WO2019116503A1 - Optical reception device and optical reception method - Google Patents

Optical reception device and optical reception method Download PDF

Info

Publication number
WO2019116503A1
WO2019116503A1 PCT/JP2017/044916 JP2017044916W WO2019116503A1 WO 2019116503 A1 WO2019116503 A1 WO 2019116503A1 JP 2017044916 W JP2017044916 W JP 2017044916W WO 2019116503 A1 WO2019116503 A1 WO 2019116503A1
Authority
WO
WIPO (PCT)
Prior art keywords
unit
likelihood
symbol
input
neural network
Prior art date
Application number
PCT/JP2017/044916
Other languages
French (fr)
Japanese (ja)
Inventor
怜典 松本
Original Assignee
三菱電機株式会社
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 三菱電機株式会社 filed Critical 三菱電機株式会社
Priority to PCT/JP2017/044916 priority Critical patent/WO2019116503A1/en
Publication of WO2019116503A1 publication Critical patent/WO2019116503A1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H03—ELECTRONIC CIRCUITRY
    • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
    • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
    • H03M13/37—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35
    • H03M13/45—Soft decoding, i.e. using symbol reliability information
    • 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
    • H04B10/66—Non-coherent receivers, e.g. using direct detection
    • H04B10/69—Electrical arrangements in the receiver
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B3/00—Line transmission systems
    • H04B3/02—Details
    • H04B3/04—Control of transmission; Equalising
    • H04B3/06—Control of transmission; Equalising by the transmitted signal

