Optical communication system, transmission device, reception device, learning device, optical communication method, and learning method

Abstract

One aspect of the present invention provides an optical communication system including a transmitter that transmits a signal converted by a filter, a photoelectric conversion element that receives the signal, and a learning unit that learns a mathematical model that performs inferring regarding to what sort of value a filter coefficient of the filter is to be updated, on the basis of a signal input to an own model and the filter coefficient of the filter. The signal input to the mathematical model is a reception signal obtained by the photoelectric conversion element. The optical communication system further includes a control unit that executes the mathematical model that has been trained, obtained by a learning device.

Description

Optical communication system, transmitting device, receiving device, learning device, optical communication method, and learning method

[0001] The present invention relates to an optical communication system, a transmitting device, a receiving device, a learning device, an optical communication method, and a learning method.

[0002] Demand for optical communications is increasing. However, the bandwidth of each device constituting a transmission path, such as a digital-to-analog converter (DAC), modulator, receiver, and analog-to-digital converter (ADC), may be insufficient for the transmitted signal. In such cases, waveform distortion occurs in the received signal due to bandwidth limitation. Waveform distortion is also caused by chromatic dispersion. As a result, the bit error rate may increase.

[0003] Therefore, a technology has been proposed that compensates for signal waveform distortion on the receiving side. Specifically, after an optical signal is received by a receiver, the electrical signal output from the receiver is input into a specified filter to compensate for waveform distortion. The filter is realized by digital signal processing. In direct detection systems, the receiver outputs an electrical signal with an intensity corresponding to the received signal strength. Therefore, when the receiver converts the signal from optical to electrical, the phase information of the signal is lost. As a result, phase information cannot be used when compensating for waveform distortion, and the effectiveness of waveform distortion compensation has been limited.

[0004] Therefore, a technique called pre-distortion has been proposed in which the waveform of a signal transmitted from the transmitting side is distorted in the first place, and the distortion is compensated for by propagating the signal through a transmission path.In other words, a technique called pre-distortion has been proposed in which the signal transmitted from the transmitting side is converted, and then the inverse conversion of the converted signal is performed on the signal by transmitting it through a transmission path (see Non-Patent Document 1).

[0005] In this technology, the transmitting side uses a filter (hereinafter referred to as the "transmit filter") to convert the signal input to the transmit filter. In order to remove distortion from the waveform obtained on the receiving side, the transfer function of the filter (hereinafter referred to as the "transmit filter function") must have the inverse characteristics of the transmission path. In other words, to effectively remove distortion from the received waveform obtained on the receiving side, it is important to calculate a filter with a transfer function close to the inverse characteristics of the transmission path and use it as the transmit filter.

[0006] The following technology (hereinafter referred to as the "reference technology") can be considered as a technique for calculating a transmit filter with a transfer function that has the inverse characteristics of the transmission line. A test signal is input to the transmit filter, and an optical signal generated based on this is transmitted toward the transmission line. The error between the received waveform obtained on the receiving side and an ideal received waveform with a large eye opening is calculated. The transmit filter coefficients are slightly updated to reduce the error. The error between the received waveform and the ideal waveform is then calculated, and the transmit filter coefficients are also slightly updated. This process is repeated until the error is below a predetermined standard. In this way, a transmit filter whose transfer function is closer to the inverse characteristics of the transmission line can be obtained. Using this, a more ideal received waveform can be obtained. An example of an algorithm that can reduce the error is the least mean square (LMS) algorithm.

[0007] RI Killey, PM Watts, M. Glick, and P. Bayvel, "Electronic dispersion compensation by signal predistortion", in Proc. Conf. Opt. Fiber Commun., 2006, pp. 1-3.

[0008] However, this technique requires the receiver to repeatedly update the transmit filter coefficients based on the received waveform, and then feed them back to the transmitter. Because this feedback from the receiver to the transmitter takes time, there is a concern that this repeated process may increase the time required to make the transmit filter's transfer function approach the inverse characteristics of the transmission path.

[0009] In view of the above circumstances, an object of the present invention is to provide a technique for reducing the time required to make the transfer function of a transmission filter approach the inverse characteristics of a transmission path.

[0010] One aspect of the present invention is an optical communication system comprising: a transmitter that transmits a signal converted by a filter; a photoelectric conversion element that receives the signal; a learning unit that learns a mathematical model that infers the value to which the filter coefficients of the filter should be updated based on a signal input to the model and the filter coefficients of the filter, wherein the signal input to the mathematical model is the received signal obtained by the photoelectric conversion element; and a control unit that executes the learned mathematical model obtained by a learning device.

[0011] One aspect of the present invention is an optical communication system comprising: a transmitter that transmits a signal converted by a filter; a photoelectric conversion element that receives the signal; a learning unit that learns a mathematical model that infers the value to which the filter coefficients of the filter should be updated based on a signal input to the model and the filter coefficients of the filter, wherein the signal input to the mathematical model is the received signal obtained by the photoelectric conversion element; and a control unit that executes the learned mathematical model obtained by a learning device.

[0012] One aspect of the present invention is a transmitting device comprising: a transmitting filter that converts an input signal; and a transmitter that transmits the signal converted by the transmitting filter; wherein the filter coefficients of the filter are updated to filter coefficients inferred by a learning unit that learns a mathematical model that infers values ​​to which the filter coefficients of the filter should be updated based on a signal input to the model and the filter coefficients of the filter; the signal input to the mathematical model is a received signal obtained by a photoelectric conversion element that receives the signal transmitted by the transmitter; and the signal is updated to filter coefficients inferred by the learned mathematical model obtained by a learning device.

[0013] One aspect of the present invention is a receiving device comprising: a photoelectric conversion element that receives a signal transmitted by a transmitter that transmits a signal converted by a transmission filter that is a filter that converts an input signal; a learning unit that learns a mathematical model that infers the value to which the filter coefficient of the filter should be updated based on the signal input to the model and the filter coefficient of the filter; and a control unit that executes the learned mathematical model obtained by a learning device, wherein the signal input to the mathematical model is a received signal obtained by the photoelectric conversion element.

[0014] One aspect of the present invention is a learning device that includes a learning unit that learns a mathematical model that infers what value the filter coefficient of a filter should be updated to based on a received signal obtained by a photoelectric conversion element that receives a signal transmitted by a transmitter that transmits a signal converted by a transmission filter, which is a filter that converts an input signal, and the filter coefficient of the filter.

[0015] One aspect of the present invention is an optical communication method executed by an optical communication system comprising: a transmitter that transmits a signal converted by a filter; an opto-electrical conversion element that receives the signal; and a learning unit that learns a mathematical model that infers what value the filter coefficients of the filter should be updated to based on a signal input to the model and the filter coefficients of the filter, wherein the signal input to the mathematical model is a received signal obtained by the opto-electrical conversion element; and a control unit that executes the learned mathematical model obtained by a learning device, the optical communication method having a receiving step in which the opto-electrical conversion element receives the signal; and an execution step in which the control unit executes the learned mathematical model.

[0016] One aspect of the present invention is a learning method executed by a learning device that includes a learning unit that learns a mathematical model that infers what value the filter coefficient of the filter should be updated to based on a received signal obtained by a photoelectric conversion element that receives a signal transmitted by a transmitter that transmits a signal converted by a transmission filter, which is a filter that converts an input signal, and the filter coefficient of the filter, and the learning method includes a learning step in which the learning unit performs the learning.

[0017] According to the present invention, it is possible to reduce the time required to make the transfer function of the transmission filter approach the inverse characteristics of the transmission path.

[0018] 1 is an explanatory diagram illustrating an optical communication system according to an embodiment. A flowchart showing an example of a processing flow executed by the optical communication system according to an embodiment. A diagram showing an example of the hardware configuration of a transmitting device according to an embodiment. A diagram showing an example of the hardware configuration of a receiving device according to an embodiment. An explanatory diagram explaining a learning device according to an embodiment. A first explanatory diagram explaining an example of a learning dataset according to an embodiment. A second explanatory diagram explaining an example of a learning dataset according to an embodiment. An explanatory diagram explaining an optical communication system according to a modified example. A flowchart showing an example of a processing flow executed by the optical communication system according to a modified example. An explanatory diagram explaining averaging processing according to a modified example. A first explanatory diagram explaining phase shift compensation processing according to a modified example. A second explanatory diagram explaining phase shift compensation processing according to a modified example. A third explanatory diagram explaining phase shift compensation processing according to a modified example. A first explanatory diagram explaining a filter coefficient inference model according to a modified example. A second explanatory diagram explaining a filter coefficient inference model according to a modified example.

[0019] (Embodiment) <Optical Communication System> Fig. 1 is an explanatory diagram illustrating an optical communication system 100 according to an embodiment. The optical communication system 100 includes a transmitting device 1, a receiving device 2, and a transmission medium 3. The transmission medium 3 is a medium through which an optical signal propagates. The transmission medium 3 is, for example, an optical fiber. The transmission medium 3 may also be, for example, air.

