Wavefront sensor and adaptive optics system

Abstract

To provide a wavefront sensor that can sense wavefront aberrations caused by the medium at a faster speed. [Solution] The wavefront sensor 1 comprises a linear focusing element 15, a line sensor 16, and a wavefront aberration estimator 20. The linear focusing element 15 focuses light 7 that has traveled through the medium 6 in a linear fashion. The line sensor 16 receives the linearly focused light 7 and acquires a one-dimensional image 33 of the light 7. The wavefront aberration parameter estimation unit 38 of the wavefront aberration estimator 20 inputs a preprocessed image 33b of the one-dimensional image 33 to a wavefront aberration parameter estimation model 32 trained by machine learning, and causes the wavefront aberration parameter 34 representing the wavefront aberration caused by the medium 6 to be output to the wavefront aberration parameter estimation model 32.

Description

Technical Field

[0001] The present disclosure relates to a wavefront sensor and an adaptive optics apparatus.

Background Art

[0002] Yohei Nishizaki et.al., “Deep learning wavefront sensing”, Optics Express, January 7, 2019, Vol.27, No.1, p.240-251 (Non-Patent Document 1) discloses wavefront sensing using deep learning.

Prior Art Literature

Non-Patent Literature

[0003]

Non-Patent Document 1

Summary of the Invention

Problem to be Solved by the Invention

[0004] An object of a first aspect of the present disclosure is to provide a wavefront sensor capable of sensing wavefront aberration caused by a medium at higher speed. An object of a second aspect of the present disclosure is to provide an adaptive optics apparatus capable of compensating for wavefront aberration caused by a medium at higher speed.

Means for Solving the Problem

[0005] The wavefront sensor of this disclosure comprises a linear focusing element, a line sensor, and a wavefront aberration estimator. The linear focusing element focuses light traveling through a medium in a linear manner. The line sensor receives the light focused linearly by the linear focusing element and acquires a one-dimensional image of the light. The wavefront aberration estimator includes an image preprocessing unit and a wavefront aberration parameter estimation unit. The image preprocessing unit preprocesses the one-dimensional image to generate a preprocessed image. The wavefront aberration parameter estimation unit inputs the preprocessed image into a wavefront aberration parameter estimation model trained by machine learning and causes the wavefront aberration parameter estimation model to output wavefront aberration parameters that represent wavefront aberrations caused by the medium.

[0006] The adaptive optics apparatus of this disclosure comprises a wavefront sensor, a spatial light modulator, and a controller. The controller is communicatively connected to the spatial light modulator and the wavefront sensor and can control the spatial light modulator. The controller receives wavefront aberrations caused by the medium from the wavefront sensor and causes the spatial light modulator to form a two-dimensional phase distribution. The compensated wavefront aberrations caused by the two-dimensional phase distribution cancel out the wavefront aberrations caused by the medium. [Effects of the Invention]

[0007] The wavefront sensor of this disclosure can sense wavefront aberrations caused by the medium at a faster speed. The adaptive optics apparatus of this disclosure can compensate for wavefront aberrations caused by the medium at a faster speed. [Brief explanation of the drawing]

