Extraction device and extraction method

Abstract

This extraction device comprises: a preprocessing unit that divides, for each symbol period, the locus of points determined on the basis of a periodically modulated electric field waveform; and a distribution processing unit that generates, as the feature amount of the electric field waveform, the frequency distribution of the points and / or the locus. The preprocessing unit may separate the electric field waveform into at least one of an intensity waveform, a real part waveform, and an imaginary part waveform. The points may be sample points of intensity, a real part component, or an imaginary part component. The points may be sample points set on a complex plane or a Poincare sphere representing the real part waveform and the imaginary part waveform.

Description

Extraction apparatus and extraction method 【0001】 The present invention relates to an extraction apparatus and an extraction method. 【0002】 To increase the capacity, reduce latency, and lower the power consumption of optical communication systems, the All-Photonics Network (APN) is being investigated (see Non-Patent Document 1). In an All-Photonics Network, optical nodes relay optical signals as optical signals, thereby directly connecting user devices on opposing networks using optical signals. 【0003】 When a user device is newly connected to the all-photonics network and a new route is requested between the user devices, the control unit assigns optical transmission paths that satisfy a specified code error rate as the new route between the user devices. Alternatively, the control unit could select an optical transmission path that satisfies the specified code error rate by actually transmitting optical signals through each path and trial and error while changing the parameters of each optical transmission path. However, route design requires an enormous amount of time. 【0004】 One method for designing routes in a short time is based on a technique for estimating the transmission quality of optical signals (see Non-Patent Literature 2). In this method, the transmission quality is estimated in a short time for each candidate new route. Transmission quality is expressed using an index that indicates the communication quality in an optical communication system. Examples of indicators of communication quality include the code error rate, the Q (Quality factor) value, and the signal-to-noise ratio (SNR). For example, if the specified value for the code error rate is met, it may be determined that the specified value for transmission quality is met. By having the control device assign optical transmission paths that meet the specified value for transmission quality as new routes between user devices, the route design time is shortened. 【0005】Non-patent document 2 discloses a method for estimating transmission quality in a short time using a pre-trained machine learning model (artificial neural network). In non-patent document 2, each pre-trained model corresponding to the optical transmission path and receiving device is cascaded. The feature quantities of the electric field waveform propagate through each of the cascaded pre-trained models. The code error rate in the pre-trained model that ultimately acquires the electric field waveform is estimated as the transmission quality. 【0006】 In an all-photonics network, user devices located far apart from each other are directly connected optically via one or more optical nodes. Each optical node is equipped with an optical switch. The optical switch relays optical signals as optical signals without performing photoelectric conversion. Here, the control unit estimates the code error rate for each candidate new path using a technique for estimating the transmission quality of optical signals. 【0007】 The control unit selects a path from among the candidate new paths that satisfies a specified value for the code error rate. The control unit assigns the selected path as the path between user devices. Based on the path between user devices, the control unit designs the optical signal path at the optical nodes. Here, the control unit outputs optical transmission path information to the estimation unit. The optical transmission path information includes, for example, the transmission distance L between each optical node (span), the intensity P of the optical signal input to each optical node, and the intensity P of the optical signal at the receiving user device. ＲＸ The estimation unit outputs the estimated code error rate, based on the optical transmission path information, as the transmission quality to the control device for each candidate new path. 【0008】 In optical transmission lines with multiple spans, optical amplifiers are inserted between the spans. These optical amplifiers compensate for optical fiber losses in the optical signal. The optical signal output from the transmitting user equipment (transmitter) is input into the optical fiber. The optical fiber outputs the optical signal to the next span. Through this repetition, the optical signal is finally received by the receiving user equipment (receiver). 【0009】S. Kaneko, M. Yoshino, N. Shibata, R. Igarashi, J. Kani, and T. Yoshida, "Photonic gateway accommodating all types of wavelength paths for digital-coherent and IM-DD user terminals in all-photonic metro-access converged networks," in Journal of Optical Communications and Networking, vol. 16, no. 3, pp. 304-316, March 2024.Ryo Igarashi, Ryo Koma, Kazutaka Hara, Jun-ichi Kani, and Tomoaki Yoshida “Fast QoT estimation method using cascaded artificial neural network for real-time path provisioning in IMDD based all-optical networks “, in Optics Express, vol. 32, no.2, pp. 1176-1187, January 2024. 【0010】 For transmission quality to be estimated, feature quantities of the electric field waveform must be extracted from that waveform. In Non-Patent Document 2, the electric field waveform, which is the target of feature quantity extraction, is input to a nonlinear filter (not shown). The transfer function of the nonlinear filter is determined according to the tap coefficients. The tap coefficients are determined by a tap coefficient calculation unit (not shown). 