Definitions

  • the present invention relates to an optical receiving apparatus and an optical receiving method for receiving multilevel optical information.
  • LAN local area network
  • LAN local area network
  • PAM pulse amplitude modulation
  • the signal equalizer compensates for linear distortion of the data signal generated in the optical transceiver and the optical transmission line.
  • a feed forward equalizer hereinafter referred to as FFE
  • FFE feed forward equalizer
  • a signal equalizer having a simple configuration a feed forward equalizer having a simple configuration.
  • FFE can not cope with non-linear distortions such as modulation instability, square wave detection distortion, and gain saturation that occur in an optical transmitter / receiver adopting the PAM method.
  • the light receiving device described in Non-Patent Document 1 uses a neural network.
  • a neural network can equalize nonlinear distortions occurring in an optical transmitter-receiver and an optical transmission path by performing nonlinear conversion by neuron elements in the intermediate layer.
  • the optical receiver described in Non-Patent Document 1 improves the equalization accuracy of non-linear distortion with respect to PAM by performing signal processing in block units composed of a plurality of symbols.
  • the optical receiver described in Non-Patent Document 1 has a problem that the circuit scale increases exponentially with the number of input symbols.
  • An object of the present invention is to solve the above-mentioned problems, and it is an object of the present invention to provide an optical receiving apparatus and an optical receiving method capable of enhancing the circuit mounting efficiency while maintaining the equalization accuracy of nonlinear distortion.
  • An optical receiving apparatus is an optical receiving apparatus that receives an optical signal including a plurality of symbols, and includes a photoelectric conversion unit, a first adjustment unit, a neural network unit, a likelihood selection unit, and a second adjustment unit. , A demodulation unit, and a first error correction decoding unit.
  • the photoelectric conversion unit converts the received optical signal into an electrical signal.
  • the first adjustment unit converts the electrical signal converted from the light signal by the photoelectric conversion unit into a parallel symbol string.
  • the neural network unit has an input layer, one or more intermediate layers, and output layers equal in number to the number of symbols input to the input layer, and the symbols and elements input in parallel from the first adjustment unit are connected The non-linear transformation of the multiplication sum with the combined weight of to generate the likelihood for each symbol.
  • the likelihood selection unit selects a likelihood for each output layer from the likelihoods generated by the neural network unit.
  • the second adjustment unit converts parallel symbol strings corresponding to the likelihoods selected by the likelihood selection unit into serial symbol strings.
  • the demodulation unit demodulates the bit string from the symbol string input from the second adjustment unit.
  • the first error correction decoding unit performs error correction decoding by hard decision based on the bit string demodulated by the demodulation unit.
  • the optical transmitter / receiver and the neural network unit for equalizing the nonlinear distortion generated in the optical transmission path have the same number of output layers as the number of input symbols, the number of elements in the output layer is not the modulation of the optical signal. It can be reduced to the number obtained by multiplying the multi-value degree by the number of input symbols. As a result, the light receiving device can increase the circuit mounting efficiency while maintaining the equalization accuracy of the non-linear distortion.
  • FIG. 1 is a block diagram showing a configuration of a neural network unit in Embodiment 1.
  • FIG. 5 is a block diagram showing a configuration of a likelihood selection unit in Embodiment 1.
  • 3 is a flowchart showing an optical reception method according to Embodiment 1.
  • FIG. It is a graph which shows the result of having simulated the relationship between a bit error rate and receiving average optical power.
  • 7 is a flowchart showing an optical reception method according to Embodiment 2.
  • FIG. 8A is a graph showing the relationship between the log likelihood of the first bit and the normalized amplitude.
  • FIG. 8B is a graph showing the relationship between the log likelihood of the second bit and the normalized amplitude.
  • FIG. 9A is a block diagram showing a hardware configuration for realizing the respective functions of the optical receiving apparatus according to Embodiment 1 and Embodiment 2.
  • FIG. 9B is a block diagram showing a hardware configuration that executes software for realizing the respective functions of the optical reception devices according to Embodiment 1 and Embodiment 2.
  • FIG. 1 is a block diagram showing the configuration of an optical receiver 1 according to Embodiment 1 of the present invention.
  • FIG. 2 is a block diagram showing the configuration of the neural network unit 4.
  • FIG. 3 is a block diagram showing the configuration of the likelihood selection unit 5.
  • the light receiving device 1 is a device that receives an optical signal including a plurality of symbols, and the photoelectric conversion unit 2, the reception signal adjustment unit 3, the neural network unit 4, the likelihood selection unit 5, and the symbol restoration unit 6.
  • the photoelectric conversion unit 2 photoelectrically converts the received optical signal to convert it into an analog electrical signal proportional to the optical signal intensity.
  • the photoelectric conversion unit 2 may include an oscillator that oscillates the local oscillation light. Furthermore, the photoelectric conversion unit 2 detects an electrical signal of analog type, and converts it into an electrical signal of digital type (hereinafter referred to as digital signal).
  • digital signal generated by the photoelectric conversion unit 2 is output to the reception signal adjustment unit 3.
  • the reception signal adjustment unit 3 is a first adjustment unit that converts the digital signal input from the photoelectric conversion unit 2 into a parallel symbol string using electrical processing. Further, the reception signal adjustment unit 3 adjusts the sampling rate of the digital signal input from the photoelectric conversion unit 2.
  • the neural network unit 4 has an input layer 40, N middle layers 41, and output layers 42-1 to 42-M, as shown in FIG. N and M are integers of 1 or more.
  • a parallel symbol string whose sampling rate has been adjusted by the received signal adjustment unit 3 is input to the neural network unit 4.
  • the neural network unit 4 shown in FIG. 2 receives 1 to 1 symbols.
  • the symbols are output from the input layer 40 to the intermediate layer 41, and output from the intermediate layer 41 to the output layers 42-1 to 42-M.
  • the neural network unit 4 non-linearly transforms the multiplication sum of the symbol and the connection weight between the neuron elements in each layer to calculate the likelihood.
  • Non-linear functions utilized for non-linear transformation include step functions, sigmoid functions, and ramp functions.
  • the output layers 42-1 to 42-M of the neural network unit 4 have the same number or more as the number of symbols input to the input layer 40, and output likelihood for each symbol.
  • the modulation level D is the number of bits per symbol, and the modulation level D of the 4-value PAM signal is "4".
  • B symbols of a 4-value PAM signal are input to the neural network unit 4
  • the number of output layers for outputting the likelihood is B
  • the total number of elements in the B output layers is 4 ⁇ B.
  • the output layers 42-1 to 42-M are out of order.
  • the likelihood selection unit 5 selects the likelihood for each output layer from the likelihood generated for each symbol by the neural network unit 4.
  • the likelihood selection unit 5 includes maximum value calculation units 50-1 to 50-M and element number extraction units 51-1 to 51-M.
  • the maximum value calculation units 50-1 to 50-M select the maximum value of likelihood for each output layer from the likelihood for each symbol generated by the neural network unit 4.
  • the element number extraction units 51-1 to 51-M extract element numbers corresponding to the likelihood of the maximum value input from the maximum value calculation units 50-1 to 50-M.
  • the symbol restoration unit 6 restores the symbol from the likelihood selected for each output layer by the likelihood selection unit 5 and outputs a parallel symbol string composed of the restored symbol.
  • the symbol restoration unit 6 generates a symbol corresponding to the element number input from the element number extraction units 51-1 to 51-M. For example, in the recovery of a symbol for which a four-level phase modulation signal is to be demodulated, the symbol recovery unit 6 receives an element number (0, 1, 2, 3) when the element number (0, 1, 2, 3) is input. ) Is converted to the symbol (1 + i, ⁇ 1 + i, -1-i, 1 ⁇ i).
  • FIG. 1 shows the configuration provided with the symbol recovery unit 6, the likelihood selection unit 5 correlates the output element number k with the symbol and outputs the symbol from the light receiving device 1
  • the symbol restoration unit 6 can be omitted. That is, when the likelihood selection unit 5 outputs a parallel symbol string corresponding to the likelihood selected for each output layer, the light receiving device 1 may not include the symbol recovery unit 6.
  • the delay adjusting unit 7 is a second adjusting unit that converts a parallel symbol string corresponding to the likelihood selected by the likelihood selecting unit 5 into a serial symbol string.
  • the delay adjusting unit 7 inputs the symbol string in parallel from the symbol restoring unit 6 and converts the input symbol string in series.
  • the demodulation unit 8 demodulates a bit string from the symbol string input from the delay adjustment unit 7. For example, the demodulator 8 converts the input symbol into a bit string in accordance with an assignment rule for changing from symbol to bit.
  • the hard decision error correction decoding unit 9 is a first error correction decoding unit that performs error correction decoding based on hard decision based on the bit string demodulated by the demodulation unit 8. For example, the hard decision error correction decoding unit 9 detects and corrects a bit error generated in the optical transmitter-receiver or the optical transmission line based on the bit assignment rule of the error correction code.
  • Hard-decision error correction codes include Reed-Solomon codes and the like.
  • the connection load update unit 10 updates the connection load of the neural network unit 4.
  • the connection weight update unit 10 updates the connection load of the neural network unit 4 so that the error between the symbol restored by the symbol restoration unit 6 and the learning value is minimized.
  • An error back-propagation method that minimizes the squared error is widely used for coupling weight updating.
  • the connection load updating unit 10 may be a component included in an apparatus provided separately from the light receiving apparatus 1.
  • the coupling load updating unit 10 may be a component provided in a learning device provided separately from the light receiving device 1.
  • the learning device updates the connection weight of the neural network unit 4 using the symbol input from the likelihood selection unit 5 or the symbol restoration unit 6.
  • FIG. 4 is a flowchart showing the light receiving method according to the first embodiment.
  • the photoelectric conversion unit 2 photoelectrically converts the received optical signal, and converts the analog electric signal obtained by the photoelectric conversion into a digital signal (step ST1).
  • the reception signal adjustment unit 3 converts the digital signal input from the photoelectric conversion unit 2 into a parallel symbol string (step ST2).
  • the neural network unit 4 non-linearly converts the multiplication sum of the symbol input in parallel from the received signal adjustment unit 3 and the connection weight between the neural elements to generate the likelihood for each symbol (step ST3). .
  • the neural network unit 4 outputs the likelihood corresponding to the correct answer symbol to the likelihood selection unit 5.
  • the output element number k is given to the neuron element in the j-th output layer, the symbol number of the symbol group input to the input layer is ⁇ , the correct answer class is C 1, and the number of output layers is M.
  • D be the modulation level.
  • the posterior probability is p jk (C 1
  • L jk corresponding to the symbol belonging to the correct answer class C 1 output from the neuron element specified by the output element number k in the j-th output layer is calculated, for example, using the following equation (1) .
  • L jk ⁇ ln [p jk (C 1
  • the likelihood selection unit 5 selects the likelihood for each output layer from the likelihood generated for each symbol by the neural network unit 4 (step ST4).
  • the maximum value calculation units 50-1 to 50-M input in parallel the likelihoods output from each of the n neuron elements in the output layer, and among the n likelihoods, choose the maximum value of.
  • Each of the maximum value calculation units 50-1 to 50-M is, for example, the maximum value ML of the likelihood output from the neuron element specified by the output element number k in the j-th output layer according to the following equation (2) Calculate j .
  • k 0,..., D ⁇ 1.
  • ML j max k (L jk ) (2)
  • the maximum value ML j of the likelihood selected from the j-th output layer by the j-th maximum value operation unit is output to the j-th element number extraction unit.
  • the j-th element number extraction unit extracts an element number corresponding to the likelihood of the maximum value ML j and outputs the element number to the symbol recovery unit 6.
  • the element number is any one of the output element numbers k for specifying the neuron element in the j-th output layer.
  • the symbol restoration unit 6 generates a symbol corresponding to the element number input from each of the element number extraction units 51-1 to 51-M, and outputs the generated symbol symbol sequence to the delay adjustment unit 7. Thereby, the parallel symbol string corresponding to the likelihood selected by the likelihood selection unit 5 is input to the delay adjustment unit 7.
  • the delay adjusting unit 7 converts the parallel symbol string corresponding to the likelihood selected by the likelihood selecting unit 5 into a serial symbol string (step ST5).
  • the symbol string serially converted by the delay adjustment unit 7 is output to the demodulation unit 8.
  • the demodulation unit 8 demodulates the bit string from the symbol string input from the delay adjustment unit 7 (step ST6).
  • the bit string demodulated by the demodulator 8 is output to the hard decision error correction decoder 9.
  • the hard decision error correction decoding unit 9 performs error correction decoding by hard decision based on the bit string demodulated by the demodulation unit 8 (step ST7).
  • a non-linear function is used to calculate the likelihood by the neural network unit 4. This cancels out non-linear distortion generated in the optical transmitter-receiver and the optical transmission line. Furthermore, in the optical receiving device described in Non-Patent Document 1, in order to calculate the likelihood, it is necessary to calculate the number obtained by raising the modulation multi-level degree by the number of input symbols. As a result, the number of elements in the output layer becomes enormous and the circuit mounting efficiency is poor.
  • the neural network unit 4 in the first embodiment an output layer is arranged for each input symbol. For this reason, the neural network unit 4 can generate the likelihood for each symbol only by calculating the number obtained by multiplying the modulation multi-value and the number of input symbols. Therefore, the light receiving device 1 can realize higher circuit mounting efficiency than the light receiving device described in Non-Patent Document 1.
  • FIG. 5 is a graph showing the simulation result of the relationship between the bit error rate and the received average optical power (dBm).
  • a 4-value PAM signal with a bit rate of 100 Gbit / s is input from the optical transmitter to the receiver without fiber transmission.
  • the photodetector provided in the optical receiver performs square detection on the input optical signal.
  • the electric signal detected by the square wave detection is input to the light receiving device 1 after passing through a 15-tap feed forward equalizer (hereinafter referred to as FFE).
  • FFE 15-tap feed forward equalizer
  • the bit error rate is calculated by the demodulator 8 based on the symbols.
  • the hard decision error correction decoding unit 9 simulates the operation by setting the forward error correction limit to 2 ⁇ 10 ⁇ 4 .
  • data a is a simulation result obtained by simulating an optical receiving apparatus having a configuration in which linear distortion and non-linear distortion are not equalized without using a neural network.
  • Data b is a simulation result obtained by simulating an optical receiving apparatus that has a 15-tap FFE and equalizes linear distortion.
  • Data c is a simulation result obtained by simulating the light receiving device described in Non-Patent Document 1. However, the light receiving apparatus is provided with a 15-tap FFE, and the neural network has 3 inputs and 64 outputs.
  • Data d is a simulation result obtained by simulating the light receiving device 1 according to the first embodiment.
  • the neural network unit 4 has 3 symbols input, 3 elements in the input layer, 1 intermediate layer, and 12 elements in the output layer.
  • the optical receiving device described in Non-Patent Document 1 has higher receiving sensitivity than an optical receiving device that does not equalize linear distortion or nonlinear distortion.
  • the light receiving device 1 according to the first embodiment has the same reception sensitivity as the light receiving device described in Non-Patent Document 1.
  • the number of elements in the output layer of the neural network unit is reduced from 64 to 12 as compared with the light receiving device described in Non-Patent Document 1. .
  • the neural network unit 4 having the same number of output layers 42-1 to 42-M as the number M of input symbols is the connection between the input symbol and the element Non-linear transformation is performed on the multiplication sum with the load to generate the likelihood for each symbol.
  • the likelihood selection unit 5 selects the likelihood for each output layer from the likelihood generated by the neural network unit 4.
  • the delay adjustment unit 7 converts parallel symbol strings corresponding to the selected likelihood into serial symbol strings.
  • the demodulator 8 demodulates the bit string from the serial symbol string.
  • the hard decision error correction decoding unit 9 performs error correction decoding by hard decision based on the bit string.
  • FIG. 6 is a block diagram showing the configuration of an optical receiver 1A according to Embodiment 2 of the present invention.
  • the light receiving device 1A is a device that receives an optical signal including a plurality of symbols, and includes a photoelectric conversion unit 2, a reception signal adjustment unit 3, a neural network unit 4, a connection weight update unit 10, a bit likelihood generation unit 11, and A soft decision error correction decoding unit 12 is provided.
  • the bit likelihood generation unit 11 generates bitwise likelihood based on the likelihood generated for each symbol by the neural network unit 4. For example, the bit likelihood generation unit 11 inputs the likelihood calculated by the neural network unit 4 according to the above equation (1), converts the inputted likelihood into a normalized amplitude, and generates the likelihood for each bit from the normalized amplitude. Calculate the degree.
  • the soft decision error correction decoding unit 12 performs error correction decoding by soft decision based on the likelihood for each bit input from the bit likelihood generation unit 11. For example, the soft decision error correction decoding unit 12 detects and corrects a bit error generated in the optical transmitter-receiver and the optical transmission line based on the bit assignment rule of the error correction code.
  • Soft decision error correction codes include low density parity check (hereinafter referred to as LDPC) codes.
  • the coupling load updating unit 10 may be a component included in an apparatus provided separately from the light receiving apparatus 1A.
  • the connection load update unit 10 may be a component provided in a learning device provided separately from the light receiving device 1A.
  • FIG. 7 is a flowchart showing an optical reception method according to the second embodiment.
  • the photoelectric conversion unit 2 photoelectrically converts the received light signal, and converts the analog electric signal obtained by the photoelectric conversion into a digital signal (step ST1a).
  • the reception signal adjustment unit 3 converts the digital signal input from the photoelectric conversion unit 2 into a parallel symbol string (step ST2a).
  • the neural network unit 4 non-linearly converts the multiplication sum of the symbol input in parallel from the received signal adjustment unit 3 and the connection weight between the neural elements to generate the likelihood for each symbol.
  • the likelihoods generated for each symbol by the neural network unit 4 are input in parallel to the bit likelihood generation unit 11.
  • the bit likelihood generation unit 11 generates a likelihood for each bit based on the likelihood input from the neural network unit 4 (step ST3a). For example, the bit likelihood generation unit 11 converts the input likelihood into a normalized amplitude, and calculates the likelihood for each bit from the normalized amplitude.
  • the normalized amplitude is an amplitude standardized so that the amplitude level corresponds to the minimum value and the maximum value of the symbol. Since the amplitude of the four-value PAM signal has a value of (-3, -1, 1, 3), the normalized amplitude falls within the range of -3 to 3.
  • FIG. 8A is a graph showing the relationship between the log likelihood of the first bit ⁇ 1 and the normalized amplitude.
  • FIG. 8B is a graph showing the relationship between the log likelihood of the bit ⁇ 2 of the second bit and the normalized amplitude.
  • the bit likelihood generation unit 11 uses the bit likelihood conversion table indicating the correspondence between the log likelihood for each bit and the normalized amplitude as shown in FIG. 8A and FIG. Convert to likelihood.
  • the normalized amplitude is A, the bit of the first bit and beta 1, the bit of the second bit and beta 2, normalized amplitude is A when transmitted as a bit of l-th bit "0"
  • the probability be p r (A
  • ⁇ 1 0)
  • ⁇ 1 1)
  • the log likelihood LLR l corresponding to the bit of the l-th bit can be represented, for example, by the following formula (3).
  • l 1,..., Log 2 D, where D is the modulation multilevel degree of the optical signal.
  • LLR l ln [ ⁇ p r (A
  • ⁇ l 0) ⁇ / ⁇ p r (A
  • ⁇ l 1) ⁇ ] (3)
  • the bit likelihood generation unit 11 refers to, for example, the bit likelihood conversion table in which the log likelihood LLR l calculated for the normalized amplitude A using the equation (3) is registered, Find log likelihood LLR l corresponding to each likelihood.
  • the log likelihood LLR l is ⁇ 1, it is possible that the lth bit is “0”.
  • the log likelihood LLR l is 1, it is possible that the l-th bit is "1".
  • 8A and 8B show bit likelihood conversion tables when four values (00, 01, 11, 10) of the four-value PAM signal are assigned to amplitudes (-3, -1, 1, 3). ing.
  • the log likelihood in bit units obtained from the bit likelihood generation unit 11 is input to the soft decision error correction decoding unit 12.
  • the soft decision error correction decoding unit 12 performs error correction decoding by soft decision on the basis of the likelihood for each bit input from the bit likelihood generation unit 11 (step ST4a). For example, the soft decision error correction decoding unit 12 detects and corrects a bit error generated in the optical transmitter-receiver and the optical transmission line based on the bit assignment rule of the error correction code.
  • the soft decision error correction code includes an LDPC code.
  • the optical receiver 1A according to the second embodiment can reduce the number of elements in the output layer of the neural network unit 4 as compared with the optical receiver described in Non-Patent Document 1. it can. Therefore, in the optical receiver 1A according to the second embodiment, it is possible to simplify the calculation of the neural network unit 4 while maintaining the equalization accuracy of the non-linear distortion.
  • the bit likelihood generation unit 11 may be omitted depending on the configuration of the neural network unit 4.
  • the neural network unit 4 shown in FIG. 6 generates the likelihood for each symbol
  • the configuration of the output layer is changed to generate the likelihood for each bit.
  • the soft decision error correction decoding unit 12 directly inputs the likelihood for each bit from the neural network unit and performs error correction decoding by soft decision.
  • the neural network unit 4 shown in FIGS.
  • the number of elements of is 4 ⁇ B.
  • the configuration of the output layer is changed to generate the likelihood for each bit, the number of elements in the output layer increases from 4 ⁇ B to 4 ⁇ B ⁇ D.
  • the bit likelihood generator 11 is omitted, and therefore, the circuit more than the optical receiver described in Non-Patent Document 1 while maintaining the equalization accuracy of the nonlinear distortion. It is possible to improve the mounting efficiency.
  • the neural network unit 4 having the same number of output layers 42-1 to 42-M as the number M of input symbols is the connection between the input symbols and the elements.
  • the likelihood is generated by non-linear transformation of the multiplication sum with the load.
  • the soft decision error correction decoding unit 12 performs error correction decoding by soft decision based on the likelihood generated by the neural network unit 4. By configuring in this manner, the neural network unit 4 equalizes the nonlinear distortion generated in the optical transmitter-receiver and the optical transmission path.
  • the optical receiver 1A can improve the circuit mounting efficiency while maintaining the equalization accuracy of the non-linear distortion.
  • FIG. 9A is a block diagram showing a hardware configuration for realizing respective functions of the light receiving device 1 according to the first embodiment and the light receiving device 1A according to the second embodiment.
  • FIG. 9B is a block diagram showing a hardware configuration for executing software for realizing the respective functions of the light receiving device 1 according to the first embodiment and the light receiving device 1A according to the second embodiment.
  • the photoelectric converter 100 is a device for converting an input optical signal into an electric signal, and corresponds to the photoelectric conversion unit 2 shown in FIGS. 1 and 6.
  • the photoelectric converter 100 may include an oscillator that oscillates local oscillation light.
  • the respective functions of are realized by the processing circuit.
  • the functions of the received signal adjustment unit 3, the neural network unit 4, the combined load update unit 10, the bit likelihood generation unit 11, and the soft decision error correction decoding unit 12 in the optical receiver 1A are realized by processing circuits. Ru. That is, the optical receiver 1 includes processing circuits for executing each of the plurality of processes in the flowchart shown in FIG. 4, and the optical receiver 1A performs each of the plurality of processes in the flowchart shown in FIG. Processing circuitry for The processing circuit may be dedicated hardware or may be a central processing unit (CPU) that executes a program stored in a memory.
  • CPU central processing unit
  • the processing circuit 101 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an application specific integrated circuit (ASIC), an FPGA (FPGA) Field-Programmable Gate Array) or a combination thereof is applicable.
  • ASIC application specific integrated circuit
  • FPGA Field-Programmable Gate Array
  • the processing circuit is the processor 102 shown in FIG. 9B, the received signal adjustment unit 3, the neural network unit 4, the likelihood selection unit 5, the symbol restoration unit 6, the delay adjustment unit 7, the demodulation unit 8 in the light receiving device 1.
  • the respective functions of the hard decision error correction decoding unit 9 and the combined load updating unit 10 are realized by software, firmware or a combination of software and firmware.
  • each function of the received signal adjustment unit 3, the neural network unit 4, the combined load update unit 10, the bit likelihood generation unit 11, and the soft decision error correction decoding unit 12 in the optical receiver 1A is also software, firmware or software It is realized by the combination of the and the firmware.
  • the software or firmware is written as a program and stored in the memory 103.
  • the processor 102 reads out and executes the program stored in the memory 103, whereby the reception signal adjustment unit 3, the neural network unit 4, the likelihood selection unit 5, the symbol restoration unit 6, and the delay adjustment unit in the light receiving device 1 7, the functions of the demodulation unit 8, the hard decision error correction decoding unit 9, and the combined load updating unit 10 are realized.
  • the optical receiver 1 includes the memory 103 for storing a program that is executed by the processor 102 as a result of each of the series of processes shown in FIG. 4.
  • These programs include the received signal adjustment unit 3, the neural network unit 4, the likelihood selection unit 5, the symbol restoration unit 6, the delay adjustment unit 7, the demodulation unit 8, the hard decision error correction decoding unit 9 and the connection weight update unit 10. Making a computer execute a procedure or method.
  • the processor 102 reads out and executes the program stored in the memory 103, whereby the reception signal adjustment unit 3, the neural network unit 4, the connection weight update unit 10, and the bit likelihood generation unit in the light receiving device 1A. 11 and each function of the soft decision error correction decoding unit 12 are realized. That is, the optical receiver 1A includes the memory 103 for storing a program that is executed by the processor 102 as a result of each of the series of processes shown in FIG. These programs cause a computer to execute the procedure or method of the received signal adjustment unit 3, the neural network unit 4, the connection weight update unit 10, the bit likelihood generation unit 11, and the soft decision error correction decoding unit 12.
  • the memory 103 is, for example, a non-volatile or volatile semiconductor memory such as a random access memory (RAM), a read only memory (ROM), a flash memory, an erasable programmable read only memory (EPROM), and an EEPROM (electrically-EPROM).
  • RAM random access memory
  • ROM read only memory
  • EPROM erasable programmable read only memory
  • EEPROM electrically-EPROM
  • a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, a DVD, etc. correspond.
  • the respective functions of received signal adjustment unit 3, neural network unit 4, likelihood selection unit 5, symbol restoration unit 6, delay adjustment unit 7, demodulation unit 8, hard decision error correction decoding unit 9 and connection weight update unit 10 The part may be realized by dedicated hardware, and a part may be realized by software or firmware.
  • the function of the reception signal adjustment unit 3 is realized by the processing circuit 101 as dedicated hardware.
  • the processor 102 is stored in the memory 103 for the neural network unit 4, likelihood selection unit 5, symbol restoration unit 6, delay adjustment unit 7, demodulation unit 8, hard decision error correction decoding unit 9 and connection weight update unit 10.
  • the function may be realized by executing a program. The same applies to the components of the light receiving device 1A.
  • the processing circuit can realize each of the above functions by hardware, software, firmware, or a combination thereof.
  • the present invention is not limited to the above embodiment, and within the scope of the present invention, variations or embodiments of respective free combinations of the embodiments or respective optional components of the embodiments.
  • An optional component can be omitted in each of the above.
  • the optical receiver according to the present invention can increase circuit mounting efficiency while maintaining the equalization accuracy of nonlinear distortion, and therefore can be used for an optical communication system that handles multilevel optical signals.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Electromagnetism (AREA)
  • Probability & Statistics with Applications (AREA)
  • Theoretical Computer Science (AREA)
  • Optical Communication System (AREA)