[0008] [Figure 1] This is a schematic diagram of the wavefront sensor according to Embodiment 1. [Figure 2] This figure shows the hardware configuration of the wavefront aberration estimator according to Embodiment 1. [Figure 3] This is a block diagram illustrating the functional configuration of the wavefront aberration estimator of Embodiment 1. [Figure 4A] This figure shows an example of pixel rearrangement. [Figure 4B] This figure shows another example of pixel rearrangement. [Figure 5]This diagram shows flowcharts of the wavefront aberration sensing methods of Embodiment 1 and Embodiment 2. [Figure 6] This figure shows flowcharts of the wavefront aberration estimation steps in Embodiment 1 and Embodiment 2. [Figure 7A] This figure shows an example of a one-dimensional image output from the line sensor of a wavefront aberration estimator. [Figure 7B] This figure shows a graph of the wavefront aberration parameters obtained by the wavefront aberration estimator. [Figure 7C] This figure shows an example of a phase map of wavefront aberration caused by the medium. [Figure 8] This diagram shows flowcharts illustrating the methods for generating wavefront aberration parameter estimation models in Embodiment 1 and Embodiment 2. [Figure 9] This is a schematic diagram of the training dataset generation device according to Embodiment 1. [Figure 10] This figure shows the hardware configuration of the training data generators in Embodiment 1 and Embodiment 2. [Figure 11] This diagram shows flowcharts illustrating the methods for generating training datasets in Embodiment 1 and Embodiment 2. [Figure 12A] This figure shows an example of wavefront aberration parameters generated by the training data generator. [Figure 12B] This figure shows an example of a two-dimensional phase distribution formed in the spatial light modulator of a training dataset generator. [Figure 12C] This figure shows an example of a one-dimensional image acquired by a line sensor in a training dataset generation device. [Figure 13] This figure shows the hardware configuration of the wavefront aberration parameter estimation model generation device of Embodiment 1 and Embodiment 2. [Figure 14] This is a block diagram illustrating the functional configuration of the wavefront aberration parameter estimation model generation apparatus in Embodiment 1 and Embodiment 2. [Figure 15] This is a block diagram illustrating the processing (learning function) of the machine learning unit in Embodiment 1 and Embodiment 2. [Figure 16]It is a diagram showing a flowchart of the step of training the wavefront aberration parameter estimation models of Embodiment 1 and Embodiment 2 by machine learning. [Figure 17] It is a schematic diagram of the adaptive optics apparatus of Embodiment 1 and Embodiment 2. [Figure 18] It is a schematic diagram of the wavefront sensor of Embodiment 2. [Figure 19] It is a schematic diagram of the training dataset generation apparatus of Embodiment 2. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present disclosure will be described. The same reference numerals are assigned to the same configurations, and descriptions thereof will not be repeated.

[0010] (Embodiment 1) <Wavefront sensor 1> With reference to FIG. 1, the wavefront sensor 1 of Embodiment 1 will be described. The wavefront sensor 1 can sense wavefront aberration caused by a medium 6. The wavefront sensor 1 can be applied to, for example, a receiver of an optical satellite communication apparatus. When the wavefront sensor 1 is applied to a receiver of an optical satellite communication apparatus, the wavefront sensor 1 is installed, for example, on the ground surface. The wavefront sensor 1 includes a linear condensing element 15, a line sensor 16, and a wavefront aberration estimator 20. The wavefront sensor 1 may further include a two-dimensional condensing element 10 and a mask 11.

[0011] With reference to FIG. 1, a light source 3 emits light 4. When the wavefront sensor 1 is applied to a receiver of an optical satellite communication apparatus, the light source 3 is included in, for example, a transmitter of the optical satellite communication apparatus and mounted on a satellite. The light source 3 is, for example, a laser light source or a super luminescent diode (SLD).

[0012] Light 4 travels through medium 6. When wavefront sensor 1 is applied to a receiver of an optical satellite communication device, medium 6 is the Earth's atmosphere. Light 4 is subjected to aberrations in medium 6 and becomes light 7 with a distorted wavefront. The wavefront aberration caused by medium 6, which is sensed by wavefront sensor 1, is the difference between the wavefront of light 4 and the wavefront of light 7. In this specification, light 4 and 7 are not limited to light such as infrared, visible, and ultraviolet light, but include millimeter waves, radio waves, and X-rays.

[0013] The two-dimensional light-gathering element 10 is positioned on the incident side of the light 7 in the optical path 8 of the light 7, relative to the mask 11, the linear light-gathering element 15, and the line sensor 16. The light 7 is incident on the two-dimensional light-gathering element 10. The two-dimensional light-gathering element 10 isotropically focuses the light 7 in a plane perpendicular to the optical path 8 of the light 7. The two-dimensional light-gathering element 10 is not particularly limited, but may be a spherical lens such as a plano-convex or biconvex lens, or it may be at least one curved mirror.

[0014] The mask 11 is positioned between the two-dimensional light-gathering element 10 and the linear light-gathering element 15 in the optical path 8 of the light 7. The mask 11 is positioned on the rear focal plane of the two-dimensional light-gathering element 10 and on the front focal plane of the linear light-gathering element 15. The mask 11 includes a central light-shielding region 12, a peripheral light-shielding region 13, and an annular aperture 14 formed between the central light-shielding region 12 and the peripheral light-shielding region 13. The light 7 passes through the annular aperture 14. The diameter of the annular aperture 14 is, for example, 1 mm. The central light-shielding region 12 blocks the image of the light source 3, enabling the line sensor 16 to detect the light 7 with high accuracy. The diameter of the central light-shielding region 12 is, for example, 100 μm.

