Estimation device, training device, and estimation method

Abstract

One aspect of the present invention relates to an estimation device comprising: a waveform feature amount estimation unit that, with respect to an optical communication path configured by connecting one or more unit sections, and an optical signal transmitted from a transmitter and inputted to the optical communication path, estimates, for each unit section constituting the optical communication path, the feature amount of an electric field waveform outputted from the unit section on the basis of the feature amount of the electric field wavelength of the optical signal inputted to the unit section; and a transmission quality estimation unit that, on the basis of the feature amount of the electric field waveform outputted from the last unit section of the optical communication path, which has been estimated by the waveform feature amount estimation unit, estimates the transmission quality of the optical communication path.

Description

Estimation device, learning device, and estimation method

[0001] The present invention relates to techniques for an estimation device, a learning device, and an estimation method.

[0002] In the following description, subscripts are represented by an underscore. For example, a small "b" at the bottom right of the letter "A" is represented as "A_b."

[0003] Conventionally, networks constructed using ROADM (reconfigurable optical add / drop multiplexer) have been proposed. In ROADM, optical signals transmitted from a communication device are forwarded at nodes and transmitted to a destination communication device. Figure 20 shows an overview of a network constructed using ROADM. Each node (nodes 1 to 5) is composed of an optical switch. Optical signals transmitted from a communication device (e.g., a first communication device) are forwarded without photoelectric conversion at the node and delivered as light to a destination communication device (e.g., a second communication device). In an all-optical network that connects such end-to-end optical paths without photoelectric conversion, when a new connection request is received from a communication device, it is necessary to select and assign an appropriate route from among numerous optical path candidates.

[0004] Each candidate optical path has different transmission distances, fiber types, and optical amplifier gains. This affects the transmission quality (e.g., bit error rate) obtained by the receiving communication device. Therefore, it is necessary to select and allocate optical paths with the highest transmission quality (e.g., those capable of error-free transmission). It may be possible to actually transmit a signal through each optical path to check the transmission quality (e.g., whether it is error-free) before allocating an appropriate path. However, this process takes an enormous amount of time. Therefore, in order to open optical paths quickly, it is effective to estimate the transmission quality of each optical path and allocate them accordingly. Below, we describe a method for estimating bit error rate as a specific example of a method for estimating transmission quality in an optical communication system.

[0005] One possible method for estimating the bit error rate in optical communication systems is to use propagation simulation. For example, the received waveform can be simulated by adding appropriate noise to the waveform after fiber propagation calculated using propagation simulation. The received code sequence is then identified using threshold judgment, and the bit error rate can be calculated from the difference with the transmitted code sequence (e.g., using a VPI simulator). Waveform changes due to fiber propagation can be calculated with high accuracy by applying an algorithm called the split-step Fourier method (SSFM) to the nonlinear Schrödinger equation. This method calculates the signal waveform by taking into account not only linear waveform changes (chromatic dispersion) due to fiber propagation but also nonlinear waveform changes (e.g., self-phase modulation). Therefore, the bit error rate can be accurately estimated even when nonlinear waveform distortion occurs.

[0006] FIG. 21 shows an example of a configuration for performing transmission simulation using SSFM in a network model consisting of N spans. In the assumed network model, optical amplifiers are inserted between each fiber span to compensate for propagation loss in the optical fiber. An optical signal (electric field waveform E_0) output from a transmitter first enters an optical fiber of length L_1 with input optical power P_1, and electric field waveform E_1 is output. Electric field waveform E_1 becomes the input electric field waveform of the next span, and light propagates in a similar manner. This process is repeated, and finally, electric field waveform E_N is received at the receiver with received optical power P_Rx. In this case, a transmission simulator based on the split-step Fourier method (SSFM) is used to calculate the bit error rate obtained at the receiver.

[0007] In SSFM, the output electric field waveform E_n+1 for the section can be calculated by providing the optical intensity and transmission distance for the input electric field waveform E_n. By repeating this process, the final received electric field waveform E_N can be obtained. The receiver then performs processes such as square-law detection, noise addition, and threshold judgment on E_N to calculate the received code sequence. The bit error rate is calculated based on a comparison of the transmitted code sequence with the received code sequence.

[0008] However, this method requires a huge amount of time to calculate nonlinear waveform changes, and also consumes a large amount of power in computational resources. Therefore, it is difficult to apply this method to bit error rate estimation for all-optical networks, which require real-time operation.

[0009] Digital coherent transmission systems, which are often used in core systems, can compensate for linear waveform distortion such as chromatic dispersion. This allows for long-distance transmission. In long-distance transmission systems, nonlinear waveform distortion can be approximated by Gaussian noise. Therefore, a high-speed transmission quality estimation system using Gaussian noise approximation has been proposed.