Abstract

In the present invention, a neural network unit (4) having the same number of output layers (42-1 through 42-M) as the number (M) of input symbols generates likelihoods for each of the symbols by means of non-linear conversion of a multiply sum of the input symbols and the connection weight between elements. A likelihood selection unit (5) selects likelihoods for each of the output layers from the likelihoods generated by the neural network unit (4). A delay adjustment unit (7) converts parallel symbol strings corresponding to the selected likelihoods into a serial symbol string. A demodulation unit (8) demodulates a bit string from the serial symbol string. A hard decision error correction decoding unit (9) performs hard decision-based error correction decoding on the basis of the bit string.

Description

光受信装置および光受信方法Optical receiving apparatus and optical receiving method
　この発明は、多値の光情報を受信する光受信装置および光受信方法に関する。 The present invention relates to an optical receiving apparatus and an optical receiving method for receiving multilevel optical information.
　近年のクラウドサービスまたは高速移動体通信の普及によって、データセンターなどのローカルエリアネットワーク（以下、ＬＡＮと記載する）で収容される通信トラヒックが逼迫している。ＬＡＮの高速伝送を実現するために、例えば、ＩＥＥＥ８０２．３では、４００Ｇイーサネット（登録商標）の標準化が進められている。この標準化には、信号強度を変化させて多値化するパルス振幅変調（以下、ＰＡＭと記載する）方式の採用が決定している。また、ＰＡＭ方式と合わせてディジタル領域の信号等化器が利用されている。 With the widespread use of cloud services or high-speed mobile communication in recent years, communication traffic accommodated in a local area network (hereinafter referred to as LAN) such as a data center is under pressure. For example, in IEEE 802.3, standardization of 400 G Ethernet (registered trademark) is in progress to realize high-speed transmission of LAN. For this standardization, it has been decided to adopt a pulse amplitude modulation (hereinafter referred to as PAM) method in which the signal strength is changed to multi-value. In addition to the PAM method, a signal equalizer in the digital domain is used.
　信号等化器は、光送受信器および光伝送路に生じたデータ信号の線形歪みを補償する。従来から、簡素な構成の信号等化器としてフィードフォワード等化器（以下、ＦＦＥと記載する）が知られている。ただし、ＦＦＥでは、ＰＡＭ方式を採用した光送受信器に生じる、変調不安定性、二乗検波歪みおよび利得飽和といった非線形歪みに対応できない。 The signal equalizer compensates for linear distortion of the data signal generated in the optical transceiver and the optical transmission line. Conventionally, a feed forward equalizer (hereinafter referred to as FFE) is known as a signal equalizer having a simple configuration. However, FFE can not cope with non-linear distortions such as modulation instability, square wave detection distortion, and gain saturation that occur in an optical transmitter / receiver adopting the PAM method.
　前述したＦＦＥにおける不具合を解消するために、例えば、非特許文献１に記載された光受信装置では、ニューラルネットワークを利用している。ニューラルネットワークは、一般に、中間層でニューロン素子による非線形変換を実施することで、光送受信器および光伝送路に生じた非線形歪みを等化することができる。
　非特許文献１に記載された光受信装置は、複数のシンボルから構成されたブロック単位で信号処理を実施することで、ＰＡＭに対する非線形歪みの等化精度を向上させている。 In order to eliminate the problem in the above-mentioned FFE, for example, the light receiving device described in Non-Patent Document 1 uses a neural network. In general, a neural network can equalize nonlinear distortions occurring in an optical transmitter-receiver and an optical transmission path by performing nonlinear conversion by neuron elements in the intermediate layer.
The optical receiver described in Non-Patent Document 1 improves the equalization accuracy of non-linear distortion with respect to PAM by performing signal processing in block units composed of a plurality of symbols.
　しかしながら、非特許文献１に記載されるニューラルネットワークでは、光信号の変調多値度をＤとし、入力されたシンボルの数をＭとした場合、出力層でＤＭ個の素子が必要となる。このように、非特許文献１に記載された光受信装置は、入力シンボル数に対して回路規模が指数関数的に増加するという課題があった。 However, in the neural network described in Non-Patent Document 1, when the modulation multilevel degree of the optical signal is D and the number of input symbols is M, D M elements are required in the output layer. Thus, the optical receiver described in Non-Patent Document 1 has a problem that the circuit scale increases exponentially with the number of input symbols.
　この発明は上記課題を解決するものであり、非線形歪みの等化精度を維持しつつ、回路実装効率を高めることができる光受信装置および光受信方法を得ることを目的とする。 An object of the present invention is to solve the above-mentioned problems, and it is an object of the present invention to provide an optical receiving apparatus and an optical receiving method capable of enhancing the circuit mounting efficiency while maintaining the equalization accuracy of nonlinear distortion.
　この発明に係る光受信装置は、複数のシンボルを含む光信号を受信する光受信装置であって、光電変換部、第１の調整部、ニューラルネットワーク部、尤度選択部、第２の調整部、復調部、および第１の誤り訂正復号部を備える。光電変換部は、受信された光信号を電気信号に変換する。第１の調整部は、光電変換部によって光信号から変換された電気信号を並列のシンボル列に変換する。ニューラルネットワーク部は、入力層、１段または複数段の中間層、および入力層に入力されるシンボル数と同数の出力層を有し、第１の調整部から並列に入力されたシンボルと素子間の結合荷重との乗算和を非線形変換してシンボルごとの尤度を生成する。尤度選択部は、ニューラルネットワーク部によって生成された尤度から、出力層ごとに尤度を選択する。第２の調整部は、尤度選択部によって選択された尤度に対応する並列のシンボル列を、直列のシンボル列に変換する。復調部は、第２の調整部から入力したシンボル列からビット列を復調する。第１の誤り訂正復号部は、復調部によって復調されたビット列に基づいて、硬判定による誤り訂正復号を行う。 An optical receiving apparatus according to the present invention is an optical receiving apparatus that receives an optical signal including a plurality of symbols, and includes a photoelectric conversion unit, a first adjustment unit, a neural network unit, a likelihood selection unit, and a second adjustment unit. , A demodulation unit, and a first error correction decoding unit. The photoelectric conversion unit converts the received optical signal into an electrical signal. The first adjustment unit converts the electrical signal converted from the light signal by the photoelectric conversion unit into a parallel symbol string. The neural network unit has an input layer, one or more intermediate layers, and output layers equal in number to the number of symbols input to the input layer, and the symbols and elements input in parallel from the first adjustment unit are connected The non-linear transformation of the multiplication sum with the combined weight of to generate the likelihood for each symbol. The likelihood selection unit selects a likelihood for each output layer from the likelihoods generated by the neural network unit. The second adjustment unit converts parallel symbol strings corresponding to the likelihoods selected by the likelihood selection unit into serial symbol strings. The demodulation unit demodulates the bit string from the symbol string input from the second adjustment unit. The first error correction decoding unit performs error correction decoding by hard decision based on the bit string demodulated by the demodulation unit.
　この発明によれば、光送受信器および光伝送路に生じた非線形歪みを等化するニューラルネットワーク部が、入力シンボル数と同数の出力層を有するので、出力層の素子数が、光信号の変調多値度と入力シンボル数とを乗算した数に抑えられる。これにより、光受信装置は、非線形歪みの等化精度を維持しつつ、回路実装効率を高めることができる。 According to the present invention, since the optical transmitter / receiver and the neural network unit for equalizing the nonlinear distortion generated in the optical transmission path have the same number of output layers as the number of input symbols, the number of elements in the output layer is not the modulation of the optical signal. It can be reduced to the number obtained by multiplying the multi-value degree by the number of input symbols. As a result, the light receiving device can increase the circuit mounting efficiency while maintaining the equalization accuracy of the non-linear distortion.
この発明の実施の形態１に係る光受信装置の構成を示すブロック図である。It is a block diagram which shows the structure of the optical receiver which concerns on Embodiment 1 of this invention. 実施の形態１におけるニューラルネットワーク部の構成を示すブロック図である。FIG. 1 is a block diagram showing a configuration of a neural network unit in Embodiment 1. 実施の形態１における尤度選択部の構成を示すブロック図である。FIG. 5 is a block diagram showing a configuration of a likelihood selection unit in Embodiment 1. 実施の形態１に係る光受信方法を示すフローチャートである。3 is a flowchart showing an optical reception method according to Embodiment 1. FIG. ビット誤り率と受信平均光パワーとの関係をシミュレーションした結果を示すグラフである。It is a graph which shows the result of having simulated the relationship between a bit error rate and receiving average optical power. この発明の実施の形態２に係る光受信装置の構成を示すブロック図である。It is a block diagram which shows the structure of the optical receiver which concerns on Embodiment 2 of this invention. 実施の形態２に係る光受信方法を示すフローチャートである。7 is a flowchart showing an optical reception method according to Embodiment 2. FIG. 図８Ａは、１ビット目のビットの対数尤度と規格化振幅との関係を示すグラフである。図８Ｂは、２ビット目のビットの対数尤度と規格化振幅との関係を示すグラフである。FIG. 8A is a graph showing the relationship between the log likelihood of the first bit and the normalized amplitude. FIG. 8B is a graph showing the relationship between the log likelihood of the second bit and the normalized amplitude. 図９Ａは、実施の形態１および実施の形態２に係る光受信装置のそれぞれの機能を実現するハードウェア構成を示すブロック図である。図９Ｂは、実施の形態１および実施の形態２に係る光受信装置のそれぞれの機能を実現するソフトウェアを実行するハードウェア構成を示すブロック図である。FIG. 9A is a block diagram showing a hardware configuration for realizing the respective functions of the optical receiving apparatus according to Embodiment 1 and Embodiment 2. FIG. 9B is a block diagram showing a hardware configuration that executes software for realizing the respective functions of the optical reception devices according to Embodiment 1 and Embodiment 2.
　以下、この発明をより詳細に説明するため、この発明を実施するための形態について、添付の図面に従って説明する。
実施の形態１．
　図１は、この発明の実施の形態１に係る光受信装置１の構成を示すブロック図である。図２は、ニューラルネットワーク部４の構成を示すブロック図である。図３は、尤度選択部５の構成を示すブロック図である。図１において、光受信装置１は、複数のシンボルを含む光信号を受信する装置であり、光電変換部２、受信信号調整部３、ニューラルネットワーク部４、尤度選択部５、シンボル復元部６、遅延調整部７、復調部８、硬判定誤り訂正復号部９および結合荷重更新部１０を備える。 Hereinafter, in order to explain the present invention in more detail, embodiments for carrying out the present invention will be described according to the attached drawings.
Embodiment 1
FIG. 1 is a block diagram showing the configuration of an optical receiver 1 according to Embodiment 1 of the present invention. FIG. 2 is a block diagram showing the configuration of the neural network unit 4. FIG. 3 is a block diagram showing the configuration of the likelihood selection unit 5. In FIG. 1, the light receiving device 1 is a device that receives an optical signal including a plurality of symbols, and the photoelectric conversion unit 2, the reception signal adjustment unit 3, the neural network unit 4, the likelihood selection unit 5, and the symbol restoration unit 6. And a delay adjustment unit 7, a demodulation unit 8, a hard decision error correction decoding unit 9, and a connection weight update unit 10.
　光電変換部２は、受信された光信号を光電変換することにより、光信号強度に比例したアナログ形式の電気信号に変換する。なお、光電変換部２は、局部発振光を発振する発振器を備えてもよい。さらに、光電変換部２は、アナログ形式の電気信号を検波してから、ディジタル形式の電気信号（以下、ディジタル信号と記載する）に変換する。光電変換部２によって生成された上記ディジタル信号は、受信信号調整部３に出力される。 The photoelectric conversion unit 2 photoelectrically converts the received optical signal to convert it into an analog electrical signal proportional to the optical signal intensity. The photoelectric conversion unit 2 may include an oscillator that oscillates the local oscillation light. Furthermore, the photoelectric conversion unit 2 detects an electrical signal of analog type, and converts it into an electrical signal of digital type (hereinafter referred to as digital signal). The digital signal generated by the photoelectric conversion unit 2 is output to the reception signal adjustment unit 3.
　受信信号調整部３は、光電変換部２から入力した上記ディジタル信号を、電気的処理を用いて並列のシンボル列に変換する第１の調整部である。さらに、受信信号調整部３は、光電変換部２から入力した上記ディジタル信号のサンプリングレートを調整する。 The reception signal adjustment unit 3 is a first adjustment unit that converts the digital signal input from the photoelectric conversion unit 2 into a parallel symbol string using electrical processing. Further, the reception signal adjustment unit 3 adjusts the sampling rate of the digital signal input from the photoelectric conversion unit 2.
　ニューラルネットワーク部４は、図２に示すように、入力層４０、Ｎ段の中間層４１、および出力層４２－１～４２－Ｍを有する。なお、ＮおよびＭは、１以上の整数である。ニューラルネットワーク部４には、受信信号調整部３によってサンプリングレートが調整された並列のシンボル列が入力される。図２に示すニューラルネットワーク部４には１～ｌ個のシンボルが入力されている。 The neural network unit 4 has an input layer 40, N middle layers 41, and output layers 42-1 to 42-M, as shown in FIG. N and M are integers of 1 or more. A parallel symbol string whose sampling rate has been adjusted by the received signal adjustment unit 3 is input to the neural network unit 4. The neural network unit 4 shown in FIG. 2 receives 1 to 1 symbols.
　シンボルは、入力層４０から中間層４１へ出力され、中間層４１から出力層４２－１～４２－Ｍへ出力される。このとき、ニューラルネットワーク部４では、シンボルと各層のニューロン素子間の結合荷重との乗算和を非線形変換して尤度を算出する。非線形変換に利用される非線形関数には、ステップ関数、シグモイド関数、およびランプ関数といったものがある。 The symbols are output from the input layer 40 to the intermediate layer 41, and output from the intermediate layer 41 to the output layers 42-1 to 42-M. At this time, the neural network unit 4 non-linearly transforms the multiplication sum of the symbol and the connection weight between the neuron elements in each layer to calculate the likelihood. Non-linear functions utilized for non-linear transformation include step functions, sigmoid functions, and ramp functions.
　ニューラルネットワーク部４の出力層４２－１～４２－Ｍは、入力層４０に入力されるシンボルの数と同数以上であって、シンボルごとに尤度を出力する。図２に示す例では、入力シンボル数ｌ＝Ｍであると、尤度を出力する出力層の数はＭ個となる。
　変調多値度Ｄは、１シンボル当たりのビット数であり、４値ＰＡＭ信号の変調多値度Ｄは“４”である。例えば、ニューラルネットワーク部４に対して、４値ＰＡＭ信号のＢ個のシンボルが入力された場合、尤度を出力する出力層の数はＢ個となり、Ｂ個の出力層のそれぞれの素子数ｎはＤ（＝４）個となり、Ｂ個の出力層における総素子数は、４×Ｂ個となる。ただし、出力層４２－１～４２－Ｍは、順序不同である。 The output layers 42-1 to 42-M of the neural network unit 4 have the same number or more as the number of symbols input to the input layer 40, and output likelihood for each symbol. In the example shown in FIG. 2, when the number of input symbols l = M, the number of output layers that output likelihood is M.