[0020] <<Transmitting Device 1 >> The transmitting device 1 includes a transmitter 10 , a transmission control unit 11 , and a DAC 120 .

[0021] The transmitter 10 outputs an optical signal. The optical signal output by the transmitter 10 (hereinafter referred to as the "transmission signal") propagates through the transmission medium 3 and reaches the receiving device 2. The transmitter 10 may have any configuration as long as it can output an optical signal. In the example of FIG. 1, the transmitter 10 includes a light source 101 and a modulator 102. The light source 101 is a laser, and the modulator 102 is an external modulator that modulates the signal output by the light source 101 with a modulation signal. Note that, although the transmitter 10 includes an external modulator in the example of FIG. 1, if the light source 101 is a directly modulated semiconductor laser, the transmitter 10 does not need to include the modulator 102. In such a case, the modulation signal input to the modulator 102 is input to the light source 101. The transmitter 10 is under the control of a transmission control unit 11.

[0022] The transmission control unit 11 is a control unit that includes a processor such as a central processing unit (CPU), a graphics processing unit (GPU), or a neural network processing unit (NPU), and a memory. The transmission control unit 11 executes a filter process, a transmission process, and a filter update process.

[0023] <<<<Filtering>>> Filtering is a process that uses a transmission filter, which is a filter that converts an input signal, and performs filtering by applying a predetermined transfer function to the signal input to the transmission filter. Therefore, a signal converted by filtering is a signal converted by the transmission filter.

[0024] The transfer function is determined by the filter coefficients of the transmit filter (hereinafter referred to as "transmit filter coefficients"). Filter coefficients are coefficients that determine the characteristics of the filter, and examples include the tap coefficients of a finite impulse response (FIR) filter. Other possible coefficients include those set for each node when a neural network is used. The input for filter processing in signal processing is an electrical signal.

[0025] The receiving device 2 is equipped with a photoelectric conversion element for receiving optical signals, and in order to obtain an ideal received waveform without waveform distortion at the photoelectric conversion element, it is necessary to set appropriate filter coefficients that give the transmitting filter the inverse characteristics of the transmission path. However, in the initial state in which the transmitting device 1 and the receiving device 2 are connected via the transmission medium 3, the transmitting device 1 does not know the state of the transmission path including the transmission medium 3, and therefore cannot calculate and set appropriate transmitting filter coefficients that have the inverse characteristics.

[0026] Therefore, in the optical communication system 100, initial filter coefficients (hereinafter referred to as "reference filter coefficients") are first set for the transmit filter. Specifically, the reference filter coefficients are predetermined filter coefficients. In fact, in the optical communication system 100, a learned filter coefficient inference model is executed, as will be described later. To explain the reference filter coefficients using the learned filter coefficient inference model, the reference filter coefficients are transmit filter coefficients used in inference using the learned filter coefficient inference model.

[0027] The filter coefficient inference model is a mathematical model that infers the values ​​to which the transmit filter coefficients should be updated based on the signal input to the model and the transmit filter coefficients. Therefore, the filter coefficient inference model can be said to be a process that infers the updated transmit filter coefficients. Hereinafter, the filter coefficients inferred by the filter coefficient inference model (hereinafter referred to as "inferred filter coefficients"). Note that learning refers to machine learning. As can be seen from the explanation so far, it can be said that the transmit filter coefficients can be updated using a trained filter coefficient inference model.

[0028] The technical significance of updating the filter coefficients will now be explained. When a transmit filter set with reference filter coefficients is used, the transmit filter does not have the inverse characteristics of the transmission path. Therefore, the received waveform contains waveform distortion due to the transfer function of the transmission path. In other words, by analyzing the distortion of the received waveform, it is possible to calculate the waveform distortion of the transmission path, and ultimately, ideal transmit filter coefficients that have the inverse characteristics of the distortion. Therefore, in the optical communication system 100, optimal transmit filter coefficients are estimated using a trained filter coefficient inference model based on the received signal containing waveform distortion.

[0029] The reference filter coefficients may be, for example, coefficients of a filter in which the input and output of a transmission filter match, that is, the transfer function is 1. Hereinafter, information indicating the reference filter coefficients will be referred to as reference filter coefficient information.

[0030] The ideal waveform is defined as the waveform of an optical signal received by the photoelectric conversion element 20, in which the transmission quality is equal to or greater than a predetermined value and the bit error is less than a predetermined value.

[0031] <<<Transmission Processing>>> The transmission processing is processing that controls the operation of the transmitter 10 and causes the transmitter 10 to output an optical signal representing the signal after conversion by the transmission filter. The output of the optical signal by the transmitter 10 is the transmission of the optical signal. In this way, the transmission processing is processing that transmits an optical signal by controlling the operation of the transmitter 10 under the control of the transmission control unit 11, and is processing that transmits an optical signal representing the signal after conversion by the transmission filter. The optical signal representing the signal after conversion by the transmission filter is the transmission signal.

[0032] <<<<Filter Update Processing>>>> As described above, in the optical communication system 100, the learned filter coefficient inference model is executed to obtain inferred filter coefficients. The filter update processing is processing for updating the transmission filter coefficients to the filter coefficients indicated by the inferred filter coefficient information (i.e., the inferred filter coefficients) based on inferred filter coefficient information, which is information indicating the inferred filter coefficients.

[0033] In the example of FIG. 1 , the transmission control unit 11 includes a control signal transmitting / receiving unit 111, a transmission signal generating unit 112, and a transmission filter unit 113. The control signal transmitting / receiving unit 111 performs processing to receive inferred filter coefficient information. The transmission signal generating unit 112 acquires a transmission code sequence and obtains an electrical signal to be input to the transmission filter. The transmission filter unit 113 acquires the inferred filter coefficient information obtained by the control signal transmitting / receiving unit 111 and performs filter update processing. In addition, the transmission filter unit 113 performs filter processing on the electrical signal obtained by the transmission signal generating unit 112.

[0034] In the example of FIG. 1 , the electrical signal obtained by the filtering process is input to the DAC 120. The DAC 120 is a digital-to-analog converter (DAC). In the example of FIG. 1 , the output of the DAC 120 is used as a modulating signal. Therefore, the modulating signal obtained by the DAC 120 is input to the transmitter 10. The transmitter 10 outputs a transmission signal obtained from the input modulating signal to the transmission medium 3. In the example of FIG. 1 , the transmission signal transmitted by the transmitter 10 propagates through the transmission medium 3 and reaches the receiving device 2 equipped with a photoelectric conversion element 20.

[0035] <Receiving Device 2> The receiving device 2 includes a photoelectric conversion element 20, a reception control unit 21, and an ADC 220.

[0036] The photoelectric conversion element 20 receives a transmission signal and obtains an electrical signal. The photoelectric conversion element 20 may be any device capable of receiving an optical signal and outputting an electrical signal, such as a photodiode. Hereinafter, the electrical signal output by the photoelectric conversion element 20 will be referred to as a received signal. More specifically, the received signal is an electrical signal obtained by converting an optical signal by the photoelectric conversion element 20.

[0037] The reception control unit 21 is a control unit that includes a processor such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or an NPU (Neural Network Processing Unit), and a memory. The reception control unit 21 controls the operation of each functional unit included in the receiving device 2.

[0038] As described above, a learned filter coefficient inference model is executed in the optical communication system 100. The learned filter coefficient inference model may be executed by any device included in the optical communication system 100, but Fig. 1 shows, as an example, the optical communication system 100 in which the reception control unit 21 executes the learned filter coefficient inference model. Here, an example of processing executed in the optical communication system 100 when the reception control unit 21 executes the learned filter coefficient inference model will be described.

[0039] The reception control unit 21 executes an inference process. The inference process is a process of inputting the signal output by the photoelectric conversion element 20 and reference filter coefficient information into a learned filter coefficient inference model and executing the learned filter coefficient inference model. Therefore, the inference process is a process of using the learned filter coefficient inference model to infer the value to which the filter coefficient of the transmission filter should be updated, based on the signal output by the photoelectric conversion element 20 and the reference filter coefficient information. Because the inference process infers the value to which the filter coefficient should be updated, it can be said that the inference process is a process of inferring the inferred filter coefficient.

[0040] The inferred filter coefficient information is input to the transmitting device 1 via a predetermined control signal channel or manually. Manual input refers to, for example, connecting a predetermined storage device in which the coefficient information is recorded to the transmitting device 1 and inputting the coefficient information to the transmitting device 1. The predetermined control signal channel may be, for example, radio wave communication between the receiving device 2 and the transmitting device 1. The predetermined control signal channel may also be, for example, communication via an electrical signal between the receiving device 2 and the transmitting device 1. The predetermined control signal channel may also be, for example, communication via a low-speed optical line between the receiving device 2 and the transmitting device 1. In a low-speed optical line, waveform distortion due to propagation is inherently small, making long-distance communication easy, even without using the transmission filter. In the example of FIG. 1 , the inferred filter coefficient information is input from the receiving device 2 to the transmitting device 1 via a predetermined control signal channel.