[0010] M. Birk et al., “The OpenROADM initiative [Invited]”, J. Opt. Commun. Netw., vol. 12, no. 6, pp. C58-67, June 2020.VPI simulator https: / / www.vpiphotonics.com / index.phpA. Ferrari et al., “GNPy: an open source application for physical layer aware open optical networks”, J. Opt. Commun. Netw., vol. 12, no. 6 , pp. C31-C40, June 2020.Pierluigi Poggiolini, Gabriella Bosco, Andrea Carena, Vittorio Curri, Yanchao Jiang, and Fabrizio Forghieri, “The GN-Model of Fiber Non-Linear Propagation and its Applications”, J. Lightw. Technol. 32(4), 694-721 (2014).

[0011] However, in direct modulation and direct detection systems, which are mainly used in access networks, long-distance transmission is difficult due to the influence of chromatic dispersion, and the Gaussian noise approximation does not hold. Therefore, in order to estimate transmission quality in real time in direct modulation and direct detection systems, a new high-speed transmission quality estimation method that does not use Gaussian noise approximation is required.

[0012] In view of the above circumstances, an object of the present invention is to provide a technique that can calculate the transmission quality of an optical path in a short time while taking into account nonlinear waveform changes in an optical communication network.

[0013] One aspect of the present invention is an estimation device that includes, for an optical communication path formed by concatenating one or more unit sections and an optical signal transmitted from a transmitter and input to the optical communication path, a waveform feature estimation unit that estimates, for each unit section that constitutes the optical communication path, a feature of an electric field waveform output from the unit section based on a feature of the electric field waveform of the optical signal input to that unit section, and a transmission quality estimation unit that estimates the transmission quality of the optical communication path based on the feature of the electric field waveform output from the last unit section of the optical communication path estimated by the waveform feature estimation unit.

[0014] One aspect of the present invention is a learning device that includes a learning control unit that generates a trained model by performing a learning process using multiple training data sets in which, for an optical communication path formed by connecting one or more unit sections and an optical signal transmitted from a transmitter and input to the optical communication path, for each unit section forming the optical communication path, the feature of the electric field waveform of the optical signal input to that unit section is used as an explanatory variable, and the feature of the electric field waveform output from the unit section is used as a target variable.

[0015] One aspect of the present invention is a learning device that includes a learning control unit that generates a trained model by performing a learning process using multiple training data sets in which, for an optical communication path formed by concatenating one or more unit sections and an optical signal transmitted from a transmitter and input to the optical communication path, the features of the electric field waveform output from the last unit section of the optical communication path are used as explanatory variables and the transmission quality of the optical communication path is used as a target variable.

[0016] One aspect of the present invention is an estimation method for an optical communication path formed by concatenating one or more unit sections and an optical signal transmitted from a transmitter and input to the optical communication path, the method comprising: a waveform feature estimation step of estimating, for each unit section constituting the optical communication path, a feature of an electric field waveform output from the unit section based on a feature of an electric field waveform of the optical signal input to that unit section; and a transmission quality estimation step of estimating the transmission quality of the optical communication path based on the feature of the electric field waveform output from the last unit section of the optical communication path estimated by the waveform feature estimation unit.

[0017] According to the present invention, it is possible to calculate the transmission quality of an optical path in an optical communication network in a short time while taking into account nonlinear waveform changes.

[0018] 1 is a diagram showing an outline of an all-optical network assumed in the present invention. FIG. 2 is a diagram showing a network model of an all-optical network 900 that is a target for transmission quality estimation by an estimation device 30. FIG. 3 is a diagram showing an outline of processing by the estimation device 30 in this embodiment. FIG. 4 is a diagram showing an example of a neural network that is a specific example of machine learning used in a waveform calculation unit 82. FIG. 5 is a diagram showing an example of a neural network that is a specific example of machine learning used in a transmission quality calculation unit 83. FIG. 6 is a diagram showing a specific example of a configuration for acquiring the above-mentioned teacher data. FIG. 7 is a diagram showing a specific example of the configuration of a feature acquisition unit 26. FIG. 8 is a diagram showing an outline of the configuration of an electric field waveform derivation unit 27. FIG. 9 is a diagram showing an example of a system configuration of a transmission quality estimation system. FIG. 10 is a schematic block diagram showing a specific example of the functional configuration of a learning device 20. FIG. 11 is a flowchart showing a specific example of the processing by the learning device 20. FIG. 12 is a schematic block diagram showing a specific example of the functional configuration of a determination device 30. FIG. 13 is a diagram showing an outline of processing by a waveform feature estimation unit 333 and a transmission quality estimation unit 334. FIG. 14 is a flowchart showing a specific example of processing by the estimation device 30. FIG. 15 is a diagram showing a network model of an all-optical network 900 that is a target for transmission quality estimation by an estimation device 30 of a second embodiment. FIG. 16 is a diagram showing an outline of processing by the estimation device 30 in the second embodiment. FIG. 10 is a diagram showing an example of a neural network, which is a specific example of machine learning used in the waveform calculation unit 82 of the second embodiment. FIG. 11 is a diagram showing an example of a neural network, which is a specific example of machine learning used in the transmission quality calculation unit 83 of the second embodiment. FIG. 12 is a diagram showing an outline of an example of the hardware configuration of an information processing device 90 applied to this embodiment. FIG. 13 is a diagram showing an outline of a network configured by ROADM. FIG. 14 is a diagram showing an example of a configuration for performing transmission simulation using SSFM in a network model configured of N spans.