【0011】 The tap coefficient calculation unit updates the tap coefficients, which reduce the difference between the electric field waveform output from the nonlinear filter and a predetermined reference electric field waveform, using an algorithm such as the least mean square (LMS) method. The tap coefficient calculation unit outputs the updated tap coefficients to the nonlinear filter. The nonlinear filter updates the electric field waveform output from the nonlinear filter based on the updated tap coefficients. 【0012】 The tap coefficient calculation unit updates the tap coefficients until the difference between the electric field waveform output from the nonlinear filter and the reference electric field waveform falls below a threshold. The final obtained tap coefficients are uniquely determined for the electric field waveform input to the nonlinear filter. Therefore, the final obtained tap coefficients can be used as feature quantities for the electric field waveform input to the nonlinear filter. 【0013】 However, if the difference between the electric field waveform input to the nonlinear filter and the reference electric field waveform is large, the results of methods such as the least squares method may not converge, and the characteristic features of the electric field waveform may not be correctly extracted from it. Therefore, the reference electric field waveform must be predetermined based on the electric field waveform of the optical signal, the characteristics of the optical signal transmitter, the transmission distance of the optical signal, and the symbol pattern of the optical signal, so that the difference between the electric field waveform input to the nonlinear filter and the reference electric field waveform is small. 【0014】 Thus, there is a problem in that, in addition to the electric field waveform of the optical signal (optical electric field waveform), it is not possible to extract the characteristic quantities of the electric field waveform of an optical signal from the electric field waveform itself without additional information such as the characteristics of the optical signal transmitting device, the transmission distance of the optical signal, and the symbol pattern of the optical signal. 【0015】 In view of the above circumstances, the present invention aims to provide an extraction device and extraction method capable of extracting characteristic quantities of the electric field waveform of an optical signal solely from that electric field waveform. 【0016】 One aspect of the present invention is an extraction device comprising a preprocessing unit that divides the trajectory of a point determined based on a periodically modulated electric field waveform into symbol periods, and a distribution processing unit that generates a frequency distribution of at least one of the points and the trajectory as a feature quantity of the electric field waveform. 【0017】 One aspect of the present invention is an extraction method performed by an extraction device, comprising the steps of: dividing the trajectory of points determined based on a periodically modulated electric field waveform into symbol periods; and generating a frequency distribution of at least one of the points and the trajectory as a feature quantity of the electric field waveform. 【0018】 This invention makes it possible to extract characteristic features of the electric field waveform of an optical signal solely from that electric field waveform. 【0019】 This figure shows an example of the configuration of the communication system in the first embodiment. This figure shows an example of the configuration of the extraction unit in the first embodiment. This figure shows an example of the trajectory of intensity sample points in the first embodiment. This figure shows an example of the trajectory of sample points in the first embodiment, for each symbol period. This is an overview view showing an example of the frequency distribution of sample points in the first embodiment, for each sample phase. This figure shows an example of the frequency distribution of sample points in the first embodiment, for each sample phase. This figure shows an example of the normalized frequency distribution in the first embodiment, for each sample phase. This figure shows an example of the configuration of the learning system in the sixth embodiment. This flowchart shows an example of the operation of the extraction unit in the first embodiment. This figure shows an example of the configuration of the extraction unit in the second embodiment. This figure shows an example of the process of adding noise to the frequency distribution of light intensity in the second embodiment. This figure shows an example of the configuration of the extraction unit in the third embodiment. This figure shows an example of the trajectory of symbol points on the IQ plane in the third embodiment, for each symbol period. This is an overview view showing an example of the frequency distribution of trajectories in the third embodiment. This figure shows an example of the frequency distribution of trajectories in the third embodiment. This figure shows an example of the configuration of the extraction unit in the fourth embodiment. This figure shows an example of the process of removing the phase offset from the trajectory of the symbol point in the IQ plane in the fourth embodiment, for each symbol period. This is an overview view showing an example of the frequency distribution of the trajectory with the phase offset removed in the fourth embodiment. This figure shows an example of the frequency distribution of the trajectory with the phase offset removed in the fourth embodiment. This figure shows an example of the configuration of the extraction unit in the fifth embodiment. This figure shows an example of a symbol point defined on the Poincaré sphere in the fifth embodiment. This figure shows an example of the trajectory of the symbol point on the Poincaré sphere for the mth period in the fifth embodiment. This is an overview view showing an example of the frequency distribution of the trajectory of the symbol point on the Poincaré sphere in the fifth embodiment. This figure shows an example of the hardware configuration of the control device in each embodiment. 【0020】Embodiments of the present invention will be described in detail with reference to the drawings. (First Embodiment) Figure 1 is a diagram showing an example configuration of a communication system 1 in the first embodiment. The communication system 1 is a system that communicates using optical signals. The communication system 1 comprises a user device 2, an optical node 3, a control device 4, an optical node 5, a user device 6, and an optical transmission line 7. The communication system 1 may further include optical nodes and an optical transmission line. The communication system 1 may also further include a user device. 