Optical signal processing device

Abstract

This eliminates the need for manual trial-and-error exploration to maximize task processing performance regarding how optical signals are input to the multiple input waveguides of an optical signal converter. [Solution] The optical signal processing device includes a light source that outputs coherent light, a demultiplexer that splits the light output from the light source, a plurality of modulators that modulate the phase of some or all of the divided light according to a control signal and output it as a plurality of phase-modulated light, a controller that generates a control signal based on input data, an optical signal converter that combines the plurality of phase-modulated light and outputs it as a plurality of converted light, a photoelectric converter that converts each of the plurality of converted light into an electrical signal, and a computer that outputs output data according to the input data based on the electrical signal. The controller has updatable internal parameters, and the internal parameters are set to generate a control signal that minimizes the error between the output data and the correct data.

Description

[Technical Field]

[0001] The disclosed technology relates to an optical signal processing device. [Background technology]

[0002] The following technologies are known regarding optical signal converters using silicon photonics technology. Patent Document 1 describes an optical signal converter comprising two or more optical signal input units for receiving optical signals, a wave combiner for combining the optical signals input from the two or more optical signal input units, and one or more optical signal output units for outputting newly generated optical signals within the wave combiner. The optical signals input from the optical signal input units are output from the optical signal output units as different optical signals due to differences in refractive index during propagation. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-80891 [Overview of the Initiative] [Problems that the invention aims to solve]

[0004] The optical signal converter described in Japanese Patent Publication No. 2022-80891 is manufactured primarily using silicon photonics technology, making it difficult to change specifications once a prototype is produced. An example of a specification that is difficult to change is the number of input sections (input waveguides) into which the optical signal is input.

[0005] For example, the dataset "Iris" (https: / / archive.ics.uci.edu / dataset / 53 / iris), used in the field of machine learning, is used to train an inference model that performs the task of classifying three types of irises from four types of features (petal length, petal width, sepal length, and sepal width). When performing this classification task using the optical signal converter described above, it is conceivable to input four optical signals, each corresponding to one of the four features, into the input section (input waveguide). In this case, the mismatch between the number of features (number of elements in the input data) and the number of input sections (input waveguides) becomes a problem. For example, if the number of input sections (input waveguides) is greater than the number of features, it is necessary to consider what kind of optical signals should be input into the surplus input sections.

[0006] In the optical signal converter described above, the performance varies greatly depending on how the optical signal is input. One possible approach is to manually find the input condition that yields the best performance (e.g., classification accuracy) among multiple input conditions. However, such manual trial-and-error search has its limitations, and becomes difficult when there are many input sections (input waveguides). Furthermore, since the optimal input conditions differ depending on the task, restarting the entire process every time the task changes is extremely costly.

[0007] The disclosed technology was developed in view of the above points, and aims to eliminate the need for manual trial-and-error exploration work to maximize the processing performance of a task regarding the method of inputting optical signals to the multiple input waveguides of an optical signal converter. [Means for solving the problem]

[0008] The optical signal processing device relating to the disclosed technology includes a light source that outputs coherent light, a demultiplexer that splits the light output from the light source, a plurality of modulators that modulate the phase of some or all of the divided light in accordance with a control signal and output it as a plurality of phase-modulated light, a controller that generates the control signal based on input data, an optical signal converter that combines the plurality of phase-modulated light and outputs it as a plurality of converted light, a photoelectric converter that converts each of the plurality of converted light into an electrical signal, and a computer that outputs output data corresponding to the input data based on the electrical signal. The optical signal converter includes a plurality of input waveguides that form input paths for the plurality of phase-modulated light, a combiner that combines the plurality of phase-modulated light, and a plurality of output waveguides that form output paths for the plurality of converted light. The controller has updatable internal parameters, and the internal parameters are set to generate the control signal such that the error between the output data and the ground truth data is minimized.

[0009] The number of input waveguides may differ from the number of elements in the input data. Preferably, the number of input waveguides is greater than the number of elements in the input data.

[0010] The input waveguide, the multiplexer, and the output waveguide may be made of a material with a higher refractive index than the surrounding material. This makes it possible to confine the light propagating inside the optical signal converter to the inside of the optical signal converter.