[0041] The process of inputting the coefficient information to the transmitting device 1 via predetermined communication is performed by, for example, the reception control unit 21. At this time, the reception control unit 21 controls the operation of a communication interface provided in, for example, the interface unit 22 (described later) and transmits the coefficient information to the communication interface. As a result, the coefficient information is transmitted from the receiving device 2 and input to the transmitting device 1.

[0042] After the inferred filter coefficient information is input to the transmitting device 1, the transmission control unit 11 acquires the inferred filter coefficient information and updates the transmission filter coefficients to the filter coefficients indicated by the inferred filter coefficient information.

[0043] 1, the reception control unit 21 includes a DSP unit 210, a filter coefficient inference model unit 211, and a control signal transmission / reception unit 212. The DSP unit 210 executes DSP (Digital Signal Processing) and outputs a received code sequence. Therefore, the DSP unit 210 can be said to perform processing to estimate a received code sequence based on a received signal.

[0044] The filter coefficient inference model unit 211 executes the inference process.

[0045] In the example of FIG. 1 , the output of the photoelectric conversion element 20 is input to the ADC 220. The output of the ADC 220 is then input to the reception control unit 21, and then to the filter coefficient inference model unit 211 and the DSP unit 210. The ADC 220 is an ADC (Analog-to-Digital Converter). Therefore, the received signal used in the filter coefficient inference model unit 211 is the signal output by the ADC 220. The received signal used in the DSP unit 210 is also the signal output by the ADC 220. However, because the ADC 220 simply converts an analog signal to a digital signal, in the example of FIG. 1 , the inference process and the estimation of the received code sequence are performed based on the received signal.

[0046] The control signal transmitting / receiving unit 212 executes a process of transmitting the inferred filter coefficient information.

[0047] 2 is a flowchart showing an example of a flow of processing executed by the optical communication system 100 according to the embodiment. The transmission control unit 11 acquires a predetermined signal (hereinafter referred to as a "test signal") (step S101). The acquired test signal is input to a transmission filter (hereinafter referred to as a "reference filter") whose filter coefficients are reference filter coefficients. The test signal is, for example, an electrical signal.

[0048] Next, the transmission control unit 11 converts the test signal using the reference filter (step S102). That is, the transmission control unit 11 performs filtering on the test signal using the reference filter.

[0049] Next, the transmission control unit 11 controls the operation of the transmitter 10 to cause the transmitter 10 to transmit an optical signal representing the converted test signal (hereinafter referred to as the "test optical signal") (step S103). That is, the transmission control unit 11 executes a transmission process to transmit the test optical signal.

[0050] Next, the photoelectric conversion element 20 receives the test optical signal transmitted in step S103 (step S104). Having received the test optical signal, the photoelectric conversion element 20 obtains a reception signal indicating the waveform of the received test optical signal.

[0051] Next, the reception control unit 21 uses the learned filter coefficient inference model to infer, based on the reference filter coefficient information and the received signal output by the photoelectric conversion element 20 in step S104, what values ​​the filter coefficients of the transmission filter should be updated to (step S105). That is, the reception control unit 21 executes the inference process. Note that the reference filter coefficient information may be, for example, pre-recorded in a storage unit provided in the receiving device 2, or may be obtained, for example, by communication with the transmitting device 1 via a predetermined control signal channel.

[0052] Next, information indicating the result of the inference (i.e., inferred filter coefficient information) is acquired by the transmission control unit 11 (step S106). Here, for example, the reception control unit 21 may transmit the inferred filter coefficient information to the transmission device 1 using predetermined communication, thereby causing the transmission control unit 11 to acquire the inferred filter coefficient information. Alternatively, the inferred filter coefficient information may be manually input from the reception device 2 to the transmission device 1, causing the transmission control unit 11 to acquire the inferred filter coefficient information.

[0053] Next, the transmission control unit 11 updates the filter coefficients of the transmission filter to the filter coefficients indicated by the acquired inferred filter coefficient information (i.e., the inferred filter coefficients) (step S107).

[0054] Next, the transmission control unit 11 acquires a signal that is not a test signal (hereinafter referred to as a "non-test signal") (step S108). The acquired non-test signal is input to the transmission filter. The non-test signal is, for example, a client signal that indicates information that one of two parties communicating using the optical communication system 100 wants to convey to the other. The non-test signal is an electrical signal.

[0055] Next, the transmission control unit 11 converts the non-test signal using a transmission filter (hereinafter referred to as the "inferred filter") whose filter coefficients are the inferred filter coefficients (step S109). That is, the transmission control unit 11 performs filtering on the non-test signal using the inferred filter.

[0056] Next, the transmission control unit 11 controls the operation of the transmitter 10 to cause the transmitter 10 to transmit an optical signal representing the converted non-test signal (hereinafter referred to as a "non-test optical signal") (step S110). That is, the transmission control unit 11 executes a transmission process to transmit the non-test optical signal.

[0057] Next, the photoelectric conversion element 20 receives the non-test optical signal transmitted in step S110 (step S111). The photoelectric conversion element 20 receives the non-test optical signal and obtains a received signal indicating the waveform of the received non-test optical signal. Since the transmission filter uses ideal filter coefficients having the inverse characteristics of the transmission path, when the non-test signal is received by the photoelectric conversion element 20, an ideal waveform without waveform distortion can be obtained, enabling communication without bit errors.

[0058] <<Effects of Executing Inference Processing>> The effects of executing the inference processing will be described. In the inference processing, filter coefficients are inferred using a trained filter coefficient inference model. "Trained" generally means that the mathematical model to be trained has been optimized.

[0059] The transfer function of a transmit filter is determined by the filter coefficients of the transmit filter, so inferring the filter coefficients of the transmit filter is equivalent to inferring the transfer function.

[0060] As described above, the filter coefficient inference model is a mathematical model that infers the filter coefficients of a transmit filter. Therefore, if a mathematical model to be learned, which is an optimized filter coefficient inference model, is used, a transfer function that is substantially identical to the transfer function obtained by the reference technology can be inferred by simply executing the mathematical model once. Therefore, when a trained filter coefficient inference model is used, it is not necessary to repeatedly feed back filter coefficients estimated on the receiving side to the transmitting side, as in the reference technology. Therefore, by executing the inference process, it is possible to reduce the time required to bring the transfer function of the transmit filter closer to the inverse characteristics of the transmission path, compared to the reference technology.

[0061] 3 is a diagram showing an example of the hardware configuration of the transmitting device 1 in an embodiment. The transmitting device 1 is equipped with a transmission control unit 11, which is a control unit including a processor 91 such as a CPU, GPU, or NPU, and a memory 92, which are connected by a bus, and executes a program. By executing the program, the transmitting device 1 functions as a device including a transmitter 10, a DAC 120, the transmission control unit 11, an interface unit 12, and a storage unit 13.

[0062] More specifically, the processor 91 reads out a program stored in the storage unit 13 and stores the read out program in the memory 92. When the processor 91 executes the program stored in the memory 92, the transmitting device 1 functions as a device including the transmitter 10, the DAC 120, the transmission control unit 11, the interface unit 12, and the storage unit 13.

[0063] The transmission control unit 11 controls the operation of each functional unit included in the transmitting device 1. The transmission control unit 11 performs, for example, filter processing. The transmission control unit 11 performs, for example, transmission processing. The transmission control unit 11 performs, for example, filter update processing. The transmission control unit 11 acquires, for example, information stored in the memory unit 13. Specifically, the process of acquiring information stored in the memory unit 13 is reading. It should be noted that the control signal transmitting / receiving unit 212 can be said to be a process that controls the operation of the interface unit 12.

[0064] The interface unit 12 includes a communication interface for connecting the transmission device 1 to an external device. The interface unit 12 communicates with the external device via a wired or wireless connection.

[0065] The external device is, for example, the receiving device 2. The interface unit 12 acquires, for example, inferred filter coefficient information through communication with the receiving device 2. The interface unit 12 transmits, for example, reference filter coefficient information to the receiving device 2 through communication with the receiving device 2.

[0066] The external device is, for example, a device that is a source of the non-test signal. The interface unit 12 acquires the non-test signal by communicating with the device that is the source of the non-test signal.

[0067] The interface unit 12 may be configured to include input devices such as a mouse, a keyboard, a touch panel, etc. The interface unit 12 may be configured as an interface that connects these input devices to the transmission device 1. In this way, the input devices of the interface unit 12 accept input of various information to the transmission device 1 via wired or wireless connections.

[0068] Note that the information or signal does not necessarily have to be input to the communication interface of the interface unit 12, but may be input to an input device of the interface unit 12. Thus, for example, inferred filter coefficient information may be input to the input device of the interface unit 12. Also, for example, a non-test signal may be input to the input device of the interface unit 12.