[0019] [First Embodiment] A first embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram showing an outline of an all-optical network envisioned in the present invention. Remote communication devices (a first communication device 810 and a second communication device 840) are communicatively connected via multiple optical nodes 820. Each optical node 820 has an internal optical switch 821 and transfers signals in a predetermined direction as optical signals without optical-to-electrical conversion. The route within each optical node 820 is controlled by a controller 830.

[0020] The controller 830 acquires the transmission quality (e.g., bit error rate) of multiple candidate optical paths, and selects and assigns an optical path having a predetermined transmission quality (e.g., an optical path having a bit error rate equal to or lower than a specified value) from among the candidate optical paths.

[0021] The controller 830 acquires the transmission quality of the optical path, for example, by the following process. First, the controller 830 provides information about the optical path (e.g., the distance L of each section, the input optical power P, and the received optical power P_Rx) to the estimation device 30. The estimation device 30 calculates the transmission quality (e.g., the bit error rate) of multiple candidate optical paths using a transmission quality estimation technique. The controller 830 acquires information about the transmission quality estimated by the estimation device 30 from the estimation device 30.

[0022] FIG. 2 illustrates a network model of an all-optical network 900, the transmission quality of which is estimated by the estimation device 30. The all-optical network 900 is the transmission quality estimation target of the estimation device 30 and is controlled by the controller 830. A transmitter (Tx) and a receiver (Rx) are connected to the all-optical network 900, specifically a first communication device 810 and a second communication device 840. The all-optical network 900 is composed of N fiber spans. An optical amplifier is inserted between each fiber span. The optical amplifier compensates for propagation loss in the optical fiber. An optical signal (electric field waveform E_0) output from the transmitter first enters an optical fiber of length L_1 with input optical power P1. An optical signal (electric field waveform E_1) is then output from the optical fiber of length L_1. E_1 becomes the input electric field waveform of the next span, and light propagates in a similar manner. This process is repeated, and finally, an optical signal (electric field waveform E_N) is received at the receiver with received optical power P_Rx. In this case, a transmission simulator based on the Split Step Fourier Method (SSFM) is used as a method for calculating the bit error rate obtained at the receiver.

[0023] 3 is a diagram showing an outline of the processing of the estimation device 30 in this embodiment. The estimation device 30 can calculate the feature quantity A_n of the output electric field waveform E_n in a certain section based on the feature quantity A_n-1 of the input electric field waveform E_n-1 for that section, the light intensity P_n, and the transmission distance L_n. The feature quantity A_n of this output electric field waveform E_n becomes the feature quantity A_n of the input electric field waveform E_n in the next section. By repeating this process, the feature quantity A_N of the electric field waveform E_N in the receiver can finally be obtained.

[0024] As described above, in this embodiment, the final transmission quality (e.g., bit error rate) is estimated using the feature A linked to the electric field waveform E instead of the electric field waveform E. The amount of calculation can be reduced by using a feature A of a lower dimension that has a one-to-one correspondence with the electric field waveform E. The feature A (transmission feature) of the electric field waveform E0 output by the transmitter is defined as A_0. A transmission feature transmitter 81 transmits the transmission feature A_0. A waveform calculator 82 is provided for each span and calculates the feature A according to the waveform change of each span. The waveform calculators 82 are cascaded as many times as the number of spans, as shown in FIG. 3 . The waveform calculator 82 calculates an output feature A_n+1 from the input feature A_n of the span. The transmission feature transmission unit 81 may store a value calculated in advance for the feature A_0 in memory and output (transmit) the stored feature A_0 to the waveform calculation unit 82, or it may calculate the feature A_0 corresponding to the electric field waveform E_0 that is transmitted each time and output (transmit) it to the waveform calculation unit 82.