[0011] The multiplexing section may have internal structures that induce or enhance at least one of the following processes: refraction, scattering, reflection, and diffraction of light propagating within it. This allows for diversification and complexity of the multiplexing behavior within the multiplexing section.

[0012] The computer may derive the output data based on the relationship between the signal levels of the multiple electrical signals corresponding to each of the multiple converted optical signals. Reducing the processing load on the computer enhances the features of the optical signal processing device, namely low power consumption and high-speed processing.

[0013] The controller may generate the control signal based on a temperature signal indicating the temperature of at least one of the internal and external temperatures of the system. The controller may also generate the control signal based on a light intensity signal indicating the amount of light input to the modulator or the optical signal converter. By generating the control signal based on the light intensity signal and the temperature signal in addition to the input data, the controller can achieve higher accuracy of the output data.

[0014] The computer has updatable internal parameters, and during the process of updating the internal parameters of the controller, the computer's internal parameters may be set so as to minimize the error between the output data and the correct data. By configuring the computer to be trainable, the accuracy of the output data can be improved. By performing computer training in parallel with controller training, it is possible to maximize processing performance. [Effects of the Invention]

[0015] According to the disclosed technology, manual trial-and-error exploration is unnecessary to maximize the processing performance of a task regarding how optical signals are input to the multiple input waveguides of an optical signal converter. [Brief explanation of the drawing]

[0016] [Figure 1] This figure shows an example of the configuration of an optical signal processing apparatus according to an embodiment of the disclosed technology. [Figure 2] This is a microscopic image of an optical signal converter according to an embodiment of the disclosed technology. [Figure 3] This figure shows an example of a computer configuration according to an embodiment of the disclosed technology. [Figure 4] This figure shows an example of a controller learning method according to an embodiment of the disclosed technology. [Figure 5] This diagram schematically shows an example of the configuration of a typical signal processing device. [Figure 6] This figure shows the configuration of an optical signal processing apparatus according to an embodiment of the disclosed technology, in a format that allows for comparison with the configuration of the DNN model shown in Figure 5. [Figure 7] This figure shows an example of a computer learning method according to an embodiment of the disclosed technology. [Figure 8] This figure shows variations in the configuration of an optical signal converter according to an embodiment of the disclosed technology. [Figure 9] This figure shows variations in the configuration of an optical signal converter according to an embodiment of the disclosed technology. [Figure 10] This figure shows variations in the configuration of an optical signal converter according to an embodiment of the disclosed technology. [Figure 11] This figure shows an example of the configuration of an optical signal processing apparatus according to another embodiment of the disclosed technology. [Modes for carrying out the invention]

[0017] Hereinafter, an example of an embodiment of the disclosed technology will be described with reference to the drawings. In each drawing, identical or equivalent components and parts will be given the same reference numerals, and redundant descriptions will be omitted.

[0018] [First Embodiment] Figure 1 shows an example of the configuration of an optical signal processing device 10 according to a first embodiment of the disclosed technology. The optical signal processing device 10 includes a light source 20, a demultiplexer 30, a plurality of modulators 40, a controller 50, an optical signal converter 60, a photoelectric converter 70, and a computer 80.

[0019] The light source 20 outputs coherent light (CW light (Continuous Wave), coherent light). The light source 20 is, for example, a laser diode. The light output from the light source 20 is input to the demultiplexer 30.

[0020] The demultiplexer 30 splits the light from the light source 20 into multiple beams of light. The number of beams split by the demultiplexer 30 is typically the same as the number M of input waveguides 61 of the optical signal converter 60. The light intensity of the split beams may be the same or different. The split beams of light are transmitted along separate paths.

[0021] The modulator 40 is provided in each path of the plurality of lights split by the demultiplexer 30. The number of modulators 40 is typically the same as the number M of the input waveguides 61 of the optical signal converter 60. The modulator 40 is a phase shifter that shifts the phase of the light in each path. The amount of phase shift in the modulator 40 is the control signal S IN , , , C , IN ,

[0024] , IN , IN supplied from the controller 50. The amount of phase shift in each modulator 40 is usually different from the amount of phase shift in other modulators 40. The modulator 40 may shift the phase of the light by, for example, heat or an electric field.