[0015] The linear focusing element 15 is positioned between the mask 11 and the line sensor 16 in the optical path 8 of the light 7. The linear focusing element 15 is an optical element in which, in a plane perpendicular to the optical path 8 of the light 7, the focusing force in the direction perpendicular to the arrangement direction of the multiple pixels of the line sensor 16 is greater than the focusing force in the arrangement direction of the multiple pixels of the line sensor 16, and thus focuses the light 7 mainly in the direction perpendicular to the arrangement direction of the multiple pixels of the line sensor 16. In this way, the linear focusing element 15 focuses the light 7 in a linear shape. The linear focusing element 15 may be, for example, a linear focusing lens such as a cylindrical lens, or it may be at least one curved mirror.

[0016] The line sensor 16 is positioned on the rear focal plane of the linear focusing element 15. The line sensor 16 includes a plurality of pixels arranged uniformly. Each of the plurality of pixels includes, for example, a CMOS (Complementary Metal Oxide Semiconductor) sensor. The line sensor 16 receives light 7 focused linearly by the linear focusing element 15 and acquires a one-dimensional image 33 of the light 7 (see Figures 2 and 7A). The one-dimensional image 33 of the light 7 represents the one-dimensional intensity distribution of the light 7 on the line sensor 16. The line sensor 16 outputs the one-dimensional image 33 to the wavefront aberration estimator 20.

[0017] The frame rate of the line sensor 16 is, for example, 10 kHz or higher. The frame rate of the line sensor 16 may be 50 kHz or higher, or 100 kHz or higher. The frame rate of the line sensor 16 is the number of one-dimensional images 33 that the line sensor 16 can output per second. As the line sensor 16, for example, the "Xposure Camera" manufactured by the Austrian Institute of Technology, which has a frame rate of 600 kHz, or the "H2-HM-16k100H-00B" manufactured by Teledyne Technologies, which has a frame rate of 1 MHz, may be used.

[0018] The wavefront aberration estimator 20 receives a one-dimensional image 33 of light 7 from the line sensor 16. The wavefront aberration estimator 20 outputs wavefront aberration parameters 34 (see Figure 2) from the one-dimensional image 33. The wavefront aberration parameters 34 are parameters that represent the wavefront aberration caused by the medium 6. The wavefront aberration can be expressed, for example, by a Zernike polynomial. Referring to Figure 7B, the wavefront aberration parameters 34 are, for example, a set of coefficients of a Zernike polynomial. In this way, the wavefront aberration estimator 20 estimates the wavefront aberration caused by the medium 6 from the one-dimensional image 33. The wavefront aberration estimator 20 may further generate a phase map of the wavefront aberration caused by the medium 6 (see Figure 7C) from the wavefront aberration parameters 34.

[0019] <Hardware Configuration> Referring to Figure 2, the hardware configuration of the wavefront aberration estimator 20 will be described. The wavefront aberration estimator 20 includes an input device 23, a processor 24, memory 25, a display 26, a network controller 27, a media drive 28, and storage 30.

[0020] The input device 23 accepts various input operations. The input device 23 is, for example, a keyboard, mouse, or touch panel.

[0021] The display 26 displays the wavefront aberration parameter 34 (see Figure 7B) or the phase map of the wavefront aberration caused by the medium 6 generated from the wavefront aberration parameter 34 (see Figure 7C). The display 26 is, for example, an LCD (Liquid Crystal Display) or an organic EL (Electroluminescence) display. The display 26 receives the wavefront aberration parameter 34 or the phase map of the wavefront aberration caused by the medium 6 and displays the wavefront aberration parameter 34 or the phase map of the wavefront aberration caused by the medium 6.

[0022] The processor 24 performs the processing necessary to realize the functions of the wavefront aberration estimator 20 by executing the wavefront aberration estimation program 31 and the like. The processor 24 is composed of, for example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit).