【0021】 Optical node 3 comprises a splitter 31, a measuring instrument 32, a communication interface 33, and an optical switch 34. Control device 4 comprises an extraction unit 41a, a candidate output unit 42, an estimation unit 43, a determination unit 44, and a communication interface 45. Optical node 5 comprises a communication interface 51 and an optical switch 52. Optical transmission path 7 comprises, for example, an optical fiber. Optical transmission path 7 may also include, for example, an optical amplifier. 【0022】 <User Device 2> User Device 2 is the communication device of the first user. In Figure 1, as an example, User Device 2 is newly connected to Optical Node 3. User Device 2 requests the control device 4 to open a new path between User Device 2 and User Device 6. In this case, User Device 2 receives the electric field waveform E of the optical signal. ０ This is transmitted to the measuring instrument 32 via the splitter 31. 【0023】 <Optical Node 3> Optical Node 3 is the first relay device for optical signals. The splitter 31 splits the optical signal (transmitted signal) transmitted from the user device 2 to the measuring instrument 32 and the optical switch 34. For example, the splitter 31 receives the electric field waveform E of the optical signal transmitted from the user device 2 that requested the opening of a new path. ０ The output is sent to the measuring instrument 32. The measuring instrument 32 receives the electric field waveform E of the optical signal transmitted from the user device 2. ０ The measuring instrument 32 measures the electric field waveform E of the transmitted optical signal. ０ This is output to the extraction unit 41a via the control channel. 【0024】The communication interface 33 acquires the internal connection information in the optical switch 34 from the communication interface 45 via the control channel. The optical switch 34 switches the internal connection (connection relationship between the input port and output port of the optical signal) in the optical switch 34 based on the internal connection information determined by the control device 4. 【0025】 <Control Device 4> The control device 4 designs the path of the optical signal. The control device 4 manages the path of the optical signal by controlling the internal connection of each optical node in the communication system 1 based on the designed path. For example, the control device 4 allocates an optical transmission path that satisfies the specified value of the transmission quality to the path of the optical signal between the user device 2 and the user device 6. 【0026】 The extraction unit 41a (extraction device) extracts the feature quantity A ０ of the electric field waveform E ０ from the electric field waveform E ０ For example, in the inference stage of machine learning, the extraction unit 41a uses the learned model to extract the feature quantity A ０ of the electric field waveform E ０ from the electric field waveform E ０ The extraction unit 41a extracts the feature quantity A ０ of the electric field waveform E ０ and outputs it to the determination unit 44. 【0027】 The candidate output unit 42 selects a plurality of path candidates as candidates for the new path based on the state information of each optical node and each optical transmission path in the communication system 1. For example, the candidate output unit 42 selects the optical node 3, the optical node 5, and the optical transmission path 7 as one of the path candidates. For the selected plurality of path candidates, the candidate output unit 42 outputs the transmission path configuration information of the path candidate (hereinafter referred to as "candidate information") to the determination unit 44. The candidate information includes, for example, the number N of spans, the length L ｎ of the optical transmission path of the nth span, the intensity P ｎ of the optical signal input to the nth span, and the intensity P ＲＸ of the optical signal received by the user device 6 after the Nth span for each path candidate. 【0028】 The estimation unit 43 is the feature quantity A of the electric field waveform E ０ ０and obtain, from the determination unit 44, candidate information about the selected path candidates. The estimation unit 43 estimates the transmission quality for each path candidate based on the feature quantity A ０ of the electric field waveform E ０ and the candidate information about the selected path candidates. The estimation unit 43 outputs the transmission quality of each path candidate to the determination unit 44. 【0029】 The determination unit 44 obtains the feature quantity A ０ of the electric field waveform E ０ from the extraction unit 41a. The determination unit 44 obtains candidate information about a plurality of selected path candidates from the candidate output unit 42. The determination unit 44 outputs the feature quantity A ０ of the electric field waveform E ０ and the candidate information about the plurality of selected path candidates to the estimation unit 43. The determination unit 44 selects an optical transmission path that satisfies a specified value of the transmission quality (e.g., bit error rate) from among the plurality of selected path candidates. 【0030】 The determination unit 44 determines the internal connection in the optical switch 34 and the internal connection in the optical switch 52 based on the selected optical transmission path. The communication interface 45 outputs the internal connection information in the optical switch 34 to the communication interface 33 via the control channel of the optical node 3. The communication interface 45 outputs the internal connection information in the optical switch 52 to the communication interface 51 via the control channel of the optical node 5. 【0031】 <Optical Node 5> The optical node 5 is a second relay device for optical signals. The communication interface 51 obtains, via the control channel, the internal connection information in the optical switch 52 from the communication interface 45. The optical switch 52 switches the internal connection (input port and output port of the optical signal) in the optical switch 52 based on the internal connection information determined by the control device 4. As a result, the optical signal path between the user device 2 and the user device 6 is opened. 