[0022] The controller 50 is based on the input data D IN to generate a control signal S C for determining the amount of phase shift in the modulator 40, and supplies this to each of the plurality of modulators 40. The control signal S C is generated individually for each modulator 40. The number of control signals S C is typically the same as the number of modulators 40 and the same as the number M of the input waveguides 61 of the optical signal converter 60. If the processing performance of the tasks executed by the optical signal processing apparatus 10 can be sufficiently ensured, the controller 50 may generate the control signal S C only for some of the plurality of modulators 40. In this case, a fixed value may be supplied as the control signal S C to the modulators 40 not controlled by the controller 50. Also, a modulator 40 may not be provided in the path not controlled by the controller 50.

[0023] The control signal S <000000​​​​​​​​​​ The controller 50 has updatable internal parameters, and the internal parameters are appropriately set by machine learning, so that the input data D IN Appropriate control signal S C This makes it possible to generate [the controller]. Details of the machine learning for the controller 50 will be described later.

[0025] The light from each path modulated by the modulator 40 is divided into multiple phase-modulated light L M The signal is input to the optical signal converter 60. The optical signal converter 60 receives multiple phase-modulated optical signals L M The light generated by combining these waves is converted into multiple light sources L C The output is as follows. The optical signal converter 60 is manufactured, for example, using silicon photonics technology. Figure 2 is a microscopic image of the optical signal converter 60 manufactured using silicon photonics technology.

[0026] The optical signal converter 60 has a plurality of input waveguides 61, a multiplexer 62, and a plurality of output waveguides 63. The input waveguides 61, the multiplexer 62, and the output waveguides 63 are made of a material with a higher refractive index than the surrounding material. This makes it possible to confine the light propagating inside the optical signal converter 60 to the inside of the optical signal converter 60. For example, the input waveguides 61, the multiplexer 62, and the output waveguides 63 may each be made of Si, with SiO2 extending around them. The surrounding area of ​​the optical signal converter 60 may be air. As materials for the input waveguides 61, the multiplexer 62, and the output waveguides 63, for example, Si, SiN, SiO2, or LiNbO3 can be used. The materials for the input waveguides 61, the multiplexer 62, and the output waveguides 63 may be the same or different. For example, the input waveguide 61 may be made of SiN, and the multiplexer 62 and the output waveguides 63 may each be made of Si.

[0027] Each of the multiple input waveguides 61 has a phase-modulated optical fiber L for each path. M The input is received. Multiple input waveguides 61 receive multiple phase-modulated optical signals L MIt forms the input path. The number M of input waveguides 61 is typically the same as the number of light divisions in the demultiplexer 30, and the input data D IN It is more than the number of elements.

[0028] The wave combiner 62 is a single structural part in which multiple input waveguides 61 are connected at one end and multiple output waveguides 63 are connected at the other end. Phase-modulated light L is input to each of the multiple input waveguides 61. M These are combined in the multiplexing section 62. Multiple phase-modulated optical waves L M These waves propagate within the combined wave section 62 while interfering with each other.

[0029] Multiple output waveguides 63 have multiple phase-modulated optical waves L in the multiplexing section 62. M Multiple converted light Ls are generated when these are combined. C This forms the output path. The number of output waveguides 63 is typically less than the number M of input waveguides 61, but it may be the same number as or greater than the number M of input waveguides 61.

[0030] The path taken by the light propagating within the multiplexing section 62 to reach each output waveguide 63 is different for each output waveguide 63. Furthermore, the process of light interference within the multiplexing section 62 is different for each path. The converted light L propagates through the output waveguide 63. C The amount of light differs for each output waveguide 63. That is, the phase-modulated light L M The converted light L is formed when it propagates through the multiplexer 62. C The light and dark patterns are output via multiple output waveguides 63. These light and dark patterns are combinations of phase shift amounts in multiple modulators 40 (i.e., input data D IN It is formed based on ). The optical signal converter 60 can be said to convert the information carried by the optical signal from phase information to intensity information and output it. The amount of power required for the conversion process by the optical signal converter 60 is zero, and the conversion processing speed is the speed of light.

[0031] Multiple converted light beams L are output from each of the multiple output waveguides 63. CThe light is transmitted via separate paths and input to a photoelectric converter 70 provided in each path. Each of the photoelectric converters 70 receives the converted light L propagating along its corresponding path. C The photoelectric converter 70 converts the light L into an electrical signal V. C The photoelectric converter 70 may include a photodiode that generates a current corresponding to the amount of light, a current-to-voltage converter that converts the current into a voltage signal, and an analog-to-digital converter that converts the voltage signal into a digital signal. The electrical signals V output from each of the photoelectric converters 70 are input to the computer 80.