[0023] Memory 25 provides a storage area for temporarily storing program code or work memory when the processor 24 executes the wavefront aberration estimation program 31, etc. Memory 25 is, for example, a volatile memory device such as DRAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory).

[0024] The network controller 27 transmits and receives programs or data to and from any device via a communication network (not shown), such as the Internet or an intranet. For example, the network controller 27 transmits the wavefront aberration parameter 34 or the phase map of the wavefront aberration caused by the medium 6 to an external computer (not shown), display (not shown), or storage device (not shown) of the wavefront aberration estimator 20 via the communication network. The network controller 27 receives the wavefront aberration parameter estimation model 32 from the wavefront aberration parameter estimation model generation device 60 (see Figure 13) via the communication network. The network controller 27 supports any communication method, such as Ethernet®, Wi-Fi (Local Area Network), or Bluetooth®.

[0025] The media drive 28 is a device that reads programs or data stored in the computer-readable medium 29. The media drive 28 may also be a device that writes programs or data to the computer-readable medium 29. The computer-readable medium 29 is a non-transitory storage medium that stores programs or data non-volatilely. The computer-readable medium 29 is, for example, an optical storage medium such as an optical disc (e.g., CD-ROM or DVD-ROM), a semiconductor storage medium such as flash memory or USB (Universal Serial Bus) memory, a magnetic storage medium such as a hard disk, FD (Flexible Disk) or storage tape, or a magneto-optical storage medium such as an MO (Magneto-Optical) disk.

[0026] The storage 30 stores the wavefront aberration estimation program 31, the wavefront aberration parameter estimation model 32, the one-dimensional image 33 of light 7, and the wavefront aberration parameters 34. The wavefront aberration parameter estimation model 32 is a model trained by machine learning, as will be described later. The wavefront aberration estimation program 31 is a program for obtaining the wavefront aberration parameters 34 from the one-dimensional image 33. The preprocessed image 33b of the one-dimensional image 33 is input to the wavefront aberration parameter estimation model 32, and the wavefront aberration parameter estimation model 32 outputs the wavefront aberration parameters 34. The storage 30 is, for example, a non-volatile memory device such as a hard disk or an SSD (Solid State Drive).

[0027] The program for implementing the functions of the wavefront aberration estimator 20 may be stored and distributed on a non-transient computer-readable medium 29 and installed on the storage 30. The program for implementing the functions of the wavefront aberration estimator 20 may also be downloaded to the wavefront aberration estimator 20 via a communication network such as the Internet or an intranet. The program for implementing the functions of the wavefront aberration estimator 20 includes a wavefront aberration estimation program 31 (see Figure 2).

[0028] In this embodiment, an example is shown in which a general-purpose computer (processor 24) implements the functions of the wavefront aberration estimator 20 by executing a program including the wavefront aberration estimation program 31. However, the embodiment is not limited to this, and all or part of the functions of the wavefront aberration estimator 20 may be implemented using an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field-Programmable Gate Array).

[0029] <Functional Configuration> Referring to Figure 3, an example of the functional configuration of the wavefront aberration estimator 20 will be explained. The wavefront aberration estimator 20 includes an image receiving unit 35 and a wavefront aberration estimation unit 36.

[0030] The image receiving unit 35 receives a one-dimensional image 33 of light 7 (see Figures 2 and 7A) from the line sensor 16. The image receiving unit 35 outputs the one-dimensional image 33 to the storage 30. The storage 30 stores the one-dimensional image 33. The image receiving unit 35 is implemented, for example, by a network controller 27 (see Figure 2).

[0031] The wavefront aberration estimation unit 36 ​​receives a one-dimensional image 33 of light 7 and outputs wavefront aberration parameters 34 (see Figure 7B). In this way, the wavefront aberration estimation unit 36 ​​estimates the wavefront aberration caused by the medium 6 from the one-dimensional image 33. The wavefront aberration estimation unit 36 ​​is a function of the wavefront aberration estimator 20, which is realized when the processor 24 (see Figure 2) executes the wavefront aberration estimation program 31. The wavefront aberration estimation unit 36 ​​includes an image preprocessing unit 37 and a wavefront aberration parameter estimation unit 38.