The modulation level D is the number of bits per symbol, and the modulation level D of the 4-value PAM signal is "4". For example, when B symbols of a 4-value PAM signal are input to the neural network unit 4, the number of output layers for outputting the likelihood is B, and the number n of elements of each of the B output layers is n. Is D (= 4), and the total number of elements in the B output layers is 4 × B. However, the output layers 42-1 to 42-M are out of order.
　尤度選択部５は、ニューラルネットワーク部４によってシンボルごとに生成された尤度から、出力層ごとに尤度を選択する。例えば、尤度選択部５は、図３に示すように、最大値演算部５０－１～５０－Ｍおよび要素番号抽出部５１－１～５１－Ｍを備える。
　最大値演算部５０－１～５０－Ｍは、ニューラルネットワーク部４によって生成されたシンボルごとの尤度から、出力層ごとに尤度の最大値を選択する。要素番号抽出部５１－１～５１－Ｍは、最大値演算部５０－１～５０－Ｍから入力した最大値の尤度に対応する要素番号を抽出する。 The likelihood selection unit 5 selects the likelihood for each output layer from the likelihood generated for each symbol by the neural network unit 4. For example, as shown in FIG. 3, the likelihood selection unit 5 includes maximum value calculation units 50-1 to 50-M and element number extraction units 51-1 to 51-M.
The maximum value calculation units 50-1 to 50-M select the maximum value of likelihood for each output layer from the likelihood for each symbol generated by the neural network unit 4. The element number extraction units 51-1 to 51-M extract element numbers corresponding to the likelihood of the maximum value input from the maximum value calculation units 50-1 to 50-M.
　シンボル復元部６は、尤度選択部５によって出力層ごとに選択された尤度からシンボルを復元し、復元したシンボルから構成される並列のシンボル列を出力する。
　シンボル復元部６は、要素番号抽出部５１－１～５１－Ｍから入力した要素番号に対応するシンボルを生成する。例えば、４値位相変調信号を復調対象としたシンボルの復元において、シンボル復元部６は、要素番号（０，１，２，３）が入力された場合、要素番号（０，１，２，３）を、シンボル（１＋ｉ，－１＋ｉ，－１－ｉ，１－ｉ）に変換する。なお、ｉは虚数であり、ｉ＝√－１である。 The symbol restoration unit 6 restores the symbol from the likelihood selected for each output layer by the likelihood selection unit 5 and outputs a parallel symbol string composed of the restored symbol.
The symbol restoration unit 6 generates a symbol corresponding to the element number input from the element number extraction units 51-1 to 51-M. For example, in the recovery of a symbol for which a four-level phase modulation signal is to be demodulated, the symbol recovery unit 6 receives an element number (0, 1, 2, 3) when the element number (0, 1, 2, 3) is input. ) Is converted to the symbol (1 + i, −1 + i, -1-i, 1−i). Here, i is an imaginary number and i = √−1.
　図１には、シンボル復元部６を備えた構成が記載されているが、尤度選択部５が、出力素子番号ｋとシンボルとを対応させてシンボルを出力することにより、光受信装置１からシンボル復元部６を省略することができる。すなわち、尤度選択部５が、出力層ごとに選択した尤度に対応する並列のシンボル列を出力する場合、光受信装置１は、シンボル復元部６を備えない構成であってもよい。 Although FIG. 1 shows the configuration provided with the symbol recovery unit 6, the likelihood selection unit 5 correlates the output element number k with the symbol and outputs the symbol from the light receiving device 1 The symbol restoration unit 6 can be omitted. That is, when the likelihood selection unit 5 outputs a parallel symbol string corresponding to the likelihood selected for each output layer, the light receiving device 1 may not include the symbol recovery unit 6.
　遅延調整部７は、尤度選択部５によって選択された尤度に対応する並列のシンボル列を直列のシンボル列に変換する第２の調整部である。遅延調整部７は、シンボル復元部６からシンボル列を並列に入力して、入力したシンボル列を直列に変換する。 The delay adjusting unit 7 is a second adjusting unit that converts a parallel symbol string corresponding to the likelihood selected by the likelihood selecting unit 5 into a serial symbol string. The delay adjusting unit 7 inputs the symbol string in parallel from the symbol restoring unit 6 and converts the input symbol string in series.
　復調部８は、遅延調整部７から入力したシンボル列からビット列を復調する。例えば、復調部８は、シンボルからビットへ変化するための割当規則に従って、入力したシンボルをビット列へ変換する。 The demodulation unit 8 demodulates a bit string from the symbol string input from the delay adjustment unit 7. For example, the demodulator 8 converts the input symbol into a bit string in accordance with an assignment rule for changing from symbol to bit.
　硬判定誤り訂正復号部９は、復調部８によって復調されたビット列に基づいて、硬判定による誤り訂正復号を行う第１の誤り訂正復号部である。
　例えば、硬判定誤り訂正復号部９は、誤り訂正符号のビット割当規則に基づいて、光送受信器または光伝送路で生じたビット誤りを検出して訂正する。硬判定の誤り訂正符号としては、リードソロモン符号などが挙げられる。 The hard decision error correction decoding unit 9 is a first error correction decoding unit that performs error correction decoding based on hard decision based on the bit string demodulated by the demodulation unit 8.
For example, the hard decision error correction decoding unit 9 detects and corrects a bit error generated in the optical transmitter-receiver or the optical transmission line based on the bit assignment rule of the error correction code. Hard-decision error correction codes include Reed-Solomon codes and the like.
　結合荷重更新部１０は、ニューラルネットワーク部４の結合荷重を更新する。例えば、結合荷重更新部１０は、シンボル復元部６によって復元されたシンボルと学習値との誤差が最小になるように、ニューラルネットワーク部４の結合荷重を更新する。結合荷重更新には、二乗誤差を最小化する誤差逆伝搬法が広く用いられる。 The connection load update unit 10 updates the connection load of the neural network unit 4. For example, the connection weight update unit 10 updates the connection load of the neural network unit 4 so that the error between the symbol restored by the symbol restoration unit 6 and the learning value is minimized. An error back-propagation method that minimizes the squared error is widely used for coupling weight updating.
　なお、結合荷重更新部１０は、光受信装置１とは別に設けられた装置が備える構成要素であってもよい。例えば、結合荷重更新部１０は、光受信装置１とは別に設けられた学習装置が備える構成要素であってもよい。この場合、学習装置は、尤度選択部５またはシンボル復元部６から入力したシンボルを用いて、ニューラルネットワーク部４の結合荷重を更新する。 The connection load updating unit 10 may be a component included in an apparatus provided separately from the light receiving apparatus 1. For example, the coupling load updating unit 10 may be a component provided in a learning device provided separately from the light receiving device 1. In this case, the learning device updates the connection weight of the neural network unit 4 using the symbol input from the likelihood selection unit 5 or the symbol restoration unit 6.
　次に動作について説明する。
　図４は、実施の形態１に係る光受信方法を示すフローチャートである。
　まず、光電変換部２が、受信された光信号を光電変換し、光電変換で得られたアナログ形式の電気信号をディジタル信号に変換する（ステップＳＴ１）。
　受信信号調整部３が、光電変換部２から入力したディジタル信号を、並列のシンボル列に変換する（ステップＳＴ２）。 Next, the operation will be described.
FIG. 4 is a flowchart showing the light receiving method according to the first embodiment.
First, the photoelectric conversion unit 2 photoelectrically converts the received optical signal, and converts the analog electric signal obtained by the photoelectric conversion into a digital signal (step ST1).
The reception signal adjustment unit 3 converts the digital signal input from the photoelectric conversion unit 2 into a parallel symbol string (step ST2).
　続いて、ニューラルネットワーク部４が、受信信号調整部３から並列に入力されたシンボルとニューラル素子間の結合荷重との乗算和を非線形変換して、シンボルごとの尤度を生成する（ステップＳＴ３）。ニューラルネットワーク部４からは、正答シンボルに対応する尤度が尤度選択部５へ出力される。 Subsequently, the neural network unit 4 non-linearly converts the multiplication sum of the symbol input in parallel from the received signal adjustment unit 3 and the connection weight between the neural elements to generate the likelihood for each symbol (step ST3). . The neural network unit 4 outputs the likelihood corresponding to the correct answer symbol to the likelihood selection unit 5.
　例えば、ｊ番目の出力層におけるニューロン素子に出力素子番号ｋが付与されており、入力層に入力されるシンボル群のシンボル数をφとし、正答クラスをＣ１とし、出力層の数をＭとし、変調多値度をＤとする。この場合、ｊ番目の出力層において、出力素子番号ｋのニューロン素子から正答クラスＣ１が得られたときの事後確率は、ｐｊｋ（Ｃ１｜φ）となる。 For example, the output element number k is given to the neuron element in the j-th output layer, the symbol number of the symbol group input to the input layer is φ, the correct answer class is C 1, and the number of output layers is M. , Let D be the modulation level. In this case, the posterior probability is p jk (C 1 | φ) when the correct answer class C 1 is obtained from the neuron element of output element number k in the j-th output layer.
　ｊ番目の出力層において出力素子番号ｋで特定されるニューロン素子から出力される、正答クラスＣ１に属するシンボルに対応する尤度Ｌｊｋは、例えば、下記式（１）を用いて算出される。ただし、ｊ＝１，・・・，Ｍであり、ｋ＝０，・・・，Ｄ－１である。
　Ｌｊｋ＝－ｌｎ［ｐｊｋ（Ｃ１｜φ）］　　　・・・（１） The likelihood L jk corresponding to the symbol belonging to the correct answer class C 1 output from the neuron element specified by the output element number k in the j-th output layer is calculated, for example, using the following equation (1) . However, j = 1,..., M and k = 0,.
L jk = −ln [p jk (C 1 | φ)] (1)
　尤度選択部５は、ニューラルネットワーク部４によってシンボルごとに生成された尤度から、出力層ごとに尤度を選択する（ステップＳＴ４）。図３に示すように、最大値演算部５０－１～５０－Ｍは、出力層におけるｎ個のニューロン素子のそれぞれから出力された尤度を並列に入力して、ｎ個の尤度のうちの最大値を選択する。最大値演算部５０－１～５０－Ｍのそれぞれは、例えば、下記式（２）に従って、ｊ番目の出力層において出力素子番号ｋで特定されるニューロン素子から出力される尤度の最大値ＭＬｊを算出する。ただし、ｋ＝０，・・・，Ｄ－１である。
　ＭＬｊ＝ｍａｘｋ（Ｌｊｋ）　　　　　　　・・・（２） The likelihood selection unit 5 selects the likelihood for each output layer from the likelihood generated for each symbol by the neural network unit 4 (step ST4). As shown in FIG. 3, the maximum value calculation units 50-1 to 50-M input in parallel the likelihoods output from each of the n neuron elements in the output layer, and among the n likelihoods, Choose the maximum value of. Each of the maximum value calculation units 50-1 to 50-M is, for example, the maximum value ML of the likelihood output from the neuron element specified by the output element number k in the j-th output layer according to the following equation (2) Calculate j . Here, k = 0,..., D−1.
ML j = max k (L jk ) (2)
　ｊ番目の最大値演算部によってｊ番目の出力層から選択された尤度の最大値ＭＬｊは、ｊ番目の要素番号抽出部へ出力される。ｊ番目の要素番号抽出部は、最大値ＭＬｊの尤度に対応する要素番号を抽出してシンボル復元部６に出力する。なお、要素番号は、ｊ番目の出力層におけるニューロン素子を特定する出力素子番号ｋのうちのいずれかの番号である。シンボル復元部６は、要素番号抽出部５１－１～５１－Ｍのそれぞれから入力された要素番号に対応するシンボルを生成し、生成したシンボルのシンボル列を遅延調整部７に出力する。これにより、尤度選択部５によって選択された尤度に対応する並列のシンボル列が、遅延調整部７に入力される。 The maximum value ML j of the likelihood selected from the j-th output layer by the j-th maximum value operation unit is output to the j-th element number extraction unit. The j-th element number extraction unit extracts an element number corresponding to the likelihood of the maximum value ML j and outputs the element number to the symbol recovery unit 6. The element number is any one of the output element numbers k for specifying the neuron element in the j-th output layer. The symbol restoration unit 6 generates a symbol corresponding to the element number input from each of the element number extraction units 51-1 to 51-M, and outputs the generated symbol symbol sequence to the delay adjustment unit 7. Thereby, the parallel symbol string corresponding to the likelihood selected by the likelihood selection unit 5 is input to the delay adjustment unit 7.
　遅延調整部７は、尤度選択部５によって選択された尤度に対応する並列のシンボル列を直列のシンボル列に変換する（ステップＳＴ５）。遅延調整部７によって直列に変換されたシンボル列は、復調部８に出力される。 The delay adjusting unit 7 converts the parallel symbol string corresponding to the likelihood selected by the likelihood selecting unit 5 into a serial symbol string (step ST5). The symbol string serially converted by the delay adjustment unit 7 is output to the demodulation unit 8.
　復調部８は、遅延調整部７から入力したシンボル列からビット列を復調する（ステップＳＴ６）。復調部８によって復調されたビット列は、硬判定誤り訂正復号部９に出力される。硬判定誤り訂正復号部９は、復調部８によって復調されたビット列に基づいて、硬判定による誤り訂正復号を行う（ステップＳＴ７）。 The demodulation unit 8 demodulates the bit string from the symbol string input from the delay adjustment unit 7 (step ST6). The bit string demodulated by the demodulator 8 is output to the hard decision error correction decoder 9. The hard decision error correction decoding unit 9 performs error correction decoding by hard decision based on the bit string demodulated by the demodulation unit 8 (step ST7).
　前述したように、光受信装置１では、ニューラルネットワーク部４による尤度の計算に非線形関数が利用される。これにより、光送受信器および光伝送路で生じる非線形歪みが打ち消される。さらに、非特許文献１に記載された光受信装置では、尤度を算出するために、変調多値度を入力シンボル数でべき乗した数の計算が必要である。これにより、出力層の素子数が膨大な数となり回路実装効率が悪かった。 As described above, in the optical receiver 1, a non-linear function is used to calculate the likelihood by the neural network unit 4. This cancels out non-linear distortion generated in the optical transmitter-receiver and the optical transmission line. Furthermore, in the optical receiving device described in Non-Patent Document 1, in order to calculate the likelihood, it is necessary to calculate the number obtained by raising the modulation multi-level degree by the number of input symbols. As a result, the number of elements in the output layer becomes enormous and the circuit mounting efficiency is poor.
　これに対し、実施の形態１におけるニューラルネットワーク部４は、入力シンボルごとに出力層が配置される。このため、ニューラルネットワーク部４では、変調多値度と入力シンボル数を乗算した数の計算を行うだけで、シンボルごとの尤度を生成することが可能である。従って、光受信装置１は、非特許文献１に記載された光受信装置よりも高い回路実装効率を実現することができる。 On the other hand, in the neural network unit 4 in the first embodiment, an output layer is arranged for each input symbol. For this reason, the neural network unit 4 can generate the likelihood for each symbol only by calculating the number obtained by multiplying the modulation multi-value and the number of input symbols. Therefore, the light receiving device 1 can realize higher circuit mounting efficiency than the light receiving device described in Non-Patent Document 1.
　図５は、ビット誤り率と受信平均光パワー（ｄＢｍ）との関係をシミュレーションした結果を示すグラフである。図５では、光送信器からビットレートが１００Ｇｂｉｔ／ｓの４値ＰＡＭ信号がファイバ伝送なしで受信器に入力されたものと仮定している。さらに、光受信器が備える光検出器は、入力した光信号を二乗検波する。二乗検波によって検出された電気信号は、１５タップのフィードフォワード等化器（以下、ＦＦＥと記載する）を通過してから、光受信装置１に入力される。ビット誤り率は、復調部８によってシンボルに基づいて算出される。硬判定誤り訂正復号部９は、前方誤り訂正限界を２×１０－４と設定することで、その動作を模擬している。 FIG. 5 is a graph showing the simulation result of the relationship between the bit error rate and the received average optical power (dBm). In FIG. 5, it is assumed that a 4-value PAM signal with a bit rate of 100 Gbit / s is input from the optical transmitter to the receiver without fiber transmission. Furthermore, the photodetector provided in the optical receiver performs square detection on the input optical signal. The electric signal detected by the square wave detection is input to the light receiving device 1 after passing through a 15-tap feed forward equalizer (hereinafter referred to as FFE). The bit error rate is calculated by the demodulator 8 based on the symbols. The hard decision error correction decoding unit 9 simulates the operation by setting the forward error correction limit to 2 × 10 −4 .