[0032] The computer 80 processes the input data D based on the electrical signal V. IN Output data D corresponding to this OUT It outputs the following. For example, if the optical signal processing device 10 is performing the Iris classification task, the input data D IN Output data D corresponding to the four elements that make up the flower (f1: petal length, f2: petal width, f3: sepal length, f4: sepal width). OUT This outputs one of the three types of irises.

[0033] Figure 3 shows an example of the configuration of computer 80. In Figure 3, computer 80 outputs data D, which is one of three types of irises (Iris_A, Iris_B, Iris_C) based on four electrical signals V1, V2, V3, and V4. OUT An example of output is shown. In this example, the computer 80 returns an argument (argmax) corresponding to the signal with the maximum level among the four electrical signals V1 to V4. The computer 80 has memory 82 that stores a table 81 which associates arguments X1 to X4 with the types of irises, and by referring to table 81, it outputs the type of iris corresponding to argmax as output data D OUT It outputs as follows. For example, when the signal level of electrical signal V2 takes its maximum value, Iris_B is output data D OUT It will be output as follows.

[0034] In this example, input data D shows the features corresponding to Iris_A. INWhen this is input, the electrical signal V1 takes its maximum value, and the input data D represents the feature corresponding to Iris_B. IN When this is input, the electrical signal V2 takes its maximum value, and the input data D represents the feature corresponding to Iris_C. IN If this is input, the electrical signal V3 will take its maximum value, and input data D will not correspond to any of Iris_A, Iris_B, or Iris_C. IN If this is input, the controller 50 learns that the electrical signal V4 takes its maximum value as "other".

[0035] Figure 3 illustrates the simplest configuration of the computer 80, but the computer 80 can be any configuration that performs some calculation on multiple electrical signals V and obtains the desired result. For example, the computer 80 outputs data D based on multiple electrical signals V. OUT It may include an inference model that outputs the above. The above inference model is built using machine learning.

[0036] Figure 4 shows an example of the learning method for controller 50. Controller 50 receives input data D IN Based on the control signal S C Generates the output voltage D from computer 80. OUT is the control signal S C The controller 50 has updatable internal parameters and output data D OUT and known correct data D C A control signal S equal to (or less than) the number of input waveguides 61, such that the error E is minimized. C The internal parameters are updated to generate the desired output. Updating the internal parameters corresponds to training the controller 50. The updated internal parameters, which enable the desired performance, are set as the final internal parameters. As a means of obtaining the optimal internal parameters, for example, a multivariate regression algorithm can be used. To suppress power consumption and system costs, it is preferable to use an algorithm with low computational cost.

[0037] An example of the operation of the optical signal processing device 10 is described below. The light source 20 outputs coherent light. The demultiplexer 30 divides the light output from the light source 20 into M parts. Each of the M divided light paths is provided with M modulators 40, which each control the light propagating along the corresponding path with a control signal S supplied from the controller 50. C Modulated based on this, and this is phase-modulated light L M It outputs as follows. The controller 50 receives input data D IN Based on this, M control signals S are used to control the modulation amount (phase shift amount) in the modulator 40. C These are generated and supplied to each of the M modulators 40.

[0038] M phase-modulated optical signals L are output from each of the M modulators 40. M The signal is input to the M input waveguides 61 of the optical signal converter 60. The optical signal converter 60 receives M phase-modulated optical signals L M The light generated by combining these is converted into N light beams L C The output is then generated from N output waveguides 63.

[0039] N conversion light L C Each of the N photoelectric converters 70 provided along the path transmits the converted light L that propagates along the corresponding path. C This is converted into an electrical signal V and supplied to the computer 80.

[0040] Computer 80 processes input data D based on N electrical signals. IN Output data D corresponding to this OUT The controller 50 has updatable internal parameters and outputs the output data D. OUT and correct data D C The control signal S that minimizes the error E between the two. C The internal parameters are updated to produce the desired result. The updated internal parameters, which allow for the desired performance, are set as the final internal parameters.