[0025] The waveform calculation unit 82 may be realized using a machine learning technique such as a neural network. That is, the waveform calculation unit 82 may acquire an output feature quantity from an input feature quantity using a trained model obtained by a training process. The feature quantity A_N of the received waveform finally obtained via the waveform calculation unit 82 is input to the transmission quality calculation unit 83.

[0026] FIG. 4 is a diagram showing an example of a neural network, which is a specific example of machine learning used in the waveform calculation unit 82. Input feature values ​​and attribute information related to transmission are input to the learning model as explanatory variables. For example, the transmission distance L may be used as the attribute information related to transmission. The input light intensity P may also be used as an explanatory variable. The learning model outputs an output feature value An+1 as a response variable. Examples of such learning models include supervised learning such as a support vector machine, a random forest, or a neural network. In such machine learning, multiple sets of training data are prepared for the training process. Each training data set includes an explanatory variable and a response variable (correct label) obtained from the explanatory variable. The training data may be data obtained by performing an actual transmission or by running a simulation.

[0027] The transmission quality calculation unit 83 estimates the transmission quality based on the feature quantity A_N of the received waveform. A specific example of the transmission quality is the bit error rate. The transmission quality calculation unit 83 calculates the transmission quality based on the feature quantity of the received waveform, the received light intensity, and the like. The transmission quality calculation unit 83 may be realized using a machine learning technique such as a neural network. In other words, the transmission quality calculation unit 83 may acquire the transmission quality from the feature quantity of the received waveform using a trained model obtained by a training process.

[0028] FIG. 5 is a diagram showing an example of a neural network, which is a specific example of machine learning used in the transmission quality calculation unit 83. A reception feature A_N is input to the learning model as an explanatory variable. The received light intensity P_Rx may also be used as an explanatory variable. Information about transmission quality is output from the learning model as a response variable. In FIG. 5, a bit error rate is used as a specific example of information about transmission quality. Examples of such learning models include supervised learning such as a support vector machine, a random forest, or a neural network. A plurality of training data sets are prepared to perform the learning process in such machine learning. Each training data set includes an explanatory variable and a response variable (correct label) obtained from the explanatory variable. The training data may be data obtained by performing actual transmission or may be data obtained by executing a simulation.

[0029] FIG. 6 is a diagram showing a specific example of a configuration for acquiring the training data described above. FIG. 6 shows an outline of a communication system 221 used to acquire the training data. In the communication system 221, an optical signal having an input electric field waveform E_0 is transmitted from a transmitter Tx to a communication path, and the optical signal is transmitted to a receiver Rx. The communication system 221 has multiple (N) unit sections 25. Each unit section 25 is provided with an optical fiber as a transmission path, the length of which is represented by L. The electric field waveform of the optical signal transmitted from the transmitter Tx is represented by electric field waveform E_0. This electric field waveform E_0 of the optical signal is the input electric field waveform in the first section corresponding to the first unit section 25. The input electric field waveform and input light intensity in the first section are input electric field waveform E_0 and input light intensity P_1. The optical signal passes through an optical path with a fiber length L_1 in the first section and becomes an output electric field waveform E_1. This output electric field waveform E_1 is input to the next unit section 25 (second section) as the input electric field waveform E_1.

[0030] In the nth unit section 25 (nth section), when the input electric field waveform E_n-1, input light intensity P_n, and fiber length L_n are given, the optical signal output from the optical fiber is the output electric field waveform E_n. The output electric field waveform E_n may be obtained by transmission simulation, by actual device verification, or by other means. The feature acquisition unit 26 acquires an input feature A_n-1 of the input electric field waveform E_n-1 and an output feature A_n of the output electric field waveform E_n. A plurality of sets of data, each set having the input feature A_n-1, input light intensity P_n, and fiber length L_n as explanatory variables and the output feature A_n as the correct label of the objective variable, are used as first training data (dataset 1).

[0031] Furthermore, the receiver Rx receives an optical signal having an output electric field waveform E_N output from the Nth unit section 25 at a received optical intensity P_Rx. The feature acquisition unit 26 acquires an output feature A_N of the output electric field waveform E_N. A plurality of sets of data, each set having the feature A_N and the optical intensity P_N as explanatory variables and the transmission quality (e.g., the value of the bit error rate) as the correct label of the objective variable, are used as second training data (data set 2).

[0032] By performing supervised learning using such datasets 1 and 2, the above-mentioned trained model is obtained. The trained model obtained by the learning process using dataset 1 is used in each waveform calculation unit 82. The trained model obtained by the learning process using dataset 2 is used in the transmission quality calculation unit 83. The waveform calculation units 82 shown in FIG. 3 are cascaded, for example, in the same number as the number of transmission fibers or unit sections in the optical communication path to be processed, and are finally connected to the transmission quality calculation unit 83. Using such a model, it is possible to estimate the transmission quality in the optical communication path to be processed.