【0032】<User Device 6> User device 6 is a communication device of the second user. User device 6 acquires optical signals from user device 2 via the opened path. User device 6 may also transmit optical signals to user device 2 via the opened path. 【0033】 Next, an example of the configuration of the extraction unit 41a will be described. Figure 2 shows an example of the configuration of the extraction unit 41a in the first embodiment. The extraction unit 41a comprises a pre-processing unit 411a, a distribution processing unit 412, a normalization unit 413, and a reduction unit 414. Note that the normalization unit 413 in the extraction unit 41a may be provided either before or after the distribution processing unit 412. 【0034】 The extraction unit 41a acquires an electric field waveform E (discrete waveform) containing arbitrary waveform distortion from the measuring instrument 32. The electric field waveform E is, for example, an intensity-modulated NRZ (Non-Return to Zero) signal. The electric field waveform E is represented as a complex time waveform. The extraction unit 41a outputs the feature quantity A of the electric field waveform E to the determination unit 44. 【0035】 Here, the preprocessing unit 411a separates the electric field waveform input from the measuring instrument 32 into the intensity trajectory (intensity time waveform), the real component trajectory (I-axis component), and the imaginary component trajectory (Q-axis component). That is, the preprocessing unit 411a separates the electric field waveform input from the measuring instrument 32 into the intensity time waveform (intensity waveform), the real component trajectory (real waveform), and the imaginary component trajectory (imaginary waveform). The intensity trajectory is calculated based on the square of the absolute value of the complex number (real component and imaginary component). 【0036】 Figure 3 shows an example of the trajectory of intensity sample points (sample point group) in the first embodiment. The preprocessing unit 411a performs a 5x oversampling on the intensity waveform 100 (intensity trajectory) as an example. The circles represent the starting point of the sample points in the mth period (first sample point). The triangles represent the points from the second to the fourth sample points in the mth period. The squares represent the ending point of the sample points in the mth period (fifth sample point). 【0037】Figure 4 shows an example of the trajectory 101 of the sample points in the first embodiment, divided by symbol period. The preprocessing unit 411a divides the trajectory 101 of the sample points (sample point group) determined based on the periodically modulated electric field waveform E into symbol periods. 【0038】 Figure 5 is an overview view showing an example of the frequency distribution of sample points in the first embodiment, for each sample phase. The distribution processing unit 412 generates a frequency distribution (probability density distribution) of at least one of the sample points and the trajectory of the sample points. 【0039】 Figure 6 shows an example of the frequency distribution of sample points in the first embodiment, for each sample phase. For example, the distribution processing unit 412 generates the frequency distribution of sample points as a feature A of the electric field waveform E. For example, the distribution processing unit 412 may generate the frequency distribution of the trajectory 101 of the sample points as a feature A of the electric field waveform E. For example, the distribution processing unit 412 may generate the frequency distribution of the sample points and the trajectory 101 of the sample points as a feature A of the electric field waveform E. The frequency distribution of sample points may be generated for each light intensity. 【0040】 Furthermore, the distribution processing unit 412 may generate frequency distributions for the locus of the real component (real part waveform) and the locus of the imaginary component (imaginary part waveform), similar to the procedure for generating the frequency distribution of the intensity locus (intensity waveform). Alternatively, feature quantities of the electric field waveform may be extracted from the electric field waveform based on the frequency distribution of the locus of the real component, the frequency distribution of the locus of the imaginary component, and the frequency distribution of the intensity locus. For example, for the same sample phase, feature quantities of the electric field waveform may be extracted from the electric field waveform based on the sum of the frequencies of the real component, the imaginary component, and the intensity. 【0041】Figure 7 shows an example of a normalized frequency distribution in the first embodiment, for each sample phase. When the distortion of the electric field waveform is used as a feature, if the distortion of the electric field waveform input to the extraction unit 41a is the same, the same feature should be output from the extraction unit 41a even if the intensity P or total number of periods M of the electric field waveform are different. Therefore, the normalization unit 413 normalizes the frequency distribution for each sample phase so that the integral value of the frequency becomes a predetermined value (for example, 1) for each symbol phase (symbol period). Alternatively, the normalization unit 413 may concatenate the frequency distributions of each sample phase as a one-dimensional array. 【0042】 To reduce the size of the trained model into which the features of the electric field waveform are input, the reduction unit 414 may reduce the dimensionality of the features using a dimensionality reduction method. The dimensionality reduction method is, for example, principal component analysis (PCA). The reduction unit 414 outputs the feature A to which the dimensionality reduction method has been applied to the determination unit 44. 【0043】 Next, a machine learning learning system will be described. Figure 8 shows an example configuration of the learning system 10 in the first embodiment. The learning system 10 is a system that generates a trained model using machine learning techniques. The learning system 10 comprises a generation device 20 and a learning device 30. The generation device 20 comprises a characteristic information generation unit 201 and an extraction unit 41a. The learning device 30 comprises a storage device 301 and a learning unit 302. 