[0041] As described above, the optical signal processing device 10 according to the disclosed technology controls the modulation amount (phase shift amount) in the modulator 40 provided corresponding to each of the input waveguides 61 of the optical signal converter 60 using a control signal S C The controller 50 has a configuration that controls the output data D OUT and correct data D C A control signal S equal to (or less than) the number of input waveguides M, such that the error E is minimized. C The internal parameters are updated to generate the input data D. This eliminates the need for manual trial-and-error exploration to maximize the processing performance of the task regarding how to input optical signals to multiple input waveguides 61. In other words, according to the optical signal processing device 10 of this embodiment, the input data D IN The mismatch between the number of elements and the number of input waveguides 61 does not pose a problem.

[0042] Here, Figure 5 shows the input data D IN Based on arbitrary output data D OUT This figure schematically shows an example of the configuration of a signal processing device related to a comparative example that performs the task of obtaining the output data D. In recent years, it has become common to implement a DNN (Deep Neural Network) model 100 having an input layer 101, an intermediate layer 102, and an output layer 103 on a GPU (Graphics Processing Unit) and have it perform various calculations. The DNN model 100 is used to obtain the output data D. OUT and correct data D C The DNN model is trained to minimize the error E. During the training of the DNN model 100, the internal parameters are optimized for each of the input layer 101, hidden layer 102, and output layer 103. With a signal processing device equipped with the DNN model 100, the increasing power consumption of the GPU and the limitations of its processing speed become problems when there is a demand for further performance improvements. In addition, the high training cost, due to the enormous amount of data required to train the DNN model 100 and the long training time involved, is also a problem.

[0043] Figure 6 is a diagram showing the configuration of an optical signal processing device 10 according to an embodiment of the disclosed technology, in a format that allows comparison with the configuration of the DNN model 100 shown in Figure 5. In the optical signal processing device 10, the controller 50 and modulator 40 correspond to the input layer, the optical signal converter 60 corresponds to the intermediate layer, and the photoelectric converter 70 and computer 80 correspond to the output layer. In the optical signal processing device 10, the target of learning is only the controller 50. Therefore, with the optical signal processing device 10, the amount of internal parameters to be optimized by learning can be greatly reduced, and the learning cost can be greatly reduced.

[0044] Furthermore, according to the optical signal processing device 10, the power consumption required for the conversion process by the optical signal converter 60 is zero, and the conversion processing speed is "the speed of light". Therefore, the optical signal processing device 10 can achieve faster processing speed and lower power consumption compared to a signal processing device equipped with a DNN model 100 implemented on a GPU. On the other hand, the optical signal converter 60, which corresponds to the intermediate layer of the DNN 100, does not have "updatable parameters", so the output data D OUT Regarding accuracy, the signal processing device equipped with the DNN model 100 tends to have higher accuracy. Therefore, the optical signal processing device 10 is intended for use in systems where high accuracy is not required, but low power consumption and high-speed processing are necessary. To enhance the low power consumption and high speed characteristics of the optical signal processing device 10, it is preferable to keep the processing load on the computer 80 as small as possible.

[0045] Figure 7 shows an example of a learning method for computer 80 when computer 80 has updatable internal parameters and a learnable configuration. Computer 80 learns output data D based on multiple electrical signals V, for example. OUT It may include an inference model that outputs the above. The above inference model is built using machine learning.

[0046] The learning process of computer 80 involves, for example, updating the internal parameters of controller 50, and output data D OUT and correct data D CThis is done by updating the internal parameters of the computer 80 to minimize the error E between the two. That is, the common input data D IN Output data D OUT Correct data D C The controller 50 and the computer 80 may be trained using the error E. By configuring the computer 80 to be trainable, the output data D OUT This enables higher accuracy. By performing the training of the computer 80 in parallel with the training of the controller 50, it is expected that processing performance will be maximized. Alternatively, the training of the controller 50 and the training of the computer 80 may be performed independently.