[0069] The interface unit 12 outputs, for example, various types of information. The interface unit 12 includes, for example, a display device such as a CRT (Cathode Ray Tube) display, a liquid crystal display, or an organic EL (Electro-Luminescence) display, and a speaker. The interface unit 12 may be configured as an interface that connects these display devices or speakers to the transmitting device 1. Therefore, the interface unit 12 may output, for example, information input to a communication interface or an input device of the interface unit 12 as an image or sound.

[0070] The storage unit 13 is configured using a computer-readable storage medium device (non-transitory computer-readable recording medium) such as a magnetic hard disk device or a semiconductor storage device. The storage unit 13 stores various information related to the transmission device 1. The storage unit 13 stores, for example, various information generated by the operation of the transmission control unit 11. The storage unit 13 may exist on a cloud, for example. The storage unit 13 may store, for example, reference filter coefficients in advance. The storage unit 13 may also store, for example, information indicating a test signal (for example, a transmission code sequence indicating a test signal) in advance.

[0071] 4 is a diagram showing an example of the hardware configuration of the receiving device 2 in an embodiment. The receiving device 2 is equipped with a reception control unit 21, which is a control unit including a processor 93 such as a CPU, GPU, or NPU, and a memory 94, which are connected via a bus, and executes a program. By executing the program, the receiving device 2 functions as a device including a photoelectric conversion element 20, an ADC 220, the reception control unit 21, an interface unit 22, and a storage unit 23.

[0072] More specifically, the processor 93 reads out a program stored in the storage unit 23 and stores the read program in the memory 94. When the processor 93 executes the program stored in the memory 94, the receiving device 2 functions as a device including the photoelectric conversion element 20, the ADC 220, the reception control unit 21, the interface unit 22, and the storage unit 23.

[0073] The reception control unit 21 controls the operation of each functional unit included in the receiving device 2. The reception control unit 21 executes, for example, an inference process. The reception control unit 21 acquires, for example, information stored in the memory unit 23. Specifically, the process of acquiring the information stored in the memory unit 23 is reading.

[0074] The interface unit 22 includes a communication interface for connecting the receiving device 2 to an external device. The interface unit 22 communicates with the external device via a wired or wireless connection.

[0075] The external device is, for example, the transmitting device 1. The interface unit 22 communicates with the transmitting device 1 to transmit, for example, inferred filter coefficient information to the transmitting device 1. The interface unit 22 may also acquire, for example, reference filter coefficient information from the transmitting device 1 through communication with the transmitting device 1.

[0076] The external device may be, for example, the learning device 4 described later. In this case, the reception control unit 21 can execute the learned filter coefficient inference model obtained by the learning device 4 via the interface unit 22.

[0077] The interface unit 22 may be configured to include input devices such as a mouse, keyboard, or touch panel. The interface unit 22 may be configured as an interface that connects these input devices to the receiving device 2. In this way, the input devices of the interface unit 22 accept input of various information to the receiving device 2 via wired or wireless connections. Note that information or signals do not necessarily have to be input to the communication interface of the interface unit 22, but may also be input to the input devices of the interface unit 22.

[0078] The interface unit 22 outputs, for example, various types of information. The interface unit 22 includes, for example, a display device such as a CRT display, a liquid crystal display, or an organic EL display, and a speaker. The interface unit 22 may be configured as an interface that connects these display devices or speakers to the receiving device 2. Therefore, the interface unit 22 may output, for example, information input to a communication interface or an input device of the interface unit 22 as an image or sound.

[0079] The storage unit 23 is configured using a computer-readable storage medium device such as a magnetic hard disk device or a semiconductor storage device. The storage unit 23 stores various information related to the receiving device 2. The storage unit 23 stores, for example, various information generated by the operation of the reception control unit 21. The storage unit 23 may exist on a cloud, for example.

[0080] It should be noted that the receiving device 2 does not necessarily need to execute the trained filter coefficient inference model via the communication interface provided in the interface unit 22. For example, the receiving device 2 may be able to execute the trained filter coefficient inference model by copying the trained filter coefficient inference model from the learning device 4 to a predetermined storage device, and then copying the trained filter coefficient inference model from the storage device to the storage unit 23 of the receiving device 2.

[0081] As can be seen from the above description, the optical communication system 100 is an optical communication system that performs optical communication using a transmitter and a photoelectric conversion element.

[0082] <Learning Device 4> As described above, a trained filter coefficient inference model is used in the optical communication system 100. Therefore, a device that performs training of the filter coefficient inference model will be described.

[0083] 5 is an explanatory diagram illustrating a learning device 4 according to an embodiment. The learning device 4 includes a learning unit 41, which is a control unit including a processor 95 such as a CPU, GPU, or NPU, and a memory 96, all connected via a bus, and executes a program. By executing the program, the learning device 4 functions as a device including the learning unit 41, an interface unit 42, and a storage unit 43.

[0084] More specifically, the processor 95 reads the program stored in the storage unit 43 and stores the read program in the memory 96. When the processor 95 executes the program stored in the memory 96, the learning device 4 functions as a device including the learning unit 41, the interface unit 42, and the storage unit 43.

[0085] The learning unit 41 executes a learning process. The learning process is a process for learning a filter coefficient inference model. The learning may be supervised learning, unsupervised learning, self-supervised learning, or semi-supervised learning.

[0086] Learning continues until a predetermined condition for terminating learning (hereinafter referred to as a "learning termination condition") is satisfied. The learning termination condition may be any condition for terminating learning. For example, the learning termination condition may be a condition that the filter coefficient inference model has been updated a predetermined number of times, or a condition that the change in the filter coefficient inference model due to the update is smaller than a predetermined change.

[0087] The learning unit 41 controls, for example, the operation of each functional unit included in the learning device 4. The learning unit 41 acquires, for example, information stored in the memory unit 43. Specifically, the process of acquiring the information stored in the memory unit 43 is reading.

[0088] The interface unit 42 includes a communication interface for connecting the learning device 4 to an external device. The interface unit 42 communicates with the external device via a wired or wireless connection.

[0089] The external device is, for example, the receiving device 2. In this case, the receiving device 2 can execute the learned filter coefficient inference model obtained by the learning device 4 via the interface unit 42.

[0090] The external device may be, for example, a device that transmits a training dataset, which is a collection of data used for training (hereinafter referred to as "training data"). The number of elements in the training dataset may be one or more. The interface unit 42 acquires the training dataset by communicating with the device that transmitted the training dataset. The training dataset acquired by the interface unit 42 may be recorded in the storage unit 43.

[0091] The interface unit 42 may be configured to include input devices such as a mouse, keyboard, or touch panel. The interface unit 42 may be configured as an interface that connects these input devices to the learning device 4. In this way, the input devices of the interface unit 42 accept input of various information to the learning device 4 via wired or wireless connections.

[0092] The information or signals do not necessarily have to be input to the communication interface of the interface unit 42, but may be input to an input device of the interface unit 42. Therefore, for example, learning data may be input to an input device of the interface unit 42, or may be input to the communication interface of the interface unit 42.

[0093] The interface unit 42 outputs, for example, various types of information. The interface unit 12 includes a display device such as a CRT display, a liquid crystal display, or an organic EL display, and a speaker. The interface unit 42 may be configured as an interface that connects these display devices or speakers to the learning device 4. Therefore, the interface unit 42 may output, for example, information or signals input to the communication interface or input device of the interface unit 42 as images or sounds. The interface unit 42 may also output, for example, information recorded in the memory unit 43 to a predetermined output destination such as a predetermined storage device.

[0094] The storage unit 43 is configured using a computer-readable storage medium device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 43 stores various information related to the learning device 4. The storage unit 43 stores, for example, various information generated by the operation of the learning unit 41. The storage unit 43 stores, for example, a learned filter coefficient inference model. The storage unit 43 stores, for example, a learning dataset. The storage unit 43 may exist, for example, on a cloud.

[0095] <<Example of Training Data Set>> An example of a training data set will be described with reference to Figs. 6 and 7. In the examples of Figs. 6 and 7, the symbol B ＤＡＣ－ｎ means the band of the DAC included in the transmitter 1. In the examples of FIGS. ｎ means the length of the transmission medium 3 such as an optical fiber. ＰＤ－ｎ means the band of the photoelectric conversion element 20.

[0096] FIG. 6 is a first explanatory diagram illustrating an example of a training data set in an embodiment. Each training data set in the training data set may include, for example, a received signal and inferred filter coefficient information. Therefore, the training data may be a pair of a received signal and inferred filter coefficient information. Note that the reference filter coefficient information may be included in the training data, but does not necessarily have to be included in the training data. For example, if the reference filter coefficient information is stored in advance in the storage unit 43, the reference filter coefficient information does not necessarily have to be included in the training data. In the example of FIG. 6, the training data is a pair of a received signal and inferred filter coefficient information.

[0097] In this case, in the learning process, for example, learning is performed using the received signal as an explanatory variable and the inferred filter coefficient information as a target variable. That is, supervised learning is performed using the received signal labeled with the inferred filter coefficient information.