【0044】 When the estimation unit 43 estimates transmission quality using a machine learning model, that machine learning model needs to be generated in advance as a trained model. Therefore, during the machine learning training phase, the generation device 20 creates a set of training data including training data (transmission path configuration information) and correct labels (transmission quality information). 【0045】 Here, the characteristic information generation unit 201 may be the actual optical nodes and optical transmission lines in the communication system 1, or it may be a simulator (computer) that simulates the actual devices. The characteristic information generation unit 201 generates the characteristic information generation unit for each parameter (length L of the optical transmission line) used as learning data (explanatory variables). ｎ , the intensity of the optical signal Pｎ , and the intensity P of the optical signal ＲＸ The characteristic information generation unit 201 generates the electric field waveform E associated with each parameter used as learning data and the transmission quality information used as the correct label (target variable). The characteristic information generation unit 201 records each parameter used as learning data and the transmission quality information used as the correct label in the storage device 301. The characteristic information generation unit 201 generates the electric field waveform E associated with each parameter used as learning data. ｎ This is output to the extraction unit 41a. 【0046】 The extraction unit 41a receives the electric field waveform E input from the characteristic information generation unit 201. ｎ Feature A ｎ It converts to the following. The extraction unit 41a selects feature A as one of the parameters used as training data. ｎ This is recorded in the storage device 301. 【0047】 During the machine learning training phase, the learning device 30 generates a trained model using the set of training data generated by the generation device 20. Here, the storage device 301 stores each parameter (feature A) used as training data. ｎ , length L of the optical transmission path ｎ , the intensity of the optical signal P ｎ , and the intensity P of the optical signal ＲＸ The system stores the transmission quality information used as the correct label, along with the training data set. The learning unit 302 generates a trained model by performing a training process using the training data set stored in the storage device 301. The storage device 301 stores the generated trained model. 【0048】 Before starting the machine learning inference stage, the estimation unit 43 retrieves the trained model from the storage device 301. During the machine learning inference stage, the estimation unit 43 calculates the feature quantity A of the electric field waveform to be inferred. ｎ And the length L of the optical transmission path ｎ And the intensity of the light signal P ｎ And the intensity of the light signal P ＲＸ The data is input to the trained model. The estimation unit 43 obtains the transmission quality (e.g., the code error rate) in the optical signal path from the trained model. In this way, the estimation unit 43 estimates the transmission quality in the optical signal path. 【0049】 Next, an example of the operation of the extraction unit 41a in the communication system 1 will be described. Figure 9 is a flowchart showing an example of the operation of the extraction unit 41a in the first embodiment. The preprocessing unit 411a divides the trajectory of the sample points, which is determined based on the periodically modulated electric field waveform, into symbol periods (step S101). The distribution processing unit 412 generates the frequency distribution of the sample points as a feature quantity of the electric field waveform. The distribution processing unit 412 may also generate the frequency distribution of the trajectory of the sample points as a feature quantity of the electric field waveform (step S102). 【0050】 As described above, the preprocessing unit 411a separates the electric field waveform into at least one of the following: the intensity trajectory (intensity waveform), the real component trajectory (real waveform), and the imaginary component trajectory (imaginary waveform). The preprocessing unit 411a divides the point trajectories (discrete electric field waveforms) determined based on the periodically modulated electric field waveform into symbol periods. The points are sample points defined in the intensity waveform 100, the real component, or the imaginary component. The point trajectories are the time waveforms of the sample points. The distribution processing unit 412 generates a frequency distribution of at least one of the points and trajectories as a feature quantity of the electric field waveform. The distribution processing unit 412 generates a frequency distribution of sample points for each phase of the sample points (sample phase) defined in the time direction of the intensity waveform 100, the time direction of the real component, or the time direction of the imaginary component. The distribution processing unit 412 may generate a frequency distribution of sample points for each phase of a sample point (sample phase) defined in the direction of light intensity of the intensity waveform 100, the direction of the real component of the real waveform, or the direction of the imaginary component of the imaginary waveform. 【0051】 This makes it possible to extract the characteristic features of the electric field waveform of an optical signal solely from that electric field waveform. This also makes it possible to suppress the complexity of the configuration of communication system 1. Furthermore, it makes it possible to suppress an increase in the cost and power consumption of communication system 1. 【0052】 The normalization unit 413 may normalize the feature quantities of the electric field waveform. For example, the normalization unit 413 may normalize the integral value of the frequency distribution of sample points so that the integral value of the frequency distribution of sample points becomes a predetermined value. 【0053】(Second Embodiment) In the second embodiment, the main difference from the first embodiment is that noise is added to the frequency distribution of the sample points. The second embodiment will be explained focusing on the differences from the first embodiment. 