[0047] Figures 8, 9, and 10 show variations in the configuration of the optical signal converter 60, respectively. As shown in Figure 8, the multiplexing section 62 of the optical signal converter 60 may contain a structure 65 inside that induces or promotes at least one of refraction, scattering, reflection, and diffraction of light propagating inside it. The structure 65 is randomly arranged inside the multiplexing section 62. By including the structure 65 in the multiplexing section 62, the manner of multiplexing in the multiplexing section 62 can be diversified and made more complex. The structure 65 may be made of a material that is reflective to light propagating in the multiplexing section 62, or it may be made of a material that is transparent to light. The structure 65 may be made of the same material as the multiplexing section 62, or it may be made of a different material.

[0048] The shape of the structure 65 is not particularly limited and may be, for example, circular, elliptical, rectangular, or triangular. The size of the structure 65 is not limited as long as a certain number of structures 65 can be arranged inside the corrugated section 62. Figure 8(a) illustrates a corrugated section 62 containing multiple structures 65 of the same size and shape. Figure 8(b) illustrates a corrugated section 62 containing multiple circular structures 65 of different sizes. Figure 8(c) illustrates a corrugated section 62 containing circular and elliptical structures 65. Figure 8(d) illustrates a corrugated section 62 containing structures 65 of random shapes and sizes.

[0049] The multiplexer 62 can take on various shapes. As shown in Figures 9(a), (c), (e), and (f), the input and output surfaces of the multiplexer 62 may be symmetrical curved surfaces. As shown in Figures 9(b), (d), and (h), the input and output surfaces of the multiplexer 62 may be planar. As shown in Figure 9(g), the input and output surfaces of the multiplexer 62 may be asymmetrical. As shown in Figures 9(d) and (h), the side surface of the multiplexer 62 may be asymmetrical.

[0050] As shown in Figures 10(a) and (b), the input and output surfaces of the multiplexer 62 may be convex curved surfaces, while the side surface may be a concave curved surface. As shown in Figure 10(a), the multiplexer 62 may have a symmetrical shape in the direction of light propagation. As shown in Figure 10(b), the multiplexer 62 may have an asymmetrical shape in the direction of light propagation.

[0051] [Second Embodiment] Figure 11 shows an example of the configuration of an optical signal processing device 10A according to a second embodiment of the disclosed technology. The optical signal processing device 10A includes a photoelectric converter 91 and a temperature sensor 92. The photoelectric converter 91 provides an optical intensity signal Q indicating the light intensity of one of the light segments divided by the demultiplexer 30. a It outputs and supplies this to the controller 50. Light intensity signal Q a The amount of light directly or indirectly indicated by this is approximately the same as the amount of light input to the modulator 40 or the optical signal converter 60.

[0052] The temperature sensor 92 provides a temperature signal T indicating the temperature of at least one of the internal and external parts of the system. a It outputs a signal and supplies it to the controller 50. The internal temperature of the system is the temperature of all or part of the elements constituting the optical signal processing device 10A. The external temperature of the system is the ambient temperature around the system.

[0053] The controller 50 receives input data D IN In addition, the light intensity signal Q a and temperature signal T a Based on the control signal S CGenerate it. The controller 50 is the light quantity signal Q a And the temperature signal T a To generate an appropriate control signal S C Is learned to be generated.

[0054] When at least one of the internal and external temperatures of the system and the amount of light input to the modulator 40 changes, an appropriate output data D OUT The control signal S for obtaining C May change. According to the optical signal processing apparatus 10A according to the present embodiment, the controller 50 is the input data D IN In addition to the light quantity signal Q a And the temperature signal T a Based on the control signal S C Is generated, so that the output data D OUT Can be realized with high precision.

[0055] In the above description, the case where the controller 50 generates the control signal S a Based on both the light quantity signal Q and the temperature signal T a Is exemplified, but the controller 50 is the light quantity signal Q C And the temperature signal T a Based on one of the control signal S a May be generated. Also, signals sensing other factors that can affect the control signal S C Other than temperature and light quantity may be input to the controller 50. C

[0056] Regarding the above embodiment, the following additional remarks are further disclosed. (Additional Remark 1) A light source that outputs light having coherence, A wavelength division multiplexer that divides the light output from the light source, A plurality of modulators that modulate the phase of a part or all of the plurality of divided lights according to a control signal and output them as a plurality of phase-modulated lights, A controller that generates the control signal based on input data, An optical signal converter that multiplexes the plurality of phase-modulated lights and outputs them as a plurality of converted lights, A photoelectric converter that converts each of the aforementioned plurality of converted light beams into an electrical signal, A computer that outputs output data corresponding to the input data based on the aforementioned electrical signal, Includes, The aforementioned optical signal converter is A plurality of input waveguides that form the input paths for the plurality of phase-modulated optical signals, A multiplexing unit that combines the aforementioned multiple phase-modulated light beams, A plurality of output waveguides that form the output paths of the plurality of converted light, Includes, The controller has updatable internal parameters, and these internal parameters are set to generate the control signal that minimizes the error between the output data and the ground truth data. Optical signal processing device.