[0098] Each training data set in the training data set is data obtained by, for example, simulation or experiment. In the case of a simulation, for example, a mathematical model (hereinafter referred to as a "system model") representing the optical communication system 100 is used. In such a simulation, for example, the values ​​of one or more parameters defined in the transmission path in the system model are changed in various ways. The simulation is a propagation simulation consisting of a transmission path that simulates a transmitter, a transmission medium, and a receiver.

[0099] Then, a pair of a received signal and inferred filter coefficient information is obtained for each set of parameter values ​​through simulation. Each pair of a received signal and inferred filter coefficient information obtained for each set of parameter values ​​is an example of training data, and a collection of training data thus obtained is an example of a training data set. Therefore, each training data included in the training data set has a different set of corresponding parameter values.

[0100] 7 is a second explanatory diagram illustrating an example of a training data set in the embodiment. More specifically, FIG. 7 is an explanatory diagram illustrating an example of parameters defined for a transmission path in the embodiment. Note that components having the same functions as those in FIG. 1 are assigned the same reference numerals and will not be described again.

[0101] 7 shows that the parameters defined for the transmission path may include, for example, the length of the transmission medium 3 through which the optical signal propagates. Also, FIG. 7 shows that the parameters defined for the transmission path may include, for example, the bandwidth of the transmission medium 3.

[0102] Fig. 7 shows that when the transmitting device 1 includes a DAC, the parameters defined for the transmission path may include the bandwidth of the DAC. Fig. 7 shows that when the receiving device 2 includes an ADC, the parameters defined for the transmission path may include the bandwidth of the ADC. Fig. 7 shows that when the transmitting device 1 includes a modulator, the parameters defined for the transmission path may include the bandwidth of the modulator. Note that the modulator may be, for example, an IQ modulator. Fig. 7 shows that the parameters defined for the transmission path may include the bandwidth of the photoelectric conversion element 20 included in the receiving device 2.

[0103] Returning to the explanation of Fig. 6, the example of Fig. 6 shows that learning data was obtained by simulation for each set (hereinafter referred to as "condition set") of the bandwidth of the DAC provided in the transmitting device 1, the length of the transmission medium 3, and the bandwidth of the photoelectric conversion element 20. More specifically, ｎ (n is an integer between 1 and N) indicates that learning data has been obtained for each of N condition sets identified by the formula:

[0104] In the example of FIG. 6, for example, the identifier P ２ The set of conditions identified by is that the bandwidth of the DAC provided in the transmitting device 1 is 50, the length of the transmission medium 3 is 30, and the bandwidth of the photoelectric conversion element 20 is 50. The example in Fig. 6 shows that the strength of the received signal at each time point obtained by simulation under the conditions indicated by this set of conditions, when expressed in discrete time, is 0.4, 0.2, 1.1, 0.3, ... in order from the earliest time point.

[0105] In the example of Fig. 6, the filter coefficient of the transmission filter is not one but a plurality of filter coefficients, and the inferred filter coefficient information indicates each of the plurality of filter coefficients in a predetermined order. Therefore, the inferred filter coefficient information indicates a sequence of filter coefficients. In the example of Fig. 6, for example, an identifier P ２ The inferred filter coefficient information obtained by simulation under the conditions indicated by the condition set identified by indicates that the values ​​of the filter coefficients in the sequence of filter coefficients are 1.2, 0.4, 0.2, 0.5, etc. In the simulation, the inferred filter coefficient information may be obtained by, for example, a reference technique.

[0106] The smaller the difference between the configuration of the transmission path used in the simulation and the configuration of the transmission path in the optical communication system 100 that is actually used instead of the simulation, the higher the accuracy of the estimation by the learned filter coefficient inference model. The configuration of the transmission path used in the simulation may be the same as the configuration of the transmission path in the optical communication system 100 that is actually used instead of the simulation.

[0107] <<Effects of Executing the Learning Process>> As explained in <<Effects of Executing the Inference Process>>, by using a trained filter coefficient inference model, the time required to make the transfer function of the transmission filter approach the inverse characteristics of the transmission path is reduced. Therefore, executing the learning process to obtain such a trained filter coefficient inference model can reduce the time required to make the transfer function of the transmission filter approach the inverse characteristics of the transmission path.

[0108] The learning device 4 configured in this manner executes the learning process, which, as described in <<Effects of Executing the Learning Process>>, makes it possible to reduce the time required for the transfer function of the transmission filter to approach the inverse characteristics of the transmission path.

[0109] The optical communication system 100 configured as above also executes inference processing, which, as explained in <<Effects of Executing Inference Processing>>, makes it possible to reduce the time required for the transfer function of the transmission filter to approach the inverse characteristics of the transmission path.

[0110] The receiving device 2 configured in this manner also executes inference processing, which, as explained in <<Effects of Executing Inference Processing>>, reduces the time required for the transfer function of the transmission filter to approach the inverse characteristics of the transmission path.

[0111] The transmitting device 1 configured in this manner also includes a transmitting filter and transmits an optical signal converted by the transmitting filter. The transmitting device 1 receives the optical signal thus transmitted and performs inference processing based on the reception result. Therefore, the transmitting device 1 can reduce the time required to optimize the transmitting filter.

[0112] (Modification) The receiving device 2 does not necessarily have to be implemented in a single housing, and may be implemented using multiple information processing devices connected to each other so as to be able to communicate via a network. Therefore, for example, the housing containing the photoelectric conversion element 20 and the housing that executes the inference process may be different housings connected to each other via a network.

[0113] Furthermore, as described above, the execution of the learned filter coefficient inference model does not necessarily have to be performed by the receiving device 2. In other words, the execution of the inference process does not necessarily have to be performed by the receiving device 2. The execution of the inference process may be performed by, for example, the transmitting device 1. More specifically, the execution of the inference process may be performed by the transmission control unit 11.

[0114] The transformation by the transmission filter may be an identity transformation. The transformation by the transmission filter before the filter coefficients are updated based on the result of the execution of the inference process may be an identity transformation.

[0115] The filter coefficient inference model may be configured, for example, by a neural network, which may convert the feature quantities of the signal output from the photoelectric conversion element 20 into a feature quantity vector and infer how to update the filter coefficients based on the resulting feature quantity vector.

[0116] <Example in which multiple reference filter coefficients are prepared> In the examples described above, when the filter coefficients are updated through a single estimation using the trained filter coefficient inference model, communication is performed using a non-test optical signal. However, for example, if the transmission path distance is greater than a certain value, the distortion of the received waveform increases, resulting in a high frequency exceeding the receiver's bandwidth, and part of the distortion information of the received waveform may be removed by the receiver. In this case, the received signal input to the trained filter coefficient inference model does not properly retain information about the transmission path. As a result, the accuracy of inference using the trained filter coefficient inference model may be reduced.

[0117] One possible technique for suppressing this degradation is to prepare multiple reference filter coefficients, sequentially calculate the transmit filter coefficients when these are set, and select from these the transmit filter coefficient that transmits the non-test signal without bit errors. Now, an optical communication system 100a will be described as an optical communication system that implements such a technique.

[0118] 8 is an explanatory diagram illustrating a modified optical communication system 100a. The optical communication system 100a differs from the optical communication system 100 in that it includes a receiving device 2a instead of the receiving device 2. The receiving device 2a differs from the receiving device 2 in that it includes a receiving control device 21a instead of the receiving control unit 21. The receiving control device 21a is capable of executing the processes executed by the receiving control unit 21, and also includes a transmission quality determination unit 213. The transmission quality determination unit 213 executes the m-th transmission quality determination process described below.

[0119] In the optical communication system 100a, unit processes are repeatedly executed until a predetermined termination condition for terminating the repetition (hereinafter referred to as the "repetition termination condition") is satisfied. Each unit process uses one of a plurality of reference filter coefficients prepared in advance. The reference filter coefficient used in each unit process is different from that used in other unit processes in at least some unit processes. Therefore, the reference filter coefficient used in the mth unit process of the repetition (m is an integer between 1 and M, inclusive; M is a predetermined integer greater than or equal to 1) is hereinafter referred to as the mth reference filter coefficient. Furthermore, the mth unit process is hereinafter referred to as the mth unit process.

[0120] <<mth Unit Processing>> In the mth unit processing, a conversion is performed on the test signal using a transmission filter whose filter coefficients are the mth reference filter coefficients, and an optical signal representing the converted signal (hereinafter referred to as the "mth reference transmission signal") is output from the transmitter 10. The mth reference transmission signal is converted into a reception signal by the photoelectric conversion element 20. Hereinafter, the reception signal obtained by the photoelectric conversion element 20 that has received the mth reference transmission signal will be referred to as the mth reference reception signal.

[0121] In the m-th unit processing, information indicating the m-th reference filter coefficient (hereinafter referred to as "m-th reference filter coefficient information") is transmitted from the transmitting device 1 to the receiving device 2a and acquired by the reception control unit 21a. As is clear from the definition, the m-th reference filter coefficient information is a type of reference filter coefficient information.