【0054】 Figure 10 shows an example of the configuration of the extraction unit 41b in the second embodiment. The extraction unit 41b comprises a preprocessing unit 411b, a distribution processing unit 412, a normalization unit 413, a reduction unit 414, and a noise unit 415. 【0055】 The frequency distribution (features) changes according to the signal-to-noise ratio (SNR) of the electric field waveform. Therefore, it is desirable that the set of training data used in subsequent machine learning training phases be easily generated so that features of electric field waveforms containing various types of noise can be extracted during the inference phase of machine learning. Accordingly, the noise unit 415 receives the frequency distribution (discretized probability density distribution) and a function representing the noise (for example, a Gaussian function) as input from the distribution processing unit 412. The amount of noise is, for example, the noise intensity σ ２ It is expressed using 【0056】 Figure 11 shows an example of the process of adding noise to the frequency distribution of light intensity in the second embodiment. The noise unit 415 adds the input noise to the input frequency distribution (discretized probability density distribution). For example, the noise unit 415 convolves the input noise (a function representing the noise) onto the input frequency distribution. The noise unit 415 outputs the frequency distribution with the added noise to the normalization unit 413. 【0057】 As described above, the noise unit 415 may add noise to the frequency distribution of sample points (discretized probability density distribution). For example, the noise unit 415 may perform a convolution operation of a predetermined function (e.g., a Gaussian function) on the frequency distribution of sample points. 【0058】 This makes it possible to extract feature characteristics from the electric field waveform of an optical signal solely from that waveform. Furthermore, feature characteristics from electric field waveforms containing various types of noise can be extracted during the inference stage of machine learning. 【0059】(Third Embodiment) In the first and second embodiments, the frequency distribution of sample points in the intensity waveform was determined for each sample phase. In the first embodiment, the frequency distribution of sample points for the in-phase component trajectory (real part waveform) or the quadrature-phase component trajectory (imaginary part waveform) may also be determined for each sample phase. 【0060】 In contrast, the main difference between the third embodiment and the first and second embodiments is that the frequency distribution of the trajectories of symbol points in the IQ (in-phase and orthogonal-phase) plane (complex plane) is extracted from the electric field waveform as a feature. The third embodiment will be explained focusing on the differences from the first and second embodiments. 【0061】 Figure 12 shows an example of the configuration of the extraction unit 41c in the third embodiment. The extraction unit 41c comprises a preprocessing unit 411c, a distribution processing unit 412, a normalization unit 413, and a reduction unit 414. 【0062】 The electric field waveform E is, for example, a signal with four values. A signal with four values ​​is, for example, a Quadrature Phase Shift Keying (QPSK) signal. The preprocessor 411a separates the electric field waveform input from the measuring instrument 32 into the trajectory of the real component (time waveform) and the trajectory of the imaginary component (time waveform). The preprocessor 411c plots the symbol points and the trajectories of the symbol points on the IQ plane based on the trajectories of the real component and the imaginary component. 【0063】 Figure 13 shows an example of the trajectory of a symbol point in the IQ plane in the third embodiment, for each symbol period. The circles represent the symbol point corresponding to the starting point of the sample point in the m-th period (the first sample point). The squares represent the symbol point corresponding to the ending point of the sample point in the m-th period (the fifth sample point). The dashed lines represent the trajectory of the symbol point. The preprocessing unit 411c divides the trajectory of the symbol point, which is determined based on the periodically modulated electric field waveform E, for each symbol period. 【0064】Figure 14 is an overview view showing an example of the frequency distribution of trajectories in the third embodiment. The distribution processing unit 412 generates a frequency distribution (probability density distribution) of at least one of the symbol points and the trajectories of the symbol points. 【0065】 Figure 15 shows an example of the frequency distribution of the trajectory in the third embodiment. For example, the distribution processing unit 412 generates the frequency distribution of the symbol points as feature quantity A of the electric field waveform E. For example, the distribution processing unit 412 may generate the frequency distribution of the trajectory 102 of the symbol points as feature quantity A of the electric field waveform E. For example, the distribution processing unit 412 may generate the frequency distribution of the symbol points and the trajectory 102 of the symbol points as feature quantity A of the electric field waveform E. The frequency distribution of the symbol points may be generated for each real component (I-axis component) or for each imaginary component (Q-axis component). 【0066】 As described above, the preprocessing unit 411c separates the electric field waveform into at least the locus of the real component and the locus of the imaginary component, which are the locus of the intensity, the locus of the real component, and the locus of the imaginary component. The preprocessing unit 411c divides the locus of points (discrete electric field waveform) determined based on the periodically modulated electric field waveform into symbol periods. The distribution processing unit 412 generates a frequency distribution of at least one of the points and locus as a feature quantity of the electric field waveform. The points are sample points defined in the complex plane (IQ plane) representing the real component and the imaginary component. The locus of points may be the time waveform of the sample points defined in the complex plane. The distribution processing unit 412 may generate a frequency distribution of sample points for each phase of the sample points in the time direction of the intensity waveform 100, the time direction of the real component waveform, or the time direction of the imaginary component waveform. The distribution processing unit 412 may generate a frequency distribution of sample points for each phase of a sample point (sample phase) defined in the direction of light intensity of the intensity waveform 100, the direction of the real component of the real waveform, or the direction of the imaginary component of the imaginary waveform. 