[0033] FIG. 7 illustrates a specific example of the configuration of the feature acquisition unit 26. The feature An obtained by the configuration illustrated in FIG. 7 is associated one-to-one with the electric field waveform E_n. The feature acquisition unit 26 includes a transmitter 261, a VNLF 262, and an LMS 263. The transmitter 261 virtually generates an optical signal transmitted from an actual communication device (transmitter Tx) and outputs it to the VNLF to simulate a phenomenon. The VNLF 262 is a nonlinear filter. The transmitter 261 outputs an ideal transmission electric field waveform E_ideal to the VNLF 262 based on an arbitrary code sequence. The ideal transmission electric field waveform may be, for example, a state in which the eye opening of the eye pattern is large. The transfer function of the VNLF 262 is represented by tap coefficients. The LMS 263 updates the tap coefficients of the VNLF using an LMS algorithm to reduce the error between E_n and E_out. The LMS 263 acquires and outputs the tap coefficient when the error between E_n and E_out converges to a predetermined threshold value (e.g., zero) or less as a feature A_n. When such a tap coefficient is used as a feature, the feature and the electric field waveform are associated one-to-one. Therefore, the original electric field waveform can be uniquely derived from the feature.

[0034] FIG. 8 is a diagram showing an outline of the configuration of the electric field waveform derivation unit 27. The electric field waveform derivation unit 27 derives an electric field waveform from a feature quantity. The electric field waveform derivation unit 27 includes a transmitter 271 and a VNLF 272. A feature quantity A_n is input to the electric field waveform derivation unit 27. A tap coefficient indicated by the input feature quantity A_n is set in the VNLF 272. The transmitter 271 outputs E_ideal to the VNLF 272. The output of the VNLF 272 is equal to E_n. As described above, the feature quantity of the electric field waveform can be acquired by using, for example, a nonlinear filter. However, other values ​​may be used as the feature quantity of the electric field waveform.

[0035] Next, a system that uses the above-described model to estimate the transmission quality of an optical communication path to be processed will be described. Fig. 9 is a diagram showing an example of the system configuration of a transmission quality estimation system. The transmission quality estimation system 100 includes a learning device 20 and an estimation device 30. The learning device 20 generates a trained model by executing a learning process using the above-described data set 1 and data set 2. The estimation device 30 estimates the transmission quality of the optical path to be processed using the trained model generated by the learning device 20.

[0036] 10 is a schematic block diagram showing a specific example of the functional configuration of the learning device 20. The learning device 20 is configured using an information processing device such as a personal computer or a server device. The learning device 20 includes a storage unit 21 and a control unit 22.

[0037] The storage unit 21 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 21 stores data used by the control unit 22. The storage unit 21 may function as, for example, a teacher data storage unit 211 and a trained model storage unit 212.

[0038] The teacher data storage unit 211 stores teacher data used in the learning process executed in the learning device 20. The teacher data stored in the teacher data storage unit 221 is the above-mentioned Data Set 1 and Data Set 2. The trained model storage unit 212 stores trained models obtained by the learning process using the teacher data stored in the teacher data storage unit 211.

[0039] The control unit 22 is configured using a processor such as a CPU and a memory. The control unit 22 functions as an information control unit 221 and a learning control unit 222 by the processor executing a program. Note that all or part of the functions of the control unit 22 may be realized using hardware such as an ASIC, PLD, or FPGA. The above program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, and a semiconductor storage device (e.g., an SSD), as well as storage devices such as a hard disk or semiconductor storage device built into a computer system. The above program may be transmitted via a telecommunications line.

[0040] The information control unit 221 controls the input and output of information. For example, the information control unit 221 acquires teacher data from another device (an information processing device or a storage medium) and records it in the teacher data storage unit 211. In this case, the other device generates the teacher data (data set 1 and data set 2). The information control unit 221 may generate either or both of data set 1 and data set 2. In this case, the information control unit 221 may acquire the dataset by simulating a communication system 221 as shown in FIG. 6 , or may acquire the dataset by functioning as the feature acquisition unit 26 for values ​​obtained from an actual device as shown in FIG. 6 to acquire features. When acquiring a dataset as described above, the information control unit 221 also functions as the feature acquisition unit 26 that acquires features from an electric field waveform. The information control unit 221 may output, for example, a trained model stored in the trained model storage unit 212 to another device (e.g., the estimation device 30).