　図５において、データａは、ニューラルネットワークを利用せず、線形歪みと非線形歪みを等化しない構成の光受信装置を模擬して得られたシミュレーション結果である。データｂは、１５タップのＦＦＥを備え、線形歪みを等化する光受信装置を模擬して得られたシミュレーション結果である。データｃは、非特許文献１に記載された光受信装置を模擬して得られたシミュレーション結果である。ただし、当該光受信装置は、１５タップのＦＦＥを備えており、ニューラルネットワークは、３入力、６４出力である。データｄは、実施の形態１に係る光受信装置１を模擬して得られたシミュレーション結果である。ニューラルネットワーク部４は、入力されるシンボル数が３シンボルであり、入力層の素子数が３であり、中間層が１段であり、出力層の素子数が１２であるものとする。 In FIG. 5, data a is a simulation result obtained by simulating an optical receiving apparatus having a configuration in which linear distortion and non-linear distortion are not equalized without using a neural network. Data b is a simulation result obtained by simulating an optical receiving apparatus that has a 15-tap FFE and equalizes linear distortion. Data c is a simulation result obtained by simulating the light receiving device described in Non-Patent Document 1. However, the light receiving apparatus is provided with a 15-tap FFE, and the neural network has 3 inputs and 64 outputs. Data d is a simulation result obtained by simulating the light receiving device 1 according to the first embodiment. The neural network unit 4 has 3 symbols input, 3 elements in the input layer, 1 intermediate layer, and 12 elements in the output layer.
　データａ、データｂおよびデータｃから明らかなように、非特許文献１に記載された光受信装置は、線形歪みまたは非線形歪みを等化しない光受信装置に比べて高い受信感度を有している。データｃとデータｄを比較すると、実施の形態１に係る光受信装置１では、非特許文献１に記載された光受信装置と同等の受信感度を有している。
　一方、実施の形態１に係る光受信装置１では、非特許文献１に記載された光受信装置と比較して、ニューラルネットワーク部の出力層の素子数が６４個から１２個に削減されている。このように、実施の形態１に係る光受信装置１では、非線形歪みの等化精度を維持しつつ、ニューラルネットワーク部４の計算を簡素化することが可能である。 As apparent from data a, data b and data c, the optical receiving device described in Non-Patent Document 1 has higher receiving sensitivity than an optical receiving device that does not equalize linear distortion or nonlinear distortion. . When data c and data d are compared, the light receiving device 1 according to the first embodiment has the same reception sensitivity as the light receiving device described in Non-Patent Document 1.
On the other hand, in the light receiving device 1 according to the first embodiment, the number of elements in the output layer of the neural network unit is reduced from 64 to 12 as compared with the light receiving device described in Non-Patent Document 1. . As described above, in the optical receiver 1 according to the first embodiment, it is possible to simplify the calculation of the neural network unit 4 while maintaining the equalization accuracy of the non-linear distortion.
　以上のように、実施の形態１に係る光受信装置１において、入力シンボル数Ｍと同数の出力層４２－１～４２－Ｍを有するニューラルネットワーク部４が、入力されたシンボルと素子間の結合荷重との乗算和を非線形変換してシンボルごとの尤度を生成する。尤度選択部５が、ニューラルネットワーク部４によって生成された尤度から出力層ごとに尤度を選択する。遅延調整部７が、選択された尤度に対応する並列のシンボル列を直列のシンボル列に変換する。復調部８が、直列のシンボル列からビット列を復調する。硬判定誤り訂正復号部９が、ビット列に基づいて、硬判定による誤り訂正復号を行う。
　このように構成することで、ニューラルネットワーク部４が、光送受信器および光伝送路に生じた非線形歪みを等化する。出力層４２－１～４２－Ｍが、入力シンボル数Ｍと同数であるので、出力層ごとのニューロン素子数が、光信号の変調多値度Ｄと入力シンボル数Ｍとを乗算した個数（＝Ｄ×Ｍ）に抑えられる。これにより、光受信装置１は、非線形歪みの等化精度を維持しつつ、回路実装効率を高めることができる。 As described above, in the light receiving device 1 according to the first embodiment, the neural network unit 4 having the same number of output layers 42-1 to 42-M as the number M of input symbols is the connection between the input symbol and the element Non-linear transformation is performed on the multiplication sum with the load to generate the likelihood for each symbol. The likelihood selection unit 5 selects the likelihood for each output layer from the likelihood generated by the neural network unit 4. The delay adjustment unit 7 converts parallel symbol strings corresponding to the selected likelihood into serial symbol strings. The demodulator 8 demodulates the bit string from the serial symbol string. The hard decision error correction decoding unit 9 performs error correction decoding by hard decision based on the bit string.
By configuring in this manner, the neural network unit 4 equalizes the nonlinear distortion generated in the optical transmitter-receiver and the optical transmission path. Since the number of output layers 42-1 to 42-M is the same as the number of input symbols M, the number of neuron elements for each output layer is the number obtained by multiplying the modulation multilevel D of the optical signal by the number of input symbols M (= It can be suppressed to D × M). Thereby, the optical receiver 1 can improve the circuit mounting efficiency while maintaining the equalization accuracy of the nonlinear distortion.
実施の形態２．
　図６は、この発明の実施の形態２に係る光受信装置１Ａの構成を示すブロック図である。図６において、図１と同一の構成要素には同一の符号を付して説明を省略する。光受信装置１Ａは、複数のシンボルを含む光信号を受信する装置であり、光電変換部２、受信信号調整部３、ニューラルネットワーク部４、結合荷重更新部１０、ビット尤度生成部１１、および軟判定誤り訂正復号部１２を備える。 Second Embodiment
FIG. 6 is a block diagram showing the configuration of an optical receiver 1A according to Embodiment 2 of the present invention. In FIG. 6, the same components as in FIG. 1 are assigned the same reference numerals and explanation thereof is omitted. The light receiving device 1A is a device that receives an optical signal including a plurality of symbols, and includes a photoelectric conversion unit 2, a reception signal adjustment unit 3, a neural network unit 4, a connection weight update unit 10, a bit likelihood generation unit 11, and A soft decision error correction decoding unit 12 is provided.
　ビット尤度生成部１１は、ニューラルネットワーク部４によってシンボルごとに生成された尤度に基づいて、ビットごとの尤度を生成する。例えば、ビット尤度生成部１１は、上記式（１）に従ってニューラルネットワーク部４により算出された尤度を入力し、入力した尤度を規格化振幅に変換し、規格化振幅からビットごとの尤度を算出する。 The bit likelihood generation unit 11 generates bitwise likelihood based on the likelihood generated for each symbol by the neural network unit 4. For example, the bit likelihood generation unit 11 inputs the likelihood calculated by the neural network unit 4 according to the above equation (1), converts the inputted likelihood into a normalized amplitude, and generates the likelihood for each bit from the normalized amplitude. Calculate the degree.
　軟判定誤り訂正復号部１２は、ビット尤度生成部１１から入力したビットごとの尤度に基づいて、軟判定による誤り訂正復号を行う。例えば、軟判定誤り訂正復号部１２は、誤り訂正符号のビット割当規則に基づいて、光送受信器および光伝送路で生じたビット誤りを検出して訂正する。軟判定誤り訂正符号には、低密度パリティ検査（以下、ＬＤＰＣと記載する）符号が挙げられる。 The soft decision error correction decoding unit 12 performs error correction decoding by soft decision based on the likelihood for each bit input from the bit likelihood generation unit 11. For example, the soft decision error correction decoding unit 12 detects and corrects a bit error generated in the optical transmitter-receiver and the optical transmission line based on the bit assignment rule of the error correction code. Soft decision error correction codes include low density parity check (hereinafter referred to as LDPC) codes.
　なお、結合荷重更新部１０は、光受信装置１Ａとは別に設けられた装置が備える構成要素であってもよい。例えば、結合荷重更新部１０は、光受信装置１Ａとは別に設けられた学習装置が備える構成要素であってもよい。 The coupling load updating unit 10 may be a component included in an apparatus provided separately from the light receiving apparatus 1A. For example, the connection load update unit 10 may be a component provided in a learning device provided separately from the light receiving device 1A.
　次に動作について説明する。
　図７は、実施の形態２に係る光受信方法を示すフローチャートである。
　まず、光電変換部２が、受信された光信号を光電変換し、光電変換で得られたアナログ形式の電気信号をディジタル信号に変換する（ステップＳＴ１ａ）。
　受信信号調整部３が、光電変換部２から入力したディジタル信号を、並列のシンボル列に変換する（ステップＳＴ２ａ）。 Next, the operation will be described.
FIG. 7 is a flowchart showing an optical reception method according to the second embodiment.
First, the photoelectric conversion unit 2 photoelectrically converts the received light signal, and converts the analog electric signal obtained by the photoelectric conversion into a digital signal (step ST1a).
The reception signal adjustment unit 3 converts the digital signal input from the photoelectric conversion unit 2 into a parallel symbol string (step ST2a).
　ニューラルネットワーク部４は、受信信号調整部３から並列に入力されたシンボルとニューラル素子間の結合荷重との乗算和を非線形変換してシンボルごとの尤度を生成する。ニューラルネットワーク部４によってシンボルごとに生成された尤度は、ビット尤度生成部１１に並列に入力される。 The neural network unit 4 non-linearly converts the multiplication sum of the symbol input in parallel from the received signal adjustment unit 3 and the connection weight between the neural elements to generate the likelihood for each symbol. The likelihoods generated for each symbol by the neural network unit 4 are input in parallel to the bit likelihood generation unit 11.
　ビット尤度生成部１１は、ニューラルネットワーク部４から入力した尤度に基づいて、ビットごとの尤度を生成する（ステップＳＴ３ａ）。例えば、ビット尤度生成部１１は、入力した尤度を規格化振幅に変換し、規格化振幅からビットごとの尤度を算出する。規格化振幅は、振幅レベルがシンボルの最小値と最大値とに対応するように規格化された振幅である。４値ＰＡＭ信号の振幅は（－３，－１，１，３）の値をとるので、規格化振幅は－３から３の範囲に収まる数値になる。 The bit likelihood generation unit 11 generates a likelihood for each bit based on the likelihood input from the neural network unit 4 (step ST3a). For example, the bit likelihood generation unit 11 converts the input likelihood into a normalized amplitude, and calculates the likelihood for each bit from the normalized amplitude. The normalized amplitude is an amplitude standardized so that the amplitude level corresponds to the minimum value and the maximum value of the symbol. Since the amplitude of the four-value PAM signal has a value of (-3, -1, 1, 3), the normalized amplitude falls within the range of -3 to 3.
　図８Ａは、１ビット目のビットβ１の対数尤度と規格化振幅との関係を示すグラフである。図８Ｂは、２ビット目のビットβ２の対数尤度と規格化振幅との関係を示すグラフである。ビット尤度生成部１１は、図８Ａおよび図８Ｂに示すようなビットごとの対数尤度と規格化振幅との対応関係を示すビット尤度変換テーブルを用いて、規格化振幅をビットごとの対数尤度に変換する。 FIG. 8A is a graph showing the relationship between the log likelihood of the first bit β1 and the normalized amplitude. FIG. 8B is a graph showing the relationship between the log likelihood of the bit β 2 of the second bit and the normalized amplitude. The bit likelihood generation unit 11 uses the bit likelihood conversion table indicating the correspondence between the log likelihood for each bit and the normalized amplitude as shown in FIG. 8A and FIG. Convert to likelihood.
　例えば、規格化振幅をＡとし、１ビット目のビットをβ１とし、２ビット目のビットをβ２とし、ｌビット目のビットを“０”として送信したときに規格化振幅がＡとなる確率をｐｒ（Ａ｜βｌ＝０）とし、さらにｌビット目のビットを“１”として送信したときに規格化振幅がＡとなる確率をｐｒ（Ａ｜βｌ＝１）とする。ｌビット目のビットに対応した対数尤度ＬＬＲｌは、例えば、下記式（３）で表すことができる。ただし、ｌ＝１，・・・，ｌｏｇ２Ｄであり、Ｄは、光信号の変調多値度である。
　ＬＬＲｌ＝ｌｎ［｛ｐｒ（Ａ｜βｌ＝０）｝／｛ｐｒ（Ａ｜βｌ＝１）｝］・・・（３） For example, the normalized amplitude is A, the bit of the first bit and beta 1, the bit of the second bit and beta 2, normalized amplitude is A when transmitted as a bit of l-th bit "0" Let the probability be p r (A | β 1 = 0) and the probability that the normalized amplitude be A when the 1st bit is transmitted as “1” be p r (A | β 1 = 1) . The log likelihood LLR l corresponding to the bit of the l-th bit can be represented, for example, by the following formula (3). Here, l = 1,..., Log 2 D, where D is the modulation multilevel degree of the optical signal.
LLR l = ln [{p r (A | β l = 0)} / {p r (A | β l = 1)}] (3)
　ビット尤度生成部１１は、例えば、上記式（３）を使用して規格化振幅Ａに対して算出された対数尤度ＬＬＲｌが登録されているビット尤度変換テーブルを参照して、シンボルごとの尤度に対応する対数尤度ＬＬＲｌを求める。対数尤度ＬＬＲｌが－１である場合、ｌビット目のビットが“０”であることが尤もらしくなる。対数尤度ＬＬＲｌが１であると、ｌビット目のビットが“１”であることが尤もらしくなる。 The bit likelihood generation unit 11 refers to, for example, the bit likelihood conversion table in which the log likelihood LLR l calculated for the normalized amplitude A using the equation (3) is registered, Find log likelihood LLR l corresponding to each likelihood. When the log likelihood LLR l is −1, it is possible that the lth bit is “0”. When the log likelihood LLR l is 1, it is possible that the l-th bit is "1".
　なお、図８Ａおよび図８Ｂでは、４値ＰＡＭ信号の４値（００，０１，１１，１０）を振幅（－３，－１，１，３）に割り当てたときのビット尤度変換テーブルを示している。ビット尤度生成部１１から得られたビット単位の対数尤度は、軟判定誤り訂正復号部１２へ入力される。 8A and 8B show bit likelihood conversion tables when four values (00, 01, 11, 10) of the four-value PAM signal are assigned to amplitudes (-3, -1, 1, 3). ing. The log likelihood in bit units obtained from the bit likelihood generation unit 11 is input to the soft decision error correction decoding unit 12.
　軟判定誤り訂正復号部１２は、ビット尤度生成部１１から入力したビットごとの尤度に基づいて、軟判定による誤り訂正復号を行う（ステップＳＴ４ａ）。例えば、軟判定誤り訂正復号部１２は、誤り訂正符号のビット割当規則に基づいて光送受信器および光伝送路で生じたビット誤りを検出して訂正する。軟判定誤り訂正符号には、ＬＤＰＣ符号が挙げられる。 The soft decision error correction decoding unit 12 performs error correction decoding by soft decision on the basis of the likelihood for each bit input from the bit likelihood generation unit 11 (step ST4a). For example, the soft decision error correction decoding unit 12 detects and corrects a bit error generated in the optical transmitter-receiver and the optical transmission line based on the bit assignment rule of the error correction code. The soft decision error correction code includes an LDPC code.
　実施の形態２に係る光受信装置１Ａは、実施の形態１と同様に、非特許文献１に記載された光受信装置と比較してニューラルネットワーク部４の出力層の素子数を削減することができる。従って、実施の形態２に係る光受信装置１Ａでは、非線形歪みの等化精度を維持しつつ、ニューラルネットワーク部４の計算を簡素化することが可能である。 Similar to the first embodiment, the optical receiver 1A according to the second embodiment can reduce the number of elements in the output layer of the neural network unit 4 as compared with the optical receiver described in Non-Patent Document 1. it can. Therefore, in the optical receiver 1A according to the second embodiment, it is possible to simplify the calculation of the neural network unit 4 while maintaining the equalization accuracy of the non-linear distortion.