[0057] (Note 2) The number of input waveguides differs from the number of elements in the input data. The optical signal processing circuit described in Appendix 1.

[0058] (Note 3) The input waveguide, the multiplexer, and the output waveguide are made of a material with a higher refractive index than the surrounding material. The optical signal processing circuit described in Appendix 1 or Appendix 2.

[0059] (Note 4) The wave multiplexer has a structure inside that induces or promotes at least one of refraction, scattering, reflection, and diffraction of light propagating within it. An optical signal processing device as described in any one of the appendices 1 to 3.

[0060] (Note 5) The computer derives the output data based on the relationship between the signal levels of the multiple electrical signals corresponding to each of the multiple converted light beams. An optical signal processing device as described in any one of the appendices 1 to 4.

[0061] (Note 6) The controller generates the control signal based on a temperature signal indicating at least one of the internal and external temperatures of the system. An optical signal processing device as described in any one of the appendices 1 to 5.

[0062] (Note 7) The controller generates the control signal based on a light intensity signal indicating the amount of light input to the modulator or the optical signal converter. An optical signal processing device as described in any one of the appendices 1 through 6.

[0063] (Note 8) The computer has updatable internal parameters, and in the process of updating the internal parameters of the controller, the computer's internal parameters are set so as to minimize the error between the output data and the correct data. An optical signal processing device as described in any one of the appendices 1 through 7. [Explanation of Symbols]

[0064] 10, 10A Optical signal processing device 20 light source 30 duplexer 40 Modulators 50 Controllers 60 Optical Signal Converter 61 Input waveguide 62 Wave-receiving section 63 Output waveguide 65 Structures 70 Photoelectric Converter 80 calculator

Claims

1. A light source that emits coherent light, A demultiplexer that splits the light output from the aforementioned light source, Multiple modulators that modulate the phase of some or all of the divided light in accordance with a control signal and output it as multiple phase-modulated lights, A controller that generates the control signal based on input data, An optical signal converter that combines the aforementioned multiple phase-modulated light signals and outputs them as multiple converted light signals, A photoelectric converter that converts each of the aforementioned plurality of converted light beams into an electrical signal, A computer that outputs output data corresponding to the input data based on the aforementioned electrical signal, Includes, The aforementioned optical signal converter is A plurality of input waveguides that form the input paths for the plurality of phase-modulated optical signals, A multiplexing unit that combines the aforementioned multiple phase-modulated light beams, A plurality of output waveguides that form the output paths of the plurality of converted light, Includes, The controller has updatable internal parameters, and these internal parameters are set to generate the control signal that minimizes the error between the output data and the ground truth data. Optical signal processing device.

2. The number of input waveguides differs from the number of elements in the input data. The optical signal processing apparatus according to claim 1.

3. The input waveguide, the multiplexer, and the output waveguide are made of a material with a higher refractive index than the surrounding material. The optical signal processing apparatus according to claim 1.

4. The wave multiplexer has a structure inside that induces or promotes at least one of refraction, scattering, reflection, and diffraction of light propagating within it. The optical signal processing apparatus according to claim 1.

5. The computer derives the output data based on the relationship between the signal levels of the multiple electrical signals corresponding to each of the multiple converted light beams. The optical signal processing apparatus according to claim 1.

6. The controller generates the control signal based on a temperature signal indicating at least one of the internal and external temperatures of the system. The optical signal processing apparatus according to claim 1.

7. The controller generates the control signal based on a light intensity signal indicating the amount of light input to the modulator or the optical signal converter. The optical signal processing apparatus according to claim 1.

8. The computer has updatable internal parameters, and in the process of updating the internal parameters of the controller, the computer's internal parameters are set so as to minimize the error between the output data and the correct data. The optical signal processing apparatus according to claim 1.

Images