[0041] The learning control unit 222 executes a learning process using the training data stored in the training data storage unit 211. Specific examples of such learning processes include supervised learning for classification, such as support vector machines, random forests, and neural networks. The learning control unit 222 generates a trained model for outputting an output feature amount A_n based on the input feature amount A_n-1, input light intensity P_n, and fiber length L_n, for example, by performing supervised learning. The learning control unit 222 records the generated trained model in the trained model storage unit 212. The trained model obtained by the learning control unit 222 is provided to the estimation device 30. The trained model may be provided by communication via a network, via a recording medium, or by any other method.

[0042] 11 is a flowchart showing a specific example of processing by the learning device 20. First, the information control unit 221 acquires training data (step S101). The training data may be input by a user, acquired via communication from another information device, or acquired from a recording medium connected to the learning device 20, for example. The learning control unit 222 executes a training process using the training data and records a trained model in the trained model storage unit 212 (step S102).

[0043] 12 is a schematic block diagram showing a specific example of the functional configuration of the determination device 30. The determination device 30 is configured using an information processing device such as a personal computer or a server device. The determination device 30 includes a communication unit 31, a storage unit 32, and a control unit 33.

[0044] The communication unit 31 is a communication device. The communication unit 31 may be configured as, for example, a network interface. The communication unit 31 communicates data with other devices via a network in accordance with the control of the control unit 33. The communication unit 31 may be a device that performs wireless communication or a device that performs wired communication. The communication unit 31 may communicate with the controller 830 via a network, for example.

[0045] The storage unit 32 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 32 stores data used by the control unit 33. The storage unit 32 may function as, for example, an estimation model storage unit 321 and an estimation result storage unit 322.

[0046] The estimation model storage unit 321 stores an estimation model used by the waveform feature estimation unit 331 when performing the estimation process and an estimation model used by the transmission quality estimation unit 332 when performing the estimation process. The estimation model may be configured using information on a trained model generated in advance by a learning process, for example. Such a learning process may be performed by another device (e.g., the learning device 20) or by the device itself (the estimation device 30). Specific examples of estimation models include a trained model acquired by the learning device 20 performing a learning process using Dataset 1, and a trained model acquired by the learning device 20 performing a learning process using Dataset 2. The estimation model does not necessarily have to be generated by a learning process. The estimation model may be configured using, for example, a lookup table that associates the explanatory variables with the target variables described above, or may be configured in another manner. The estimation result storage unit 322 stores the estimation results of the waveform feature estimation unit 331 and the transmission quality estimation unit 332.

[0047] The control unit 33 is configured using a processor such as a CPU and a memory. The control unit 33 functions as an information control unit 331, a feature acquisition unit 332, a waveform feature estimation unit 333, and a transmission quality estimation unit 334 by the processor executing a program. Note that all or part of the functions of the control unit 33 may be realized using hardware such as an ASIC, a PLD, or an FPGA. The above program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, and a semiconductor storage device (e.g., an SSD), as well as storage devices such as a hard disk or semiconductor storage device built into a computer system. The above program may be transmitted via a telecommunications line.

[0048] The information control unit 331 acquires data from other devices such as the controller 830. The information control unit 331 transmits information indicating the estimation result obtained by the transmission quality estimation unit 334 to other devices such as the controller 830. Such exchange of information between the information control unit 331 and other devices may be performed by communication using the communication unit 31, for example.

[0049] The feature acquisition unit 332 acquires feature quantities corresponding to the input electric field waveform in the optical communication path to be estimated. The feature acquisition unit 332 may be configured similarly to the feature acquisition unit 26 shown in FIG. 7 . The feature acquisition unit 332 may acquire the initial electric field waveform E_0 transmitted from the transmitter in any manner. For example, the feature acquisition unit 332 may be stored in advance as a fixed value, or may be acquired from the controller 830. The acquired electric field waveform E_0 may also be stored in advance in association with the type (e.g., model number) of the transmitter in the optical communication path to be processed. Such associated information may be stored in the storage unit 32 of the determination device 30 or in the controller 830. In this case, the controller 830 may determine the type of transmitter in the optical communication path to be processed and acquire the electric field waveform E_0 based on the determined information, or may notify the determination device 30 of the transmitter type determination result. Upon receiving the notification, the feature acquisition unit 332 of the determination device 30 may acquire the electric field waveform E_0 corresponding to the transmitter type from the storage unit 32. Furthermore, instead of the electric field waveform E_0 described above, the determination device 30 or the controller 830 may store in advance a feature A_0 generated in accordance with the electric field waveform E_0. In this case, the feature A_0 may also be stored in advance in association with the type of transmitter.