　図６に示した光受信装置１Ａは、ビット尤度生成部１１を備えたが、ニューラルネットワーク部４の構成によっては、ビット尤度生成部１１を省略してもよい。例えば、図６に示したニューラルネットワーク部４ではシンボルごとに尤度を生成したが、出力層の構成を、ビットごとの尤度を生成するように変更する。軟判定誤り訂正復号部１２は、ニューラルネットワーク部からビットごとの尤度を直接入力して、軟判定による誤り訂正復号を行うことになる。 Although the optical receiver 1A shown in FIG. 6 includes the bit likelihood generation unit 11, the bit likelihood generation unit 11 may be omitted depending on the configuration of the neural network unit 4. For example, although the neural network unit 4 shown in FIG. 6 generates the likelihood for each symbol, the configuration of the output layer is changed to generate the likelihood for each bit. The soft decision error correction decoding unit 12 directly inputs the likelihood for each bit from the neural network unit and performs error correction decoding by soft decision.
　例えば、ニューラルネットワーク部４に対して、変調多値度Ｄが４である４値ＰＡＭ信号のＢ個のシンボルが入力される場合、図１および図６に示したニューラルネットワーク部４では、出力層の素子数が４×Ｂとなる。
　一方、ビットごとの尤度を生成するように出力層の構成を変更すると、出力層の素子数は、４×Ｂから４×Ｂ×Ｄに増加する。ただし、このニューラルネットワーク部を備える光受信装置では、ビット尤度生成部１１が省略されるので、非線形歪みの等化精度を維持しつつ、非特許文献１に記載された光受信装置よりも回路実装効率を高めることが可能である。 For example, when B symbols of a four-value PAM signal having a modulation multi-level D of 4 are input to the neural network unit 4, the neural network unit 4 shown in FIGS. The number of elements of is 4 × B.
On the other hand, when the configuration of the output layer is changed to generate the likelihood for each bit, the number of elements in the output layer increases from 4 × B to 4 × B × D. However, in the optical receiver including the neural network unit, the bit likelihood generator 11 is omitted, and therefore, the circuit more than the optical receiver described in Non-Patent Document 1 while maintaining the equalization accuracy of the nonlinear distortion. It is possible to improve the mounting efficiency.
　以上のように、実施の形態２に係る光受信装置１Ａにおいて、入力シンボル数Ｍと同数の出力層４２－１～４２－Ｍを有するニューラルネットワーク部４が、入力されたシンボルと素子間の結合荷重との乗算和を非線形変換して尤度を生成する。軟判定誤り訂正復号部１２は、ニューラルネットワーク部４によって生成された尤度に基づいて、軟判定による誤り訂正復号を行う。このように構成することで、ニューラルネットワーク部４が、光送受信器および光伝送路に生じた非線形歪みを等化する。出力層４２－１～４２－Ｍが、入力シンボル数Ｍと同数であるので、出力層の素子数が光信号の変調多値度Ｄと入力シンボル数Ｍとを乗算した個数（＝Ｄ×Ｂ）に抑えられる。これにより、光受信装置１Ａは、非線形歪みの等化精度を維持しつつ、回路実装効率を高めることができる。 As described above, in the optical receiver 1A according to the second embodiment, the neural network unit 4 having the same number of output layers 42-1 to 42-M as the number M of input symbols is the connection between the input symbols and the elements. The likelihood is generated by non-linear transformation of the multiplication sum with the load. The soft decision error correction decoding unit 12 performs error correction decoding by soft decision based on the likelihood generated by the neural network unit 4. By configuring in this manner, the neural network unit 4 equalizes the nonlinear distortion generated in the optical transmitter-receiver and the optical transmission path. Since the output layers 42-1 to 42-M have the same number as the number M of input symbols, the number of elements of the output layer is the number of the modulation multi-level D of the optical signal multiplied by the number M of input symbols (= D × B It is suppressed to). Thereby, the optical receiver 1A can improve the circuit mounting efficiency while maintaining the equalization accuracy of the non-linear distortion.
　図９Ａは、実施の形態１に係る光受信装置１および実施の形態２に係る光受信装置１Ａのそれぞれの機能を実現するハードウェア構成を示すブロック図である。図９Ｂは、実施の形態１に係る光受信装置１および実施の形態２に係る光受信装置１Ａのそれぞれの機能を実現するソフトウェアを実行するハードウェア構成を示すブロック図である。
　図９Ａおよび図９Ｂにおいて、光電変換器１００は、入力した光信号を電気信号に変換する装置であって、図１および図６に示した光電変換部２に相当する。光電変換器１００は、内部に局部発振光を発振する発振器を備えていてもよい。 FIG. 9A is a block diagram showing a hardware configuration for realizing respective functions of the light receiving device 1 according to the first embodiment and the light receiving device 1A according to the second embodiment. FIG. 9B is a block diagram showing a hardware configuration for executing software for realizing the respective functions of the light receiving device 1 according to the first embodiment and the light receiving device 1A according to the second embodiment.
9A and 9B, the photoelectric converter 100 is a device for converting an input optical signal into an electric signal, and corresponds to the photoelectric conversion unit 2 shown in FIGS. 1 and 6. The photoelectric converter 100 may include an oscillator that oscillates local oscillation light.
　光受信装置１における、受信信号調整部３、ニューラルネットワーク部４、尤度選択部５、シンボル復元部６、遅延調整部７、復調部８、硬判定誤り訂正復号部９および結合荷重更新部１０のそれぞれの機能は、処理回路によって実現される。
　また、光受信装置１Ａにおける、受信信号調整部３、ニューラルネットワーク部４、結合荷重更新部１０、ビット尤度生成部１１および軟判定誤り訂正復号部１２のそれぞれの機能は、処理回路によって実現される。
　すなわち、光受信装置１は、図４に示したフローチャートにおける複数の処理のそれぞれを実行するための処理回路を備え、光受信装置１Ａは、図７に示したフローチャートにおける複数の処理のそれぞれを実行するための処理回路を備える。
　処理回路は、専用のハードウェアであってもよく、メモリに記憶されたプログラムを実行するＣＰＵ（Ｃｅｎｔｒａｌ　Ｐｒｏｃｅｓｓｉｎｇ　Ｕｎｉｔ）であってもよい。 Received signal adjustment unit 3, neural network unit 4, likelihood selection unit 5, symbol recovery unit 6, delay adjustment unit 7, demodulation unit 8, hard decision error correction decoding unit 9, and connection weight update unit 10 in optical receiver 1. The respective functions of are realized by the processing circuit.
Also, the functions of the received signal adjustment unit 3, the neural network unit 4, the combined load update unit 10, the bit likelihood generation unit 11, and the soft decision error correction decoding unit 12 in the optical receiver 1A are realized by processing circuits. Ru.
That is, the optical receiver 1 includes processing circuits for executing each of the plurality of processes in the flowchart shown in FIG. 4, and the optical receiver 1A performs each of the plurality of processes in the flowchart shown in FIG. Processing circuitry for
The processing circuit may be dedicated hardware or may be a central processing unit (CPU) that executes a program stored in a memory.
　処理回路が図９Ａに示す専用のハードウェアである場合、処理回路１０１は、例えば、単一回路、複合回路、プログラム化したプロセッサ、並列プログラム化したプロセッサ、ＡＳＩＣ（Ａｐｐｌｉｃａｔｉｏｎ　Ｓｐｅｃｉｆｉｃ　Ｉｎｔｅｇｒａｔｅｄ　Ｃｉｒｃｕｉｔ）、ＦＰＧＡ（Ｆｉｅｌｄ－Ｐｒｏｇｒａｍｍａｂｌｅ　Ｇａｔｅ　Ａｒｒａｙ）またはこれらを組み合わせたものが該当する。 When the processing circuit is dedicated hardware shown in FIG. 9A, the processing circuit 101 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an application specific integrated circuit (ASIC), an FPGA (FPGA) Field-Programmable Gate Array) or a combination thereof is applicable.
　処理回路が図９Ｂに示すプロセッサ１０２であると、光受信装置１における、受信信号調整部３、ニューラルネットワーク部４、尤度選択部５、シンボル復元部６、遅延調整部７、復調部８、硬判定誤り訂正復号部９および結合荷重更新部１０のそれぞれの機能は、ソフトウェア、ファームウェアまたはソフトウェアとファームウェアとの組み合わせによって実現される。
　また、光受信装置１Ａにおける、受信信号調整部３、ニューラルネットワーク部４、結合荷重更新部１０、ビット尤度生成部１１および軟判定誤り訂正復号部１２のそれぞれの機能も、ソフトウェア、ファームウェアまたはソフトウェアとファームウェアとの組み合わせによって実現される。
　ソフトウェアまたはファームウェアはプログラムとして記述され、メモリ１０３に記憶される。 If the processing circuit is the processor 102 shown in FIG. 9B, the received signal adjustment unit 3, the neural network unit 4, the likelihood selection unit 5, the symbol restoration unit 6, the delay adjustment unit 7, the demodulation unit 8 in the light receiving device 1. The respective functions of the hard decision error correction decoding unit 9 and the combined load updating unit 10 are realized by software, firmware or a combination of software and firmware.
In addition, each function of the received signal adjustment unit 3, the neural network unit 4, the combined load update unit 10, the bit likelihood generation unit 11, and the soft decision error correction decoding unit 12 in the optical receiver 1A is also software, firmware or software It is realized by the combination of the and the firmware.
The software or firmware is written as a program and stored in the memory 103.
　プロセッサ１０２は、メモリ１０３に記憶されたプログラムを読み出して実行することで、光受信装置１における、受信信号調整部３、ニューラルネットワーク部４、尤度選択部５、シンボル復元部６、遅延調整部７、復調部８、硬判定誤り訂正復号部９および結合荷重更新部１０のそれぞれの機能を実現する。すなわち、光受信装置１は、プロセッサ１０２によって実行されたときに、図４に示す一連の処理のそれぞれが結果的に実行されるプログラムを記憶するためのメモリ１０３を備える。
　これらのプログラムは、受信信号調整部３、ニューラルネットワーク部４、尤度選択部５、シンボル復元部６、遅延調整部７、復調部８、硬判定誤り訂正復号部９および結合荷重更新部１０の手順または方法をコンピュータに実行させるものである。 The processor 102 reads out and executes the program stored in the memory 103, whereby the reception signal adjustment unit 3, the neural network unit 4, the likelihood selection unit 5, the symbol restoration unit 6, and the delay adjustment unit in the light receiving device 1 7, the functions of the demodulation unit 8, the hard decision error correction decoding unit 9, and the combined load updating unit 10 are realized. That is, the optical receiver 1 includes the memory 103 for storing a program that is executed by the processor 102 as a result of each of the series of processes shown in FIG. 4.
These programs include the received signal adjustment unit 3, the neural network unit 4, the likelihood selection unit 5, the symbol restoration unit 6, the delay adjustment unit 7, the demodulation unit 8, the hard decision error correction decoding unit 9 and the connection weight update unit 10. Making a computer execute a procedure or method.
　同様に、プロセッサ１０２は、メモリ１０３に記憶されたプログラムを読み出して実行することで、光受信装置１Ａにおける、受信信号調整部３、ニューラルネットワーク部４、結合荷重更新部１０、ビット尤度生成部１１および軟判定誤り訂正復号部１２のそれぞれの機能を実現する。すなわち、光受信装置１Ａは、プロセッサ１０２によって実行されたときに、図７に示す一連の処理のそれぞれが結果的に実行されるプログラムを記憶するためのメモリ１０３を備える。これらのプログラムは、受信信号調整部３、ニューラルネットワーク部４、結合荷重更新部１０、ビット尤度生成部１１および軟判定誤り訂正復号部１２の手順または方法をコンピュータに実行させるものである。 Similarly, the processor 102 reads out and executes the program stored in the memory 103, whereby the reception signal adjustment unit 3, the neural network unit 4, the connection weight update unit 10, and the bit likelihood generation unit in the light receiving device 1A. 11 and each function of the soft decision error correction decoding unit 12 are realized. That is, the optical receiver 1A includes the memory 103 for storing a program that is executed by the processor 102 as a result of each of the series of processes shown in FIG. These programs cause a computer to execute the procedure or method of the received signal adjustment unit 3, the neural network unit 4, the connection weight update unit 10, the bit likelihood generation unit 11, and the soft decision error correction decoding unit 12.
　メモリ１０３には、例えば、ＲＡＭ（Ｒａｎｄｏｍ　Ａｃｃｅｓｓ　Ｍｅｍｏｒｙ）、ＲＯＭ（Ｒｅａｄ　Ｏｎｌｙ　Ｍｅｍｏｒｙ）、フラッシュメモリ、ＥＰＲＯＭ（Ｅｒａｓａｂｌｅ　Ｐｒｏｇｒａｍｍａｂｌｅ　Ｒｅａｄ　Ｏｎｌｙ　Ｍｅｍｏｒｙ）、ＥＥＰＲＯＭ（Ｅｌｅｃｔｒｉｃａｌｌｙ－ＥＰＲＯＭ）などの不揮発性または揮発性の半導体メモリ、磁気ディスク、フレキシブルディスク、光ディスク、コンパクトディスク、ミニディスク、ＤＶＤなどが該当する。 The memory 103 is, for example, a non-volatile or volatile semiconductor memory such as a random access memory (RAM), a read only memory (ROM), a flash memory, an erasable programmable read only memory (EPROM), and an EEPROM (electrically-EPROM). A magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, a DVD, etc. correspond.
　受信信号調整部３、ニューラルネットワーク部４、尤度選択部５、シンボル復元部６、遅延調整部７、復調部８、硬判定誤り訂正復号部９および結合荷重更新部１０のそれぞれの機能について一部を専用のハードウェアで実現し、一部をソフトウェアまたはファームウェアで実現してもよい。
　例えば、受信信号調整部３については、専用のハードウェアとしての処理回路１０１でその機能を実現する。ニューラルネットワーク部４、尤度選択部５、シンボル復元部６、遅延調整部７、復調部８、硬判定誤り訂正復号部９および結合荷重更新部１０については、プロセッサ１０２が、メモリ１０３に記憶されたプログラムを実行することによってその機能を実現してもよい。これは、光受信装置１Ａの構成要素においても同様である。
　このように、処理回路は、ハードウェア、ソフトウェア、ファームウェア、または、これらの組み合わせによって上記機能のそれぞれを実現することができる。 The respective functions of received signal adjustment unit 3, neural network unit 4, likelihood selection unit 5, symbol restoration unit 6, delay adjustment unit 7, demodulation unit 8, hard decision error correction decoding unit 9 and connection weight update unit 10 The part may be realized by dedicated hardware, and a part may be realized by software or firmware.
For example, the function of the reception signal adjustment unit 3 is realized by the processing circuit 101 as dedicated hardware. The processor 102 is stored in the memory 103 for the neural network unit 4, likelihood selection unit 5, symbol restoration unit 6, delay adjustment unit 7, demodulation unit 8, hard decision error correction decoding unit 9 and connection weight update unit 10. The function may be realized by executing a program. The same applies to the components of the light receiving device 1A.
Thus, the processing circuit can realize each of the above functions by hardware, software, firmware, or a combination thereof.
　なお、本発明は上記実施の形態に限定されるものではなく、本発明の範囲内において、実施の形態のそれぞれの自由な組み合わせまたは実施の形態のそれぞれの任意の構成要素の変形もしくは実施の形態のそれぞれにおいて任意の構成要素の省略が可能である。 The present invention is not limited to the above embodiment, and within the scope of the present invention, variations or embodiments of respective free combinations of the embodiments or respective optional components of the embodiments. An optional component can be omitted in each of the above.
　この発明に係る光受信装置は、非線形歪みの等化精度を維持しつつ、回路実装効率を高めることができるので、多値の光信号を扱う光通信システムに利用可能である。 The optical receiver according to the present invention can increase circuit mounting efficiency while maintaining the equalization accuracy of nonlinear distortion, and therefore can be used for an optical communication system that handles multilevel optical signals.
　１，１Ａ　光受信装置、２　光電変換部、３　受信信号調整部、４　ニューラルネットワーク部、５　尤度選択部、６　シンボル復元部、７　遅延調整部、８　復調部、９　硬判定誤り訂正復号部、１０　結合荷重更新部、１１　ビット尤度生成部、１２　軟判定誤り訂正復号部、４０　入力層、４１　中間層、４２－１～４２－Ｍ　出力層、５０－１～５０－Ｍ　最大値演算部、５１－１～５１－Ｍ　要素番号抽出部、１００　光電変換器、１０１　処理回路、１０２　プロセッサ、１０３　メモリ。 1, 1A optical receiver, 2 photoelectric conversion unit, 3 reception signal adjustment unit, 4 neural network unit, 5 likelihood selection unit, 6 symbol recovery unit, 7 delay adjustment unit, 8 demodulation unit, 9 hard decision error correction decoding unit , 10 joint weight update unit, 11 bit likelihood generation unit, 12 soft decision error correction decoding unit, 40 input layers, 41 intermediate layers, 42-1 to 42-M output layers, 50-1 to 50-M maximum value calculation 51-1 to 51-M Element number extraction unit, 100 photoelectric converter, 101 processing circuit, 102 processor, 103 memory.