【0067】 This makes it possible to extract characteristic features of the electric field waveform of an optical signal solely from that electric field waveform. 【0068】(Fourth Embodiment) In the fourth embodiment, the main difference from the first to third embodiments is that the phase offset of the symbol points in the IQ plane is removed. The fourth embodiment will be explained focusing on the differences from the first to third embodiments. 【0069】 The waveform distortion of the electric field waveform does not change with respect to the phase offset of the symbol point in the IQ plane. Therefore, since the code error rate does not change with respect to the phase offset, information about the phase offset is unnecessary for feature extraction. Thus, in the fourth embodiment, the phase offset of the symbol point is removed. 【0070】 Figure 16 shows an example of the configuration of the extraction unit 41d in the fourth embodiment. The extraction unit 41d comprises a preprocessing unit 411d, a distribution processing unit 412, a normalization unit 413, a reduction unit 414, and a removal unit 416. 【0071】 Figure 17 shows an example of the process for removing the phase offset from the trajectory of a symbol point in the IQ plane in the fourth embodiment, for each symbol period. In Figure 17, a phase offset of 45 degrees is given to the trajectory of the symbol point exemplified in Figure 13 as an example. The preprocessing unit 411d divides the trajectory of the symbol point, which is determined based on the electric field waveform E to which a 45-degree phase offset is given, for each symbol period. 【0072】 The removal unit 416 removes the phase offset included in the trajectory of the symbol point so that the phase of the symbol point (circle) corresponding to the starting point of the sample point in the m-th period becomes a predetermined phase (for example, 0 degrees). This makes it possible to obtain a constant frequency distribution even if the electric field waveform E input to the preprocessing unit 411d contains a phase offset. 【0073】 Figure 18 is an overview view showing an example of the frequency distribution of a trajectory with phase offset removed in the fourth embodiment. The distribution processing unit 412 generates a frequency distribution (probability density distribution) of at least one of the symbol points with phase offset removed and the trajectory of the symbol points with phase offset removed. 【0074】Figure 19 shows an example of the frequency distribution of a trajectory with phase offset removed in the fourth embodiment. For example, the distribution processing unit 412 generates the frequency distribution of symbol points with phase offset removed as feature quantity A of the electric field waveform E. For example, the distribution processing unit 412 may generate the frequency distribution of the trajectory 103 of symbol points with phase offset removed as feature quantity A of the electric field waveform E. For example, the distribution processing unit 412 may generate the frequency distribution of symbol points with phase offset removed and the trajectory 103 of symbol points with phase offset removed as feature quantity A of the electric field waveform E. 【0075】 As described above, the removal unit 416 may remove the phase offset of the symbol point trajectory. This makes it possible to extract the feature quantities of the electric field waveform of the optical signal solely from that electric field waveform. Furthermore, it is possible to suppress an increase in the number of training data sets. 【0076】 The removal unit 416 may be provided between the pre-processing unit 411a and the distribution processing unit 412 in the extraction unit 41a of the first embodiment. Similarly, the removal unit 416 may be provided between the pre-processing unit 411b and the distribution processing unit 412 in the extraction unit 41b of the second embodiment. Furthermore, the removal unit 416 may be provided between the pre-processing unit 411c and the distribution processing unit 412 in the extraction unit 41c of the third embodiment. 【0077】 (Fifth Embodiment) In the fifth embodiment, the main difference from the first to fourth embodiments is that the frequency distribution of the trajectories of the symbol points on the Poincaré sphere is extracted as a feature. The fifth embodiment will be explained focusing on the differences from the first to fourth embodiments. 【0078】 Figure 20 shows an example of the configuration of the extraction unit 41e in the fifth embodiment. The extraction unit 41e comprises a pre-processing unit 411e, a distribution processing unit 412, a normalization unit 413, and a reduction unit 414. The extraction unit 41e may also include a noise unit 415. The extraction unit 41e may also include a removal unit 416 prior to the distribution processing unit 412. 【0079】Figure 21 shows an example of symbol points defined in the Poincaré sphere 8 in the fifth embodiment. Each symbol point is arranged three-dimensionally in the Poincaré sphere 8. 【0080】 Figure 22 shows an example of the trajectory of the symbol point on the Poincaré sphere 8 in the fifth embodiment, for the mth period. The trajectory of the symbol point is determined by the time change of the position of the symbol point on the Poincaré sphere 8. 【0081】 Figure 23 is an overview view showing an example of the frequency distribution of the trajectories of symbol points on the Poincaré sphere 8 in the fifth embodiment. The distribution processing unit 412 generates a frequency distribution (probability density distribution) of at least one of the symbol points and the trajectories 104 of the symbol points for each symbol period. 