[0032] The image preprocessing unit 37 preprocesses the one-dimensional image 33 of light 7 to generate a preprocessed image 33b. For example, the image preprocessing unit 37 applies preprocessing to the one-dimensional image 33, such as white balance correction, contrast correction, resizing, normalization, noise reduction, or filtering. The image preprocessing unit 37 outputs the preprocessed image 33b to the wavefront aberration parameter estimation unit 38.

[0033] If the wavefront aberration parameter estimation model 32 includes a convolutional neural network (CNN), the image preprocessing unit 37 may convert the one-dimensional image 33 into a two-dimensional image and generate the two-dimensional image as a preprocessed image 33b. For example, the one-dimensional image 33 can be converted into a two-dimensional image by rearranging (reshaping) the pixels of the one-dimensional image 33 two-dimensionally. Figures 4A and 4B show examples of reshaping methods. The pixels of the one-dimensional image 33 are rearranged as shown by the dotted arrows in Figures 5A and 5B to convert it into a two-dimensional image.

[0034] The wavefront aberration parameter estimation unit 38 outputs wavefront aberration parameters 34 (see Figure 7B) from the preprocessed image 33b generated by the image preprocessing unit 37. The wavefront aberration parameter estimation unit 38 includes a wavefront aberration parameter estimation model 32. The wavefront aberration parameter estimation model 32 includes a neural network 32N and parameters 32P. The neural network 32N is a neural network classified as a deep neural network (DNN). The neural network 32N may also include a convolutional neural network (CNN).

[0035] Specifically, the wavefront aberration parameter estimation unit 38 receives a preprocessed image 33b from the image preprocessing unit 37. The wavefront aberration parameter estimation unit 38 inputs the preprocessed image 33b to the wavefront aberration parameter estimation model 32 and causes the wavefront aberration parameter 34 corresponding to the one-dimensional image 33 to be output by the wavefront aberration parameter estimation model 32. The wavefront aberration parameter estimation unit 38 may generate a phase map of wavefront aberration caused by the medium 6 (see Figure 7C) from the wavefront aberration parameter 34 and output the phase map of wavefront aberration caused by the medium 6. The wavefront aberration estimation unit 36 ​​outputs the wavefront aberration parameter 34 or the phase map of wavefront aberration caused by the medium 6 to at least one of, for example, the display 26 (see Figure 2) or the storage 30 (see Figure 2).

[0036] <Method for sensing wavefront aberration caused by medium 6> Referring to Figures 5 and 6, a method for sensing wavefront aberration caused by the medium 6 using the wavefront sensor 1 of this embodiment will be described. The wavefront aberration sensing method of this embodiment includes acquiring a one-dimensional image 33 of the light 7 that has traveled through the medium 6 (step S1) and estimating the wavefront aberration caused by the medium 6 from the one-dimensional image 33 (step S2).

[0037] Step S1 (see Figure 5) will be explained with reference to Figures 1 and 2. When the wavefront sensor 1 is applied to the receiver of an optical satellite communication device, the light source 3 is, for example, mounted on a satellite and is located far enough away from the Earth's surface where the wavefront sensor 1 is installed. Therefore, the light 4 emitted by the light source 3 reaches the medium 6 as a plane wave. The light 4 is subjected to aberrations in the medium 6 and becomes light 7 with a distorted wavefront. The wavefront aberration caused by the medium 6 sensed by the method of this embodiment is the difference between the wavefront of light 4 and the wavefront of light 7.

[0038] The two-dimensional focusing element 10 focuses the light 7 isotropically in a plane perpendicular to the optical path 8 of the light 7. The mask 11 is positioned on the back focal plane of the two-dimensional focusing element 10 and on the front focal plane of the linear focusing element 15. The central light-shielding region 12 of the mask 11 blocks the image of the light source 3. The light 7 passes through the annular aperture 14 of the mask 11. The linear focusing element 15 focuses the light 7 linearly onto the line sensor 16. The line sensor 16 receives the light 7 that has been linearly focused by the linear focusing element 15. The line sensor 16 acquires a one-dimensional image 33 of the light 7 (see Figures 2 and 7A). The line sensor 16 outputs the one-dimensional image 33 to the wavefront aberration estimator 20. The image receiving unit 35 (see Figure 3) receives the one-dimensional image 33 from the line sensor 16 and outputs it to the storage 30. Storage 30 stores a one-dimensional image 33.