Claims (10)

  1. 　複数のシンボルを含む光信号を受信する光受信装置であって、
    　受信された前記光信号を電気信号に変換する光電変換部と、
    　前記光電変換部によって前記光信号から変換された電気信号を並列のシンボル列に変換する第１の調整部と、
    　入力層、１段または複数段の中間層、および前記入力層に入力されるシンボル数と同数の出力層を有し、前記第１の調整部から並列に入力されたシンボルと素子間の結合荷重との乗算和を非線形変換してシンボルごとの尤度を生成するニューラルネットワーク部と、
    　前記ニューラルネットワーク部によって生成された尤度から、前記出力層ごとに尤度を選択する尤度選択部と、
    　前記尤度選択部によって選択された尤度に対応する並列のシンボル列を直列のシンボル列に変換する第２の調整部と、
    　前記第２の調整部から入力したシンボル列からビット列を復調する復調部と、
    　前記復調部によって復調されたビット列に基づいて、硬判定による誤り訂正復号を行う第１の誤り訂正復号部と
    　を備えたことを特徴とする光受信装置。     An optical receiver for receiving an optical signal including a plurality of symbols, comprising:
    A photoelectric conversion unit that converts the received optical signal into an electrical signal;
    A first adjustment unit configured to convert an electrical signal converted from the optical signal by the photoelectric conversion unit into a parallel symbol string;
    An input layer, one or more intermediate layers, and an output layer having the same number as the number of symbols input to the input layer, and the connection weight between the symbol and the element input in parallel from the first adjustment unit A neural network unit that generates a likelihood for each symbol by nonlinearly converting the multiplication sum with
    A likelihood selection unit that selects a likelihood for each of the output layers from the likelihoods generated by the neural network unit;
    A second adjusting unit configured to convert a parallel symbol string corresponding to the likelihood selected by the likelihood selecting unit into a serial symbol string;
    A demodulation unit that demodulates a bit string from the symbol string input from the second adjustment unit;
    And a first error correction decoding unit for performing error correction decoding by hard decision based on the bit string demodulated by the demodulation unit.
  2. 　前記尤度選択部は、前記出力層ごとに選択した尤度に対応する並列のシンボル列を出力し、
    　前記第２の調整部は、前記尤度選択部から入力した並列のシンボル列を直列のシンボル列に変換すること
    　を特徴とする請求項１記載の光受信装置。     The likelihood selection unit outputs parallel symbol strings corresponding to the likelihoods selected for each of the output layers,
    The light receiving apparatus according to claim 1, wherein the second adjustment unit converts the parallel symbol string input from the likelihood selection unit into a serial symbol string.
  3. 　前記尤度選択部によって前記出力層ごとに選択された尤度からシンボルを復元し、復元したシンボルから構成される並列のシンボル列を出力するシンボル復元部を備え、
    　前記第２の調整部は、前記シンボル復元部から入力した並列のシンボル列を直列のシンボル列に変換すること
    　を特徴とする請求項１記載の光受信装置。     The symbol selection unit includes a symbol recovery unit that recovers a symbol from the likelihood selected for each of the output layers by the likelihood selection unit, and outputs a parallel symbol string composed of the recovered symbols.
    The light receiving apparatus according to claim 1, wherein the second adjusting unit converts the parallel symbol string input from the symbol restoring unit into a serial symbol string.
  4. 　複数のシンボルを含む光信号を受信する光受信装置であって、
    　受信された前記光信号を電気信号に変換する光電変換部と、
    　前記光電変換部によって前記光信号から変換された電気信号を並列のシンボル列に変換する第１の調整部と、
    　入力層、１段または複数段の中間層、および前記入力層に入力されるシンボル数と同数の出力層を有し、前記第１の調整部から並列に入力されたシンボルと素子間の結合荷重との乗算和を非線形変換して尤度を生成するニューラルネットワーク部と、
    　前記ニューラルネットワーク部によって生成された尤度に基づいて、軟判定による誤り訂正復号を行う第２の誤り訂正復号部と
    　を備えたことを特徴とする光受信装置。     An optical receiver for receiving an optical signal including a plurality of symbols, comprising:
    A photoelectric conversion unit that converts the received optical signal into an electrical signal;
    A first adjustment unit configured to convert an electrical signal converted from the optical signal by the photoelectric conversion unit into a parallel symbol string;
    An input layer, one or more intermediate layers, and an output layer having the same number as the number of symbols input to the input layer, and the connection weight between the symbol and the element input in parallel from the first adjustment unit A neural network unit that generates a likelihood by nonlinearly converting the multiplication sum with
    And a second error correction decoding unit that performs error correction decoding by soft decision based on the likelihood generated by the neural network unit.
  5. 　前記ニューラルネットワーク部は、ビットごとの尤度を生成し、
    　前記第２の誤り訂正復号部は、前記ニューラルネットワーク部から入力したビットごとの尤度に基づいて、軟判定による誤り訂正復号を行うこと
    　を特徴とする請求項４記載の光受信装置。     The neural network unit generates bitwise likelihoods,
    The optical receiving apparatus according to claim 4, wherein the second error correction decoding unit performs error correction decoding by soft decision based on the likelihood for each bit input from the neural network unit.
  6. 　前記ニューラルネットワーク部によってシンボルごとに生成された尤度に基づいて、ビットごとの尤度を生成するビット尤度生成部を備え、
    　前記第２の誤り訂正復号部は、前記ビット尤度生成部から入力したビットごとの尤度に基づいて、軟判定による誤り訂正復号を行うこと
    　を特徴とする請求項４記載の光受信装置。     And a bit likelihood generation unit configured to generate a likelihood for each bit based on the likelihood generated for each symbol by the neural network unit.
    The optical receiving apparatus according to claim 4, wherein the second error correction decoding unit performs error correction decoding by soft decision based on the likelihood for each bit input from the bit likelihood generation unit.
  7. 　前記光信号は、多値振幅変調フォーマットの信号であること
    　を特徴とする請求項１または請求項４記載の光受信装置。     5. The light receiving apparatus according to claim 1, wherein the light signal is a signal of a multi-value amplitude modulation format.
  8. 　前記ニューラルネットワーク部の前記結合荷重を更新する結合荷重更新部を備えたこと
    　を特徴とする請求項１または請求項４記載の光受信装置。     5. The light receiving apparatus according to claim 1, further comprising: a connection load updating unit that updates the connection load of the neural network unit.
  9. 　複数のシンボルを含む光信号を受信する光受信装置の光受信方法であって、
    　光電変換部が、受信された前記光信号を電気信号に変換するステップと、
    　第１の調整部が、前記光電変換部によって前記光信号から変換された電気信号を並列のシンボル列に変換するステップと、
    　入力層、１段または複数段の中間層、および前記入力層に入力されるシンボル数と同数の出力層を有するニューラルネットワーク部が、前記第１の調整部から入力されたシンボルと素子間の結合荷重との乗算和を非線形変換してシンボルごとの尤度を生成するステップと、
    　尤度選択部が、前記ニューラルネットワーク部によって生成された尤度から、前記出力層ごとに尤度を選択するステップと、
    　第２の調整部が、前記尤度選択部によって選択された尤度に対応する並列のシンボル列を直列のシンボル列に変換するステップと、
    　復調部が、前記第２の調整部から入力したシンボル列からビット列を復調するステップと、
    　第１の誤り訂正復号部が、前記復調部によって復調されたビット列に基づいて、硬判定による誤り訂正復号を行うステップと
    　を備えたことを特徴とする光受信方法。     What is claimed is: 1. A light receiving method of a light receiving apparatus for receiving a light signal including a plurality of symbols, comprising:
    A photoelectric conversion unit converting the received optical signal into an electrical signal;
    The first adjustment unit converting the electrical signal converted from the optical signal by the photoelectric conversion unit into a parallel symbol string;
    A neural network unit having an input layer, one or more intermediate layers, and output layers equal in number to the number of symbols input to the input layer is connected between the symbol and the element input from the first adjustment unit. Non-linear transformation of the multiplication sum with the load to generate a likelihood for each symbol;
    Selecting a likelihood for each of the output layers from the likelihood generated by the neural network unit;
    The second adjusting unit converts a parallel symbol string corresponding to the likelihood selected by the likelihood selecting unit into a serial symbol string;
    The demodulation unit demodulates a bit string from the symbol string input from the second adjustment unit;
    A first error correction decoding unit performing error correction decoding by hard decision based on the bit string demodulated by the demodulation unit.
  10. 　複数のシンボルを含む光信号を受信する光受信装置の光受信方法であって、
    　光電変換部が、受信された前記光信号を電気信号に変換するステップと、
    　第１の調整部が、前記光電変換部によって前記光信号から変換された電気信号を並列のシンボル列に変換するステップと、
    　入力層、１段または複数段の中間層、および前記入力層に入力されるシンボル数と同数の出力層を有するニューラルネットワーク部が、前記第１の調整部から入力されたシンボルと素子間の結合荷重との乗算和を非線形変換して尤度を生成するステップと、
    　第２の誤り訂正復号部が、前記ニューラルネットワーク部によって生成された尤度に基づいて、軟判定による誤り訂正復号を行うステップと
    　を備えたことを特徴とする光受信方法。     What is claimed is: 1. A light receiving method of a light receiving apparatus for receiving a light signal including a plurality of symbols, comprising:
    A photoelectric conversion unit converting the received optical signal into an electrical signal;
    The first adjustment unit converting the electrical signal converted from the optical signal by the photoelectric conversion unit into a parallel symbol string;
    A neural network unit having an input layer, one or more intermediate layers, and output layers equal in number to the number of symbols input to the input layer is connected between the symbol and the element input from the first adjustment unit. Non-linear transformation of the multiplication sum with the load to generate a likelihood;
    And a second error correction decoding unit performing soft correction based on the likelihood generated by the neural network unit.
PCT/JP2017/044916 2017-12-14 2017-12-14 Optical reception device and optical reception method WO2019116503A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2017/044916 WO2019116503A1 (en) 2017-12-14 2017-12-14 Optical reception device and optical reception method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2017/044916 WO2019116503A1 (en) 2017-12-14 2017-12-14 Optical reception device and optical reception method