[0122] In the m-th unit process, the reception control unit 21a executes the m-th inference process. The m-th inference process is an inference process executed in the m-th unit process. Specifically, the m-th inference process is a process of inputting the m-th reference filter coefficient information and the m-th reference received signal into a trained filter coefficient inference model and executing the trained filter coefficient inference model.

[0123] That is, the mth inference process is a process that uses a learned filter coefficient inference model to infer what value the filter coefficient of the transmission filter should be updated to based on the mth reference received signal and the mth reference filter coefficient information. As is clear from the definition, the mth inference process is a type of inference process.

[0124] When the mth inference process is performed, the mth filter update process is executed in the mth unit process. The mth filter update process is a process of updating the transmission filter coefficient to the filter coefficient indicated by the mth inferred filter coefficient information. The mth inferred filter coefficient information is inferred filter coefficient information obtained by executing the mth inference process. As is clear from the definition, the mth filter update process is a type of filter update process.

[0125] In the mth unit process, a conversion is performed on the test signal using the transmit filter whose filter coefficients have been updated by the mth filter update process, and the transmitter 10 outputs an optical signal representing the converted signal (hereinafter referred to as the "mth non-reference transmit signal").

[0126] The mth non-reference transmission signal is converted into a reception signal by the photoelectric conversion element 20. Hereinafter, the reception signal obtained by the photoelectric conversion element 20 that has received the mth non-reference transmission signal will be referred to as the mth non-reference reception signal.

[0127] In the m-th unit process, the reception control unit 21a executes the m-th transmission quality determination process, which is a process for determining whether the transmission quality of the optical communication system 100a satisfies a predetermined standard based on the m-th non-reference received signal.

[0128] The value indicating the transmission quality may be, for example, a bit error rate, a Q factor, or a signal-to-noise ratio (SNR). When the transmission quality satisfies a predetermined standard, the optical communication system 100a is a system with no bit errors.

[0129] The above-mentioned repetition end condition includes a condition that the transmission quality satisfies a predetermined standard. Therefore, when it is determined that the transmission quality satisfies the predetermined standard by the m-th transmission quality determination process, the repetition of the unit process ends.

[0130] If the m-th transmission quality determination process does not determine that the transmission quality satisfies the predetermined standard, the m-th reference filter update process is executed. The m-th reference filter update process is a process for updating the transmission filter coefficients to the (m+1)-th reference filter coefficients. As is clear from the definition, the m-th reference filter update process is a type of filter update process.

[0131] After the m-th reference filter update process, the (m+1)-th unit process is executed.

[0132] The reception control unit 21a includes a transmission quality determination unit 213. The transmission quality determination unit 213 executes the mth transmission quality determination process. The mth reference filter coefficient information is input to the reception device 2a via a predetermined control signal channel or manually. For example, the control signal transmission / reception unit 111 may transmit the mth reference filter coefficient information, and in this case, for example, the control signal transmission / reception unit 212 may receive the mth reference filter coefficient information. For example, in this manner, the mth reference filter coefficient information may be transmitted from the transmission device 1 to the reception device 2a.

[0133] In the example of Figure 8, the transmission quality determination unit 213 may output information indicating the result of the determination made by the transmission quality determination process (hereinafter referred to as "determination result information") to the control signal transmitting / receiving unit 212. Having acquired the determination result information, the control signal transmitting / receiving unit 212 transmits the determination result information to the transmitting device 1. The transmitted determination result information is acquired by the transmission control unit 11. The transmission control unit 11 performs an operation according to the determination result indicated by the acquired determination result information. Specifically, in the mth unit process, if the determination result information does not indicate that the transmission quality satisfies the predetermined standard, the transmission control unit 11 performs the mth reference filter update process. On the other hand, in the mth unit process, if the determination result information indicates that the transmission quality satisfies the predetermined standard, the transmission control unit 11 does not perform the mth reference filter update process, and the repetition of the unit process ends.

[0134] 9 is a flowchart showing an example of the flow of processing executed in the optical communication system 100a of the modified example. The transmission control unit 11 acquires a test signal (step S201). Next, the transmission control unit 11 converts the test signal using a transmission filter (step S202). The filter coefficient of the transmission filter here is the mth reference filter coefficient in the mth unit processing.

[0135] Next, the transmission control unit 11 controls the operation of the transmitter 10 to cause the transmitter 10 to transmit an optical signal representing the converted test signal (step S203). Here, the optical signal transmitted by the transmitter 10 is the m-th reference transmission signal in the m-th unit process.

[0136] Next, the photoelectric conversion element 20 receives the optical signal transmitted in step S203 (step S204). The photoelectric conversion element 20 receives the optical signal and obtains a reception signal indicating the waveform of the received optical signal. The reception signal obtained here is the m-th reference reception signal in the m-th unit processing.

[0137] Next, the transmission control unit 11 transmits reference filter coefficient information indicating the filter coefficients used in step S202 to the receiving device 2a via the control signal transmitting / receiving unit 111 (step S205). Next, the reception control unit 21a receives the reference filter coefficient information transmitted in step S205 via the control signal transmitting / receiving unit 212 (step S206). Note that the reference filter coefficient information transmitted and received in steps S205 and S206 is the m-th reference filter coefficient information in the m-th unit processing.

[0138] Next, the reception control unit 21a uses the learned filter coefficient inference model to infer what values ​​the filter coefficients of the transmission filter should be updated to, based on the reference filter coefficient information acquired in step S206 and the received signal obtained in step S204 (step S207). That is, the reception control unit 21a executes inference processing based on the reference filter coefficient information acquired in step S206 and the received signal obtained in step S204. Note that the inference processing in step S207 is the m-th inference processing in the m-th unit processing.

[0139] Next, the transmission control unit 11 executes a filter update process to update the transmission filter coefficients to the filter coefficients indicated by the result of the inference process (step S208). Note that the filter update process of step S208 is the m-th filter update process in the m-th unit process.

[0140] Next, the transmission control unit 11 acquires a test signal (step S209). Next, the transmission control unit 11 converts the test signal using a transmission filter (step S210). The filter coefficients of the transmission filter here are the transmission filter coefficients updated by the m-th filter update process in the m-th unit process.

[0141] Next, the transmission control unit 11 controls the operation of the transmitter 10 to cause the transmitter 10 to transmit an optical signal representing the converted test signal (step S211). Here, the optical signal transmitted by the transmitter 10 is the mth non-reference transmission signal in the mth unit process.

[0142] Next, the photoelectric conversion element 20 receives the optical signal transmitted in step S211 (step S212). The photoelectric conversion element 20 receives the optical signal and obtains a received signal indicating the waveform of the received optical signal. The received signal obtained here is the mth non-reference received signal in the mth unit processing.

[0143] Next, the reception control unit 21a executes a transmission quality determination process (step S213). The transmission quality determination process is a process for determining whether the transmission quality of the optical communication system 100a satisfies a predetermined standard based on the received signal. Therefore, the m-th transmission quality determination process is a type of transmission quality determination process. Note that the transmission quality determination process executed in step S213 is the m-th transmission quality determination process in the m-th unit process.

[0144] Next, the reception control unit 21a determines whether the repetition end condition is satisfied (step S214). If the repetition end condition is satisfied (step S214: YES), determination result information is sent to the transmitting device 1 (step S215). The determination result information in step S215 indicates that the repetition end condition is satisfied, and is therefore information indicating that the transmission quality satisfies a predetermined standard. Since the repetition end condition is satisfied, the repetition of the unit process ends. In other words, the process ends.

[0145] On the other hand, if the repetition end condition is not satisfied (step S214: NO), information indicating that the repetition end condition is not satisfied is transmitted from the reception control unit 21a to the transmission control unit 11 (step S216). Here, the information indicating that the repetition end condition is not satisfied is the determination result information. Having acquired the determination result information, the transmission control unit 11 updates the transmission filter coefficients to other reference filter coefficients (step S217). Here, the other reference filter coefficients are, for example, unused reference filter coefficients. For example, if the filter coefficient of the transmission filter used in step S202 was the mth reference filter coefficient, the other reference filter coefficient is the (m+1)th reference filter coefficient.

[0146] The series of steps from step S201 to step S217 is an example of a unit process, and after step S217, the process of step S201 starts. That is, the next unit process starts. For example, if step S217 is the process of the m-th unit process, the (m+1)-th unit process starts from the next step S201.

[0147] The processes of steps S205 and S206 may be executed at any timing as long as they are executed before step S207 is executed.

[0148] The m-th reference filter coefficient may be selected from a plurality of coefficients prepared in advance according to a predetermined rule, and the (m+1)-th filter coefficient may be a filter coefficient calculated in advance according to a predetermined rule based on the m-th filter coefficient.

[0149] In addition, if multiple reference filter coefficients are prepared in advance, they may be prepared for each length of the transmission medium 3 through which the optical signal transmitted from the transmitting device 1 passes during propagation to the receiving device 2a.