[0039] Step S2 (see Figures 5 and 6) will be explained. The wavefront aberration estimation unit 36 ​​reads a one-dimensional image 33 from the storage 30. The wavefront aberration estimation unit 36 ​​estimates the wavefront aberration caused by the medium 6 from the one-dimensional image 33.

[0040] Specifically, the image preprocessing unit 37 preprocesses the one-dimensional image 33 to generate a preprocessed image 33b (step S3). The image preprocessing unit 37 performs preprocessing on the one-dimensional image 33, such as white balance correction, contrast correction, resizing, normalization, noise reduction, or filtering. The image preprocessing unit 37 outputs the preprocessed image 33b to the wavefront aberration parameter estimation unit 38. If the wavefront aberration parameter estimation model 32 includes a convolutional neural network (CNN), the image preprocessing unit 37 may convert the one-dimensional image 33 into a two-dimensional image and generate the two-dimensional image as the preprocessed image 33b. For example, the one-dimensional image 33 can be converted into a two-dimensional image by rearranging (reshaping) the pixels of the one-dimensional image 33 two-dimensionally.

[0041] The wavefront aberration parameter estimation unit 38 outputs wavefront aberration parameters 34 (see Figure 7B) corresponding to the one-dimensional image 33 from the preprocessed image 33b generated by the image preprocessing unit 37 (step S4). Specifically, the wavefront aberration parameter estimation unit 38 receives the preprocessed image 33b from the image preprocessing unit 37. The wavefront aberration parameter estimation unit 38 inputs the preprocessed image 33b to the wavefront aberration parameter estimation model 32 and causes the wavefront aberration parameters 34 corresponding to the one-dimensional image 33 to be output by the wavefront aberration parameter estimation model 32. The wavefront aberration parameter estimation unit 38 may also generate a phase map of wavefront aberration caused by the medium 6 (see Figure 7C) from the wavefront aberration parameters 34 and output the phase map of wavefront aberration caused by the medium 6.

[0042] The wavefront aberration estimation unit 36 ​​outputs the wavefront aberration parameter 34 (see Figure 7B) or the phase map of the wavefront aberration caused by the medium 6 (see Figure 7C) to at least one of, for example, the display 26 (see Figure 2) or the storage 30 (see Figure 2). The display 26 receives the wavefront aberration parameter 34 or the phase map of the wavefront aberration caused by the medium 6 and displays it. The storage 30 stores the wavefront aberration phase map caused by the wavefront aberration parameter 34 or the medium 6.

[0043] The wavefront aberration estimation program 31 causes the processor 24 to execute a method for sensing wavefront aberrations caused by the medium 6 in this embodiment. The non-transient computer-readable medium 29 in this embodiment may store the wavefront aberration estimation program 31.

[0044] <Method for generating wavefront aberration parameter estimation model 32> Referring to Figures 8 to 16, the method for generating the wavefront aberration parameter estimation model 32 of this embodiment will be described. Referring to Figure 8, the method for generating the wavefront aberration parameter estimation model 32 of this embodiment includes the step of generating a training dataset 62 (step S11) and the step of training the wavefront aberration parameter estimation model 67 using machine learning with the training dataset 62 (step S20).

[0045] <Step S11> Step S11 (see Figure 8) will be explained with reference to Figures 9 to 12C. In step S11, the training dataset 62 (see Figure 10) is generated using the training dataset generation device 40 (see Figure 9). Step S11 is performed by the processor 51 (see Figure 10) of the training dataset generation device 40 executing the training dataset generation program 61 (see Figure 10).

[0046] <Training Dataset Generator 40> The training dataset generation device 40 comprises a light source 41, a collimator lens 44, a spatial light modulator 45, a linear focusing element 15, a line sensor 16, a training data generator 48, and a controller 49. The training dataset generation device 40 may further include an aperture 43. The training dataset generation device 40 may further include a two-dimensional focusing element 10 and a mask 11.