【0082】 As described above, the preprocessing unit 411e separates the electric field waveform into at least the locus of the real component and the locus of the imaginary component, which are the locus of the intensity (intensity waveform), the locus of the real component (real waveform), and the locus of the imaginary component (imaginary waveform). The preprocessing unit 411e divides the locus of points (discrete electric field waveform) determined based on the periodically modulated electric field waveform into symbol periods. The distribution processing unit 412 generates a frequency distribution of at least one of the points and locus as a feature quantity of the electric field waveform. The points are sample points defined on the Poincaré sphere 8, which represent the locus of the real component and the locus of the imaginary component. The locus of points may also be the time waveform of the sample points defined on the Poincaré sphere 8. The distribution processing unit 412 may generate a frequency distribution of sample points for each phase of the sample points in the time direction of the intensity waveform 100, the time direction of the real waveform, or the time direction of the imaginary waveform. The distribution processing unit 412 may generate a frequency distribution of sample points for each phase of a sample point defined in the light intensity direction of the intensity waveform 100, the real component direction of the real waveform, or the imaginary component direction of the imaginary waveform (sample phase). Alternatively, the distribution processing unit 412 may generate a frequency distribution of sample points for each phase of a sample point defined in the S1 axis direction, the S2 axis direction, or the S3 axis direction. 【0083】 This makes it possible to extract characteristic features of the electric field waveform of an optical signal solely from that electric field waveform. 【0084】(Hardware Configuration) Figure 24 shows an example of the hardware configuration of the control device 4 in each embodiment. The control device 4 is implemented as software by a processor 11 such as a CPU (Central Processing Unit) executing a program stored in a storage device 12 having a non-volatile recording medium (non-temporary recording medium) and a memory 13. The program may be recorded on a computer-readable recording medium. A computer-readable recording medium is a non-temporary recording medium such as a portable medium such as a flexible disk, magneto-optical disk, ROM (Read Only Memory), CD-ROM (Compact Disc Read Only Memory), or a storage device such as a hard disk or solid-state drive (SSD) built into a computer system. The communication unit 14 performs predetermined communication processing. 【0085】 The control device 4 may be implemented via hardware including an electronic circuit (or circuitry) using, for example, an LSI (Large Scale Integrated Circuit), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). 【0086】 Although embodiments of this invention have been described in detail above with reference to the drawings, the specific configuration is not limited to these embodiments and includes designs and the like that do not depart from the spirit of this invention. Furthermore, each embodiment may be combined. 【0087】 This invention is applicable to optical communication systems. 【0088】1...Communication system, 2...User device, 3...Optical node, 4...Control device, 5...Optical node, 6...User device, 7...Optical transmission line, 8...Poincaré sphere, 10...Learning system, 20...Generator, 30...Learning device, 31...Splitter, 32...Measuring instrument, 33...Communication interface, 34...Optical switch, 41a, 41b, 41c, 41d, 41e...Extraction unit, 42...Candidate output unit, 43...Estimation unit, 44...Decision unit, 45...Communication interface, 51...Communication interface, 52...Optical switch, 100...Intensity waveform, 101...Trajectory, 102...Trajectory, 103...Trajectory, 104...Trajectory, 201...Characteristic information generation unit, 301...Storage device, 302...Learning unit, 411a, 411b, 411c, 411d, 411e...Preprocessing unit, 412...Distribution processing unit, 413...Normalization unit, 414...Reduction unit, 415...Noise unit, 416...Removal unit

Claims

1. An extraction device comprising: a preprocessing unit that divides the trajectory of a point, determined based on a periodically modulated electric field waveform, into symbol periods; and a distribution processing unit that generates a frequency distribution of at least one of the points and the trajectory as a feature quantity of the electric field waveform.

2. The extraction apparatus according to claim 1, wherein the preprocessing unit separates the electric field waveform into at least one of the locus of intensity, the locus of the real component, and the locus of the imaginary component, the point is a sample point of the intensity, the real component, or the imaginary component, and the locus of the point is the time waveform of the sample point.

3. The extraction apparatus according to claim 1, wherein the preprocessing unit separates the electric field waveform into at least the locus of the real component and the locus of the imaginary component from the locus of the intensity, the locus of the real component and the locus of the imaginary component, and the points are sample points defined on the complex plane or Poincaré sphere representing the real component and the imaginary component.

4. The extraction apparatus according to claim 3, wherein the locus of the point is the locus of the sample point in the complex plane or the Poincaré sphere.

5. The extraction apparatus according to claim 3, further comprising a removal unit for removing the phase offset contained in the trajectory of the real component and the trajectory of the imaginary component.

6. The extraction apparatus according to claim 1, further comprising a normalization unit for normalizing the feature quantities.

7. The extraction device according to claim 1, further comprising a noise unit that performs a process of convolving a predetermined function onto the feature quantity.

8. An extraction method performed by an extraction device, comprising the steps of: dividing the trajectory of a point determined based on a periodically modulated electric field waveform into symbol periods; and generating a frequency distribution of at least one of the points and the trajectory as a feature quantity of the electric field waveform.

Images