[0150] <Averaging Process> The reception control unit 21 may also perform averaging. The averaging process is performed, for example, when an optical signal is transmitted K times (K is an integer greater than or equal to 2) from the transmitting device without changing the filter coefficients of the transmission filter and the signal input to the transmission filter. In this case, the photoelectric conversion element 20 receives the optical signal K times and outputs a signal a total of K times corresponding to each reception. In this case, the averaging process obtains an average of the K signals output from the photoelectric conversion element 20. The signal obtained by averaging K times has reduced noise compared to a single average. Therefore, if the signal obtained by such averaging process is used as input to the filter coefficient inference model, the estimation accuracy is higher than that of estimation process based on signals that have not been subjected to averaging process.

[0151] Furthermore, the averaging process may be performed, for example, when the signal input to the transmit filter is a signal in which a predetermined waveform is repeated K times with a period T. In this case, the averaging process divides the signal output from the photoelectric conversion element 20 into K waveforms by dividing it at the period T, and then takes the average of the K divided waveforms. This averaging reduces noise compared to the signal before averaging. Therefore, if a signal in which the averaged waveform is repeated K times is input to the estimation process, the accuracy of the estimation will be higher than that of estimation processes based on a signal that has not been subjected to averaging (i.e., the signal directly output from the photoelectric conversion element 20).

[0152] 10 is an explanatory diagram illustrating the averaging process in the modified example. In the example of FIG. 10, ０ From time t １ The signal for one period up to time t １ From time t ２ The signal for one period up to time t ２ From time t ３ A signal for one period is obtained by averaging the signals for one period up to and including the three signals in. In the example of Figure 10, the signal obtained by averaging the three signals above, removing noise, and converting them into a discrete received waveform is shown as signal A101.

[0153] Incidentally, in the above-mentioned filter coefficient inference model, it is conceivable to use a received signal as an explanatory variable. When training a model using a received signal without phase shift, the received signal input to the filter coefficient inference model in the inference stage must also be a signal without phase shift. However, because the phase of the received signal changes randomly over time, there is a concern that if the received signal is input to the model as is, the accuracy of the inference will decrease. To avoid this, it is effective to create a received signal without phase shift by phase shift compensation and use this as input to the filter coefficient inference model. Below, a method of phase shift compensation will be explained.

[0154] <Phase Shift Compensation Processing> The reception control unit 21 may execute phase shift compensation processing, which is processing for compensating for a shift in the sampling phase.

[0155] Fig. 11 is a first explanatory diagram illustrating a phase shift compensation process in the modified example. Fig. 12 is a second explanatory diagram illustrating a phase shift compensation process in the modified example. Fig. 13 is a third explanatory diagram illustrating a phase shift compensation process in the modified example.

[0156] 11 is a block diagram illustrating the phase shift compensation process for ease of explanation. The blocks in the block diagram represent processes. In FIG. 11, the symbol t represents time.

[0157] In the phase shift compensation process, first oversampling, second oversampling, cross-correlation acquisition process, and downsampling are performed.

[0158] The first oversampling is a process of performing oversampling on a received signal. The received signal is a signal output by the photoelectric conversion element 20 that receives an optical signal representing a test signal converted by a conversion process, and is assumed to be discretized. The discretization is performed, for example, by an analog-to-digital converter (ADC) shown in FIG. 1. In the example of FIG. 11, a time series C101 represents an example of a received signal input to the first oversampling. In the example of FIG. 11, a time series C102 represents an example of an oversampled received signal.

[0159] The second oversampling is a process of oversampling a test signal previously stored in the storage unit 23. The test signal input to the second oversampling is a series in which the first bit is the first sample. In the example of Fig. 11, time series C103 represents an example of the test signal input to the second oversampling. In the example of Fig. 11, time series C104 represents an example of the oversampled test signal.

[0160] 11 also shows the symbol positions of the time series C103, indicating that symbols exist at the sampling positions in the time series C103. Here, the sample points are used as the reference for the sample phase, which becomes the symbol positions of the received signal. In other words, the test signal C103 is a signal with no phase shift. On the other hand, the received signal C101 does not have sample points at the symbol positions, meaning that a phase shift occurs.

[0161] The cross-correlation acquisition process is a process for obtaining conditions that maximize the correlation between the oversampled received signal and the oversampled test signal. Fig. 12 is an explanatory diagram illustrating the cross-correlation acquisition process. More specifically, Fig. 12 shows the timing at which the cross-correlation occurs between the time series C102 and the time series C104 in Fig. 11. In Fig. 12, the symbol q represents the amount of sample shift in the positive direction of the time axis.

[0162] Time series C105 in Fig. 12 is time series C104 shifted by one sample in the positive direction of the time axis. Time series C106 in Fig. 12 is time series C104 shifted by another one sample in the positive direction of the time axis from time series C105. Time series C107 in Fig. 12 is time series C104 shifted by another bit in the positive direction of the time axis from time series C106. Time series C108 in Fig. 12 is time series C104 shifted by another bit in the positive direction of the time axis from time series C107.

[0163] Fig. 12 shows that the time series C 108 has the greatest correlation with time series C 102. In other words, Fig. 12 indicates that shifting time series C 104 by 4 bits in the positive direction of the time axis maximizes the correlation with time series C 102.

[0164] In this way, the cross-correlation acquisition process obtains the condition that maximizes the correlation between the oversampled received signal and the oversampled test signal. Note that the condition obtained in the example of Figure 12 is the condition that shifts the time series C104 by 4 bits in the positive direction of the time axis.

[0165] Downsampling is a process of downsampling an oversampled received signal using the oversampled received signal and the result of the cross-correlation acquisition process, and the output of the downsampling is phase-shift-free.

[0166] FIG. 13 is an explanatory diagram specifically illustrating downsampling, and more specifically, a diagram illustrating the process of downsampling a received signal that has been oversampled using the conditions obtained in the example of FIG. 12 . The conditions obtained in the example of FIG. 12 are that the time series C104 is shifted by four bits in the positive direction of the time axis. Therefore, in downsampling, the section from the fourth bit onwards of the time series C102 is extracted, and the extracted signal is downsampled. This results in a signal without phase shift. The time series C109 in FIG. 13 is the result of extracting the section from the fourth bit onwards of the time series C102 and downsampling the extracted signal.

[0167] By using the received signal whose phase shift has been compensated for by the above processing as an input to the filter coefficient estimation model, highly accurate estimation can be performed.

[0168] The reception control unit 21 or 21a may perform both averaging and phase shift compensation. In this case, the phase shift compensation is performed on the signal averaged by the averaging. In other words, the phase shift compensation is performed on the result of the averaging.

[0169] <Inference Based on Multiple Signals> The filter coefficient inference model is a mathematical model that infers the values ​​to which the transmit filter coefficients should be updated based on the signals input to the model and the transmit filter coefficients. Here, the signal input to the model does not necessarily have to be one. Specifically, the filter coefficient inference model may use G (G is an integer greater than or equal to 2) different filter coefficients to infer the values ​​to which the transmit filter coefficients should be updated based on a total of G received signals obtained for each filter coefficient and each filter coefficient.

[0170] When the filter coefficient inference model infers, based on G signals, what values ​​the transmit filter coefficients should be updated to, the filter coefficient inference model also infers, based on G signals, what values ​​the transmit filter coefficients should be updated to in the optical communication system 100. That is, in the optical communication system 100, the learned filter coefficient inference model infers, based on G signals, what values ​​the transmit filter coefficients should be updated to.

[0171] Therefore, in such a case, G mutually different reference filter coefficients, from a first reference filter coefficient to a Gth reference filter coefficient, are prepared in advance in the optical communication system 100. The transmission control unit 11 converts the test signal using a transmission filter whose filter coefficient is the gth reference filter coefficient, and causes the transmitter 10 to transmit an optical signal representing the converted test signal, from g=1 to g=G. Here, the test signal may be the same regardless of g, where g is an integer between 1 and G.

[0172] The photoelectric conversion element 20 receives these K signals in total and obtains corresponding received signals. In the inference process, the learned filter coefficient inference model is executed to infer the values ​​to which the transmit filter coefficients should be updated based on these G received signals and the G reference filter coefficients from the first reference filter coefficient to the Gth reference filter coefficient.

[0173] The effect of the filter coefficient inference model being inferred based on G received signals will be described with reference to Figures 14 and 15. Figure 14 is a first explanatory diagram illustrating a filter coefficient inference model in a modified example.

[0174] Consider two types of communication in which the transmit filter coefficients are the same but the bands of the transmitter 10 or the transmission medium 3 are different. In the example of Fig. 14, the transfer function of the transmitter 10 or the transmission medium 3 is transfer function A in one communication, and transfer function B, which is different from transfer function A, in the other communication.