[0047] The light source 41 is, for example, a laser light source or a light-emitting element such as a superluminescent diode (SLD). The light source 41 emits light 42. Referring to Figure 1, when the wavefront sensor 1 is applied to the receiver of an optical satellite communication device, the light source 3 is, for example, mounted on a satellite and is located far enough away from the Earth's surface where the wavefront sensor 1 is installed, so the light source 3 can be considered a point source. Therefore, an aperture 43 may be placed between the light source 41 and the collimator lens 44 so that the light source 41 can be considered a point source.

[0048] The collimator lens 44 is positioned between the light source 41 and the spatial light modulator 45. The collimator lens 44 may also be positioned between the aperture 43 and the spatial light modulator 45. Referring to Figure 1, when the wavefront sensor 1 is applied to the receiver of an optical satellite communication device, the light source 3 is, for example, mounted on a satellite and is sufficiently far from the Earth's surface where the wavefront sensor 1 is installed, so the light 4 emitted by the light source 3 reaches the medium 6 as a plane wave. Therefore, the collimator lens 44 collimates the light 42 emitted by the light source 41, converting the light 42 into a plane wave, so that the light 42 emitted by the light source 41 reaches the spatial light modulator 45 as a plane wave. The collimator lens 44 is not particularly limited, but may be a plano-convex lens.

[0049] The spatial light modulator 45 is positioned between the collimator lens 44 and the two-dimensional focusing element 10 on the optical path 47 of light 42 and 46. The spatial light modulator 45 is positioned on the front focal plane of the two-dimensional focusing element 10. The spatial light modulator 45 may be a transmissive spatial light modulator such as a liquid crystal spatial light modulator (LC-SLM), or a reflective spatial light modulator such as an LCOS (Liquid Crystal on Silicon) spatial light modulator or a microelectromechanical system (MEMS) spatial light modulator. The spatial light modulator 45 can be controlled by a controller 49. The spatial light modulator receives a wavefront aberration signal from the controller 49 and forms a two-dimensional phase distribution. Light 42 travels through the spatial light modulator 45 and becomes light 46. The wavefront aberration resulting from the two-dimensional phase distribution formed in the spatial light modulator 45 is the difference between the wavefront of light 42 and the wavefront of light 46.

[0050] The two-dimensional focusing element 10 (see Figure 9) of the training dataset generation device 40 is configured similarly to the two-dimensional focusing element 10 of the wavefront sensor 1 (see Figure 1). The two-dimensional focusing element 10 of the training dataset generation device 40 is positioned between the spatial light modulator 45 and the mask 11 in the optical path 8 of the light 46. The light 46 is incident on the two-dimensional focusing element 10. The two-dimensional focusing element 10 isotropically focuses the light 46 in a plane perpendicular to the optical path 47 of the light 46. The two-dimensional focusing element 10 is not particularly limited, but may be a spherical lens such as a plano-convex or biconvex lens, or it may be at least one curved mirror.

[0051] The mask 11 of the training dataset generator 40 (see Figure 9) is configured similarly to the mask 11 of the wavefront sensor 1 (see Figure 1). The mask 11 of the training dataset generator 40 is positioned on the rear focal plane of the two-dimensional light-gathering element 10 and on the front focal plane of the linear light-gathering element 15. The mask 11 of the training dataset generator 40 includes a central light-shielding region 12, a peripheral light-shielding region 13, and an annular aperture 14 formed between the central light-shielding region 12 and the peripheral light-shielding region 13. Light 46 passes through the annular aperture 14. The central light-shielding region 12 of the mask 11 blocks the image of the light source 41, enabling the line sensor 16 to detect light 46 with high accuracy.

[0052] The linear focusing element 15 (see Figure 9) of the training dataset generation device 40 is configured similarly to the linear focusing element 15 (see Figure 1) of the wavefront sensor 1. The linear focusing element 15 of the training dataset generation device 40 is positioned between the mask 11 and the line sensor 16 in the optical path 8 of the light 46. The linear focusing element 15 focuses the light 7 mainly in a direction perpendicular to the arrangement direction of multiple pixels of the line sensor 16. In this way, the linear focusing element 15 focuses the light 46 linearly. The linear focusing element 15 may be, for example, a linear focusing lens such as a cylindrical lens, or it may be at least one curved mirror.