[0050] FIG. 13 is a diagram illustrating an outline of the processing of the waveform feature estimation unit 333 and the transmission quality estimation unit 334. The waveform feature estimation unit 333 uses a trained model to estimate a feature A_1 of an electric field waveform E_1 output from the first unit section 25 based on a feature A_0 of an optical signal transmitted by a transmitter in the optical communication path to be estimated. The waveform feature estimation unit 333 estimates features across all N sections in the optical communication path to be estimated, and estimates a feature A_N of an electric field waveform E_N that is finally output. The waveform feature estimation unit 333 performs estimation by acquiring values ​​from the controller 830 according to explanatory variables of the trained model. For example, as shown in FIG. 13 , the input light intensity P and the length L of the optical fiber may be further used as explanatory variables. The waveform feature estimation unit 333 outputs a feature A_N of an electric field waveform E_N of an optical signal that is finally output in the optical communication path to be estimated to the transmission quality estimation unit 334.

[0051] The transmission quality estimation unit 334 estimates the transmission quality of the optical communication path to be estimated using the trained model based on the feature acquired by the feature acquisition unit 332. The transmission quality estimation unit 334 records the estimation result in the estimation result storage unit 322. The transmission quality estimation unit 334 performs estimation by acquiring values ​​from the controller 830 according to the explanatory variables of the trained model. For example, as shown in FIG. 13 , the input optical intensity P_Rx may be further used as an explanatory variable.

[0052] 14 is a flowchart showing a specific example of processing by the estimation device 30. First, the information control unit 331 acquires optical path information (e.g., the distance L of each section, the input light intensity P, and the received light intensity P_Rx) of the optical communication path to be estimated from the controller 830 (step S201). The waveform feature amount estimation unit 333 estimates a feature amount A_N of the optical communication path to be estimated. The transmission quality estimation unit 334 estimates the transmission quality using the feature amount A_N (step S202). The information control unit 331 transmits information indicating the estimation result to the controller 830 (step S203).

[0053] In the transmission quality estimation system 100 configured as described above, when estimating the transmission quality of an optical communication path to be estimated, the feature amount A is used instead of the electric field waveform E of the optical signal. The number of dimensions of the feature amount A is lower than that of the electric field waveform E. This makes it possible to reduce the amount of calculation required in the estimation process.

[0054] [Second Embodiment] In the first embodiment, it was assumed that the optical fiber was of a single type. However, to compensate for chromatic dispersion in order to extend the transmission distance of IMDD transmission, different types of fiber, such as dispersion compensating fiber (DCF), may be inserted in some sections. The output electric field and its characteristic quantities vary depending not only on the input optical power P and the transmission distance L but also on the type of optical fiber. Therefore, in a system in which different optical fibers are mixed, it is necessary to estimate the transmission quality depending on the type of each optical fiber. For example, in order to take into account the magnitude of chromatic dispersion of each optical fiber, the value of the group velocity dispersion parameter β_2 must be taken into account.

[0055] Furthermore, in the first embodiment, the bit error rate is used as a specific example of transmission quality, but in the second embodiment, the Q value is estimated as a specific example of transmission quality.

[0056] 15 is a diagram showing a network model of an all-optical network 900, the transmission quality of which is estimated by the estimation device 30 of the second embodiment. In the all-optical network 900 of the second embodiment, dispersion compensation fiber is used only in the unit section where n=2, and single-mode fiber (SMF) is used in the other unit sections. Therefore, the values ​​of β_2-1, β_2-3 to β_2-N are the same, but these values ​​are different from the value of β_2-2.

[0057] 16 is a diagram showing an outline of the processing of the estimation device 30 in the second embodiment. The estimation device 30 in the second embodiment calculates a feature amount A_n of an output electric field waveform E_n in a certain section based on the feature amount of an input electric field waveform E_n-1 for that section, the light intensity P_n, the transmission distance L_n, and the value of β_2.

[0058] 17 is a diagram showing an example of a neural network, which is a specific example of machine learning used in the waveform calculation unit 82 of the second embodiment. Input features and attribute information related to transmission are input to the learning model as explanatory variables, and the attribute information related to transmission includes the value of β_2.

[0059] Fig. 18 is a diagram showing an example of a neural network, which is a specific example of machine learning used in the transmission quality calculation unit 83 of the second embodiment. In the second embodiment shown in Fig. 18, a Q value is used as a specific example of information related to transmission quality.

[0060] In the transmission quality estimation system 100 of the second embodiment configured as described above, when estimating the transmission quality of the optical communication path to be estimated, it is possible to more accurately estimate the transmission quality even in a system in which different optical fibers are mixed.