[0175] In the example of Fig. 14, the transmission filter coefficient is filter coefficient 1 in both of the two types of communication. Also in the example of Fig. 14, the transmission medium 3 includes an optical amplifier and a fiber. More specifically, one of the two types of communication, which has a transfer function A, includes an optical amplifier A and a fiber A. And the other of the two types of communication, which has a transfer function B, includes an optical amplifier B and a fiber B.

[0176] In such cases, since the transfer function of the transmitter 10 or the transmission medium 3 is different, even if the signal converted by the transmission filter is the same, the signal received by the photoelectric conversion element 20 will be different. However, although the signals themselves are not substantially identical, the optical intensity (envelope) may be substantially identical. Since the photoelectric conversion element 20 outputs a signal indicating the optical intensity of the received signal, in such cases the signal output by the photoelectric conversion element 20 (i.e., the received signal) will be substantially identical. The detection method is, for example, direct detection.

[0177] The filter coefficient inference model is a mathematical model that performs inference based on received signals. Therefore, if the received signals are the same, the same inference results are obtained. However, because the transfer functions of the two types of communication are different, the transmit filter coefficients inferred by the filter coefficient inference model for each of the above communications should be different for each of the above communications. Thus, even if the signals received by the photoelectric conversion element 20 are different and not substantially identical, the accuracy of the inference of the filter coefficient inference model may be reduced if the light intensities are substantially identical.

[0178] 14, "received optical field" represents a signal received by the photoelectric conversion element 20. In addition, "received optical intensity" represents a received signal.

[0179] Fig. 15 is a second explanatory diagram illustrating a filter coefficient inference model in a modified example. Here, a case will be considered in which two filter coefficients (i.e., K=2) are used in one communication and the other communication in Fig. 14. Specifically, in the example of Fig. 15, a case will be considered in which two different filter coefficients, filter coefficient 1 and filter coefficient 2, are used. Filter coefficient 1 is the same as filter coefficient 1 in Fig. 14.

[0180] In this case, as described in FIG. 14 , the envelope of the signal converted by filter coefficient 1 and then received by photoelectric conversion element 20 is substantially identical for one of the two types of communication. On the other hand, since filter coefficient 2 is different from filter coefficient 1, it is unlikely that the envelopes for one of the two types of communication will be substantially identical. In the example of FIG. 15 , they are not substantially identical. Therefore, the received signals for the two types of communication clearly show the difference in transfer function more clearly than in the case of FIG. 14 . Therefore, the inference accuracy of the filter coefficient inference model is higher in the case of FIG. 15 than in the case of FIG. 14 .

[0181] In this way, the filter coefficient inference model may be a mathematical model that infers to what value the filter coefficient of the transmit filter should be updated, based on K receive signals from a receive signal obtained when the filter coefficient of the transmit filter is a first filter coefficient to a receive signal obtained when the filter coefficient of the transmit filter is a K-th filter coefficient, and on K filter coefficients from the first filter coefficient to the K-th filter coefficient, where the K filter coefficients are different from one another.

[0182] The same applies to the optical communication system 100a, in which K different reference filter coefficients are used in each unit process. Based on the K total received signals obtained from the reference filter coefficients and the reference filter coefficients, each unit process infers the coefficients to which the transmit filter coefficients should be updated. The K filter coefficients may be different for each unit process.

[0183] The transmitting device 1 may be implemented using a plurality of information processing devices connected to each other via a network so that they can communicate with each other. In this case, the processes executed by the transmission control unit 11 may be distributed among the plurality of information processing devices.

[0184] As described above, the receiving device 2 may be implemented using a plurality of information processing devices connected to each other so as to be able to communicate via a network. In this case, the processes executed by the reception control unit 21 may be distributed among the plurality of information processing devices. Accordingly, the receiving device 2a may also be implemented using a plurality of information processing devices connected to each other so as to be able to communicate via a network. In this case, the processes executed by the reception control unit 21a may be distributed among the plurality of information processing devices.

[0185] The learning device 4 may be implemented using a plurality of information processing devices connected to each other via a network so that they can communicate with each other. In this case, the processes executed by the learning unit 41 may be distributed among the plurality of information processing devices.

[0186] All or part of the functions of the optical communication system 100, the optical communication system 100a, the transmitting device 1, the receiving device 2, and the learning device 4 may be realized using hardware such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). The program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. The program may be transmitted via a telecommunications line.

[0187] The mth reference transmission signal is an example of a first-type transmission signal. The mth non-reference transmission signal is an example of a second-type transmission signal. The mth reference reception signal is an example of a first-type reception signal. The mth non-reference reception signal is an example of a second-type reception signal.

[0188] The transmission control unit 11 is an example of a control unit that executes the learned mathematical model, and the reception control unit 21 is an example of a control unit that executes the learned mathematical model.

[0189] 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.

[0190] 100, 100a...optical communication system, 1...transmitting device, 2, 2a...receiving device, 10...transmitter, 11...transmission control section, 20...photoelectric conversion element, 21, 21a...receiving control section, 12...interface section, 13...storage section, 22...interface section, 23...storage section, 4...learning device, 41...learning section, 42...interface section, 43...storage section, 91...processor, 92...memory, 93...processor, 94...memory, 95...processor, 96...memory

Claims

1. An optical communication system comprising: a transmitter that transmits a signal converted by a filter; a photoelectric conversion element that receives the signal; and a learning unit that learns a mathematical model that infers the value to which the filter coefficients of the filter should be updated based on a signal input to the model and the filter coefficients of the filter, wherein the signal input to the mathematical model is the received signal obtained by the photoelectric conversion element, and a control unit that executes the learned mathematical model obtained by a learning device.

2. The optical communication system of claim 1, wherein the control unit executes the learned mathematical model to infer what value the filter coefficient of the filter should be updated to based on a first-type received signal, which is a received signal obtained by receiving a first-type transmitted signal that is a signal transmitted by the transmitter and converted by one of a plurality of pre-prepared filters, and the transmitter transmits a second-type transmitted signal, which is a signal converted by the filter whose filter coefficient is the filter coefficient updated in accordance with the result of the inference; the control unit determines whether the transmission quality of its own system satisfies a predetermined standard based on a second-type received signal, which is a received signal obtained by receiving the second-type transmitted signal by the photoelectric conversion element; and if the result of the determination is that the standard is not satisfied, the transmitter transmits a signal converted by another one of the plurality of pre-prepared filters.

3. The optical communication system according to claim 1 or 2, wherein the mathematical model infers the values ​​to which the filter coefficients of the filter should be updated based on K received signals from the received signal obtained when the filter coefficient of the filter is a first filter coefficient to the received signal obtained when the filter coefficient of the filter is a Kth filter coefficient (K is an integer equal to or greater than 2) and K filter coefficients from the first filter coefficient to the Kth filter coefficient, and the K filter coefficients are different from one another.

4. A transmitting device comprising: a transmitting filter which is a filter that converts an input signal; and a transmitter that transmits the signal converted by the transmitting filter, wherein the filter coefficients of the filter are updated to filter coefficients inferred by a learning unit that learns a mathematical model that infers values ​​to which the filter coefficients of the filter should be updated based on a signal input to the model and the filter coefficients of the filter, and the signal input to the mathematical model is a received signal obtained by a photoelectric conversion element that receives the signal transmitted by the transmitter, and the filter coefficients are updated to filter coefficients inferred by the learned mathematical model obtained by a learning device.

5. A receiving device comprising: a photoelectric conversion element that receives a signal transmitted by a transmitter that transmits a signal converted by a transmission filter that is a filter that converts an input signal; a learning unit that learns a mathematical model that infers the value to which the filter coefficient of the filter should be updated based on the signal input to the model and the filter coefficient of the filter, wherein the signal input to the mathematical model is the received signal obtained by the photoelectric conversion element; and a control unit that executes the learned mathematical model obtained by a learning device.

6. A learning device comprising: a learning unit that learns a mathematical model that infers what value the filter coefficient of the filter should be updated to, based on a received signal obtained by a photoelectric conversion element that receives a signal transmitted by a transmitter that transmits a signal converted by a transmission filter, which is a filter that converts an input signal, and the filter coefficient of the filter.

7. An optical communication method executed by an optical communication system comprising: a transmitter that transmits a signal converted by a filter; a photoelectric conversion element that receives the signal; and a learning unit that learns a mathematical model that infers the value to which the filter coefficients of the filter should be updated based on a signal input to the model and the filter coefficients of the filter, wherein the signal input to the mathematical model is a received signal obtained by the photoelectric conversion element; and a control unit that executes the learned mathematical model obtained by a learning device, the optical communication method comprising: a receiving step in which the photoelectric conversion element receives the signal; and an execution step in which the control unit executes the learned mathematical model.

8. A learning method performed by a learning device having a learning unit that learns a mathematical model that infers what value the filter coefficient of a filter should be updated to based on a received signal obtained by a photoelectric conversion element that receives a signal transmitted by a transmitter that transmits a signal converted by a transmission filter that is a filter that converts an input signal and the filter coefficient of the filter, the learning method comprising: a learning step in which the learning unit performs the learning.

Images