Publications (1)

Publication Number Publication Date
WO2019116503A1 true WO2019116503A1 (en) 2019-06-20

Family

ID=66819080

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2017/044916 WO2019116503A1 (en) 2017-12-14 2017-12-14 Optical reception device and optical reception method

Country Status (1)

Country Link
WO (1) WO2019116503A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112398543A (en) * 2019-08-19 2021-02-23 上海诺基亚贝尔股份有限公司 Method, apparatus, system, device and computer readable medium for optical communication

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9166701B1 (en) * 2008-08-14 2015-10-20 Cirrex Systems, Llc System and method for receiving optical signals
JP2017126051A (en) * 2016-01-07 2017-07-20 日本電気株式会社 Template generation device, template generation method, template generation program, and phrase detection system
JP2017520974A (en) * 2014-09-03 2017-07-27 三菱電機株式会社 System and method for recovering carrier phase in optical communications

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9166701B1 (en) * 2008-08-14 2015-10-20 Cirrex Systems, Llc System and method for receiving optical signals
JP2017520974A (en) * 2014-09-03 2017-07-27 三菱電機株式会社 System and method for recovering carrier phase in optical communications
JP2017126051A (en) * 2016-01-07 2017-07-20 日本電気株式会社 Template generation device, template generation method, template generation program, and phrase detection system

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
RIOS-MULLER, R. ET AL.: "Experimental estimation of optical nonlinear memory channel conditional distribution using deep neural networks", OPTICAL FIBER COMMUNICATIONS CONFERENCE AND EXHIBITION (OFC, March 2017 (2017-03-01), pages 1 - 3, XP033101367 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112398543A (en) * 2019-08-19 2021-02-23 上海诺基亚贝尔股份有限公司 Method, apparatus, system, device and computer readable medium for optical communication
CN112398543B (en) * 2019-08-19 2023-08-25 上海诺基亚贝尔股份有限公司 Methods, apparatus, systems, devices, and computer readable media for optical communications

Similar Documents

Publication Publication Date Title
CN109075807B (en) System and method for Faster Than Nyquist (FTN) transmission
Schädler et al. Soft-demapping for short reach optical communication: A comparison of deep neural networks and volterra series
US8185796B2 (en) Mitigation of fiber nonlinearities in multilevel coded-modulation schemes
US9203558B1 (en) Soft-decision decoding of differential 16-QAM
CN113169752B (en) Learning in a communication system
CN103516647A (en) Low-complexity, highly-spectrally-efficient communications
JP5724031B2 (en) Method for decoding an optical data signal
JP6753931B2 (en) Methods for optical receivers, optical transmitters and optical receivers
Karlsson et al. Multidimensional modulation and coding in optical transport
TWI317578B (en) Trellis processing apparatus and its associated method
Schibisch et al. Online label recovery for deep learning-based communication through error correcting codes
CN114337911A (en) Communication method based on neural network and related device
CN112514337A (en) Distribution matching circuit, distribution dematching circuit, distribution matching method, distribution dematching method, and optical transmission system
KR20230098577A (en) Non-linear neural network equalizer for high-speed data channels
Schaedler et al. Neural network-based soft-demapping for nonlinear channels
WO2019116503A1 (en) Optical reception device and optical reception method
US8578229B1 (en) High-throughput pipelined and scalable architecture for a K-Best MIMO detector
US11855699B2 (en) Optical transmission system, optical transmitting apparatus and optical receiving apparatus
WO2019082239A1 (en) Equalization device and equalization method
Prinz et al. Successive Interference Cancellation for Bandlimited Channels with Direct Detection
JP6450209B2 (en) Optical transmitter, optical receiver, optical transmission device, optical transmission system, optical transmission method, optical reception method, and optical transmission method
WO2023166630A1 (en) Symbol determination device, symbol determination method and program
KR101839749B1 (en) Method and apparatus for cooperation iterative decoding
JP5656617B2 (en) Receiving apparatus and method
CN112398765B (en) Signal processing method, receiver and digital processing chip

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: 17934674

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17934674

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: JP