[0061] FIG. 19 is a diagram illustrating an outline of an example hardware configuration of an information processing device 90 applied to this embodiment. The information processing device 90 includes a processor 91, a main storage device 92, a communication interface 93, an auxiliary storage device 94, an input / output interface 95, and an internal bus 96. The processor 91, the main storage device 92, the communication interface 93, the auxiliary storage device 94, and the input / output interface 95 are communicably connected to each other via the internal bus 96. The information processing device 90 may be applied to, for example, the learning device 20 and the estimation device 30. In this case, for example, the communication unit 31 may be configured using the communication interface 93. For example, the memory units 21 and 32 may be configured using the auxiliary storage device 94. Furthermore, the control unit 22 and the control unit 33 may be configured using the processor 91 and the main storage device 92.

[0062] (Variations) In the above description, the learning device 20 and the estimation device 30 are configured as separate devices, but they may be configured as an integrated device. The learning device 20 may be implemented using multiple information processing devices. For example, the learning device 20 may be implemented using a device such as a cloud. For example, in the learning device 20, the memory unit 21 and the control unit 22 may be implemented in different information processing devices. For example, the memory unit 21 of the learning device 20 may be distributed and implemented across multiple information processing devices. The estimation device 30 may be implemented using multiple information processing devices. For example, the estimation device 30 may be implemented using a device such as a cloud. For example, in the estimation device 30, the memory unit 32 and the control unit 33 may be implemented in different information processing devices. For example, the memory unit 32 of the estimation device 30 may be distributed and implemented across multiple information processing devices.

[0063] Although an embodiment of the present invention has been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs within the scope of the gist of the present invention.

[0064] The present invention is applicable to techniques for estimating transmission quality.

[0065] 100... estimation system, 20... learning device, 21... storage unit, 22... control unit, 30... estimation device, 21... communication unit, 32... storage unit, 33... control unit

Claims

1. An estimation device comprising: an optical communication path formed by connecting one or more unit intervals; an optical signal transmitted from a transmitter and input to the optical communication path; and a waveform feature quantity estimation unit that estimates a feature quantity of an electric field waveform output from the unit interval based on a feature quantity of the electric field waveform of the optical signal input to the unit interval for each unit interval constituting the optical communication path; and a transmission quality estimation unit that estimates the transmission quality of the optical communication path based on the feature quantity of the electric field waveform output from the last unit interval of the optical communication path estimated by the waveform feature quantity estimation unit.

2. The estimation device according to claim 1, wherein the waveform feature quantity estimation unit estimates the feature quantity using a learned model obtained by performing learning processing with the feature quantity of the electric field waveform of the input optical signal as an explanatory variable and the feature quantity of the electric field waveform of the output optical signal as an objective variable.

3. The estimation device according to claim 1, wherein the transmission quality estimation unit estimates the transmission quality using a learned model obtained by performing learning processing with the feature quantity of the electric field waveform output from the last unit interval of the optical communication path as an explanatory variable and the transmission quality of the optical communication path as an objective variable.

4. The estimation device according to claim 2, wherein the waveform feature quantity estimation unit estimates the feature quantity using a learned model with at least one of the optical intensity of the input optical signal and the fiber length of the unit interval as an additional explanatory variable.

5. The estimation device according to claim 3, wherein the transmission quality estimation unit estimates the transmission quality using a learned model with the optical intensity of the input optical signal as an additional explanatory variable.

6. A learning device comprising: an optical communication path formed by connecting one or more unit intervals; an optical signal transmitted from a transmitter and input to the optical communication path; and a learning control unit that generates a learned model by performing learning processing using a plurality of teacher data with the feature quantity of the electric field waveform of the optical signal input to the unit interval as an explanatory variable and the feature quantity of the electric field waveform output from the unit interval as an objective variable for each unit interval constituting the optical communication path.

7. A learning device comprising a learning control unit that generates a learned model by performing learning processing using a plurality of teacher data, where a feature amount of an electric field waveform output from the last unit interval of an optical communication path is an explanatory variable, and the transmission quality of the optical communication path is an objective variable, with respect to an optical communication path constituted by connecting one or more unit intervals and an optical signal transmitted from a transmitter and input to the optical communication path.

8. An estimation method having a waveform feature amount estimation step of estimating a feature amount of an electric field waveform output from a unit interval based on a feature amount of an electric field waveform of the optical signal input to the unit interval for each unit interval constituting the optical communication path, with respect to an optical communication path constituted by connecting one or more unit intervals and an optical signal transmitted from a transmitter and input to the optical communication path; and a transmission quality estimation step of estimating the transmission quality of the optical communication path based on the feature amount of the electric field waveform output from the last unit interval of the optical communication path estimated by the waveform feature amount estimation unit.

Images