のシンボルを示す判定結果を出力するステップと、を有し、
    　前記第ｍ層の領域選択回路は、前記第ｍ－１層の領域選択回路で選択された領域に応じた領域選択回路が処理を行う、光通信方法。     a received symbol determination unit that performs symbol determination of a multilevel signal on a received optical signal and restores data based on the result of the symbol determination;
    The received symbol determination unit includes a first layer to an Mth layer (M is an integer equal to or greater than 1) region selection layers, and a first layer region determination layer,
    The area selection layer of the m-th layer (m is an integer of 1 or more and M or less) has the same number of area selection circuits as the number of area divisions in the m-1-th area selection layer ("1" when m = 1) with
    In the optical communication device, the area determination layer includes an area determination circuit corresponding to the total number of outputs of the area selection circuits in the area selection layer of the M-th layer,
    A region selection circuit of the m-th layer selects a region to which a symbol of the received multi-level signal belongs from among a plurality of regions according to the number of layers (1 to M) for the received multi-level signal. selecting based on the result of learning and outputting to the next layer according to the selected area;
    The area determination circuit selects an area to which the symbol of the multi-level signal received from the M-th area selection layer belongs based on a result of machine learning, and selects the symbol of the multi-level signal according to the selected area. and a step of outputting a determination result indicating
    The optical communication method according to the optical communication method, wherein the area selection circuit of the m-th layer performs processing by an area selection circuit corresponding to the area selected by the area selection circuit of the (m−1)-th layer.
  6. 　請求項１から４のいずれか一項に記載の光送受信装置としてコンピューターを機能させるためのコンピュータープログラム。 A computer program for causing a computer to function as the optical transceiver according to any one of claims 1 to 4.
PCT/JP2021/047580 2021-12-22 2021-12-22 Optical transceiving device, optical communication method, and computer program WO2023119487A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2021/047580 WO2023119487A1 (en) 2021-12-22 2021-12-22 Optical transceiving device, optical communication method, and computer program

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2021/047580 WO2023119487A1 (en) 2021-12-22 2021-12-22 Optical transceiving device, optical communication method, and computer program

Publications (1)

Publication Number Publication Date
WO2023119487A1 true WO2023119487A1 (en) 2023-06-29

Family

ID=86901631

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2021/047580 WO2023119487A1 (en) 2021-12-22 2021-12-22 Optical transceiving device, optical communication method, and computer program

Country Status (1)

Country Link
WO (1) WO2023119487A1 (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH09214578A (en) * 1996-01-30 1997-08-15 Fujitsu Ltd Carrier reproduction circuit
JPH10136046A (en) * 1996-10-31 1998-05-22 Jisedai Digital Television Hoso Syst Kenkyusho:Kk Soft judgment system and reception device
JP2017163485A (en) * 2016-03-11 2017-09-14 国立大学法人名古屋大学 Optical receiver and optical symbol label identification method in optical receiver
JP2021034816A (en) * 2019-08-21 2021-03-01 株式会社東芝 Symbol determination device of input signal, method, and program

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH09214578A (en) * 1996-01-30 1997-08-15 Fujitsu Ltd Carrier reproduction circuit
JPH10136046A (en) * 1996-10-31 1998-05-22 Jisedai Digital Television Hoso Syst Kenkyusho:Kk Soft judgment system and reception device
JP2017163485A (en) * 2016-03-11 2017-09-14 国立大学法人名古屋大学 Optical receiver and optical symbol label identification method in optical receiver
JP2021034816A (en) * 2019-08-21 2021-03-01 株式会社東芝 Symbol determination device of input signal, method, and program

Similar Documents

Publication Publication Date Title
Zhang et al. Non-data-aided k-nearest neighbors technique for optical fiber nonlinearity mitigation
CN108667523B (en) Optical fiber nonlinear equalization method based on data-aided KNN algorithm
US8446989B2 (en) Signal separating device and signal separating method
US11050494B2 (en) Signal-multiplexing apparatus and method based on machine learning
Zhao et al. Low complexity OSNR monitoring and modulation format identification based on binarized neural networks
Yun et al. Reinforcement learning for link adaptation in MIMO-OFDM wireless systems
WO2023119487A1 (en) Optical transceiving device, optical communication method, and computer program
Xu et al. RC-struct: A structure-based neural network approach for MIMO-OFDM detection
US20160277175A1 (en) Clock Recovery and Equalizer Estimation In a Multi-Channel Receiver
Xu et al. Detect to learn: Structure learning with attention and decision feedback for MIMO-OFDM receive processing
CN110870213A (en) Space-time precoding for faster than Nyquist signal transmission
Wu et al. Automatic digital modulation recognition based on support vector machines
JP4340759B2 (en) Digital data transmission device
Bajaj et al. Performance analysis of recurrent neural network-based digital pre-distortion for optical coherent transmission
Zecchin et al. In-Context Learning for MIMO Equalization Using Transformer-Based Sequence Models
JP3843891B2 (en) Multilevel modulation apparatus, multilevel demodulation apparatus, multilevel modulation / demodulation communication system, modulation / demodulation program, and modulation / demodulation method
Anderson et al. Deep modulation (Deepmod): A self-taught PHY layer for resilient digital communications
Choqueuse et al. ParamNet: A multi-layer parametric network for joint channel estimation and symbol detection
EP4026056A1 (en) Signal demapping
Jovanovic et al. Rate adaptive autoencoder-based geometric constellation shaping
CN114679223B (en) Machine learning-based mode division multiplexing signal demodulation optimization method
Mumtaz et al. Autonomous modulation classification using single inception module based convolutional neural network
CN109526006B (en) Resource optimization method for maximizing URLLC service quantity of network function virtualization
WO2022249465A1 (en) Training apparatus, control method, and non-transitory computer-readable storage medium
CN109495402B (en) Resource optimization method for minimizing physical layer resources of network function virtualization

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21968924

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2023568880

Country of ref document: JP

Kind code of ref document: A