[0053] The line sensor 16 (see Figure 9) of the training dataset generator 40 is configured similarly to the line sensor 16 of the wavefront sensor 1 (see Figure 1). The line sensor 16 is positioned on the rear focal plane of the linear focusing element 15. The line sensor 16 receives light 46 focused linearly by the linear focusing element 15 and acquires a one-dimensional image 65 (see Figure 10) of the light 46. The line sensor 16 outputs the one-dimensional image 65 of the light 46 to the wavefront aberration estimator 20. The frame rate of the line sensor 16 is, for example, 10 kHz or higher. The frame rate of the line sensor 16 may be 50 kHz or higher, or 100 kHz or higher.

[0054] The training data generator 48 generates wavefront aberration parameters 64 (see Figure 10). The training data generator 48 outputs the wavefront aberration parameters 64 to the controller 49. The wavefront aberration parameters 64 are, for example, a set of coefficients of a Zernike polynomial (see Figure 12A). The training data generator 48 also receives a one-dimensional image 65 of light 46 (see Figure 10) from the line sensor 16. Furthermore, the training data generator 48 generates training data 63 (see Figure 10), which is a pair of wavefront aberration parameters 64 and the one-dimensional image 65 corresponding to the wavefront aberration parameters 64. The training data generator 48 generates a training dataset 62 (see Figure 10) consisting of multiple training data sets 63. The multiple training data sets 63 differ from each other in terms of wavefront aberration parameters 64.

[0055] The controller 49 receives the wavefront aberration parameter 64 from the training data generator 48 and generates a wavefront aberration signal corresponding to the wavefront aberration parameter 64. The controller 49 outputs the wavefront aberration signal to the spatial light modulator 45. The controller 49 controls the spatial light modulator 45.

[0056] The controller 49 is a microcomputer that includes, for example, a processor, RAM (Random Access Memory), and a memory device such as ROM (Read Only Memory). A CPU or GPU may be used as the processor. RAM functions as working memory for temporarily storing data processed by the processor. The memory device stores, for example, a program executed by the processor. In this embodiment, the controller 49 controls the spatial light modulator 45 by having the processor execute the program stored in the memory device. Instead of a microcomputer, an FPGA may be used as the controller 49. Various processes in the controller 49 are not limited to being executed by software, but may also be implemented by dedicated hardware (electronic circuits).

[0057] <Hardware configuration of training data generator 48> Referring to Figure 10, the hardware configuration of the training data generator 48 will be described. The training data generator 48 includes an input device 50, a processor 51, memory 52, a display 53, a network controller 54, a media drive 55, and storage 57.

[0058] The input device 50 accepts various input operations. The input device 50 is, for example, a keyboard, a mouse, or a touch panel.

[0059] The display 53 displays information necessary for processing the training dataset 62 and the training data generator 48. The display 53 is, for example, an LCD or an organic EL display.

[0060] The processor 51 performs the processing necessary to realize the functions of the training data generator 48 by executing the training dataset generation program 61, etc. The processor 51 is composed of, for example, a CPU or a GPU.

[0061] Memory 52 provides a storage area for the processor 51 to temporarily store program code or work memory when executing the training dataset generation program 61, etc. Memory 52 is, for example, a volatile memory device such as DRAM or SRAM.

[0062] The network controller 54 transmits and receives programs or data to and from any device via a communication network (not shown), such as the Internet or an intranet. For example, the network controller 54 transmits the training dataset 62 via the communication network to a wavefront aberration parameter estimation model generation device 60 (see Figure 13), a display (not shown), or a storage device (not shown). The network controller 54 supports any communication method, such as Ethernet®, wireless LAN, or Bluetooth®.

[0063] The media drive 55 is a device that reads programs or data stored in the computer-readable medium 56. The media drive 55 may also be a device that writes programs or data to the computer-readable medium 56. The computer-readable medium 56 is a non-transitory storage medium that stores programs or data non-volatilely. The computer-readable medium 56 is, for example, an optical storage medium such as an optical disc (e.g., CD-ROM or DVD-ROM), a semiconductor storage medium such as flash memory or USB memory, a magnetic storage medium such as a hard disk, floppy disk or storage tape, or a magneto-optical storage medium such as an MO disk.

[0064] The storage 57 stores data such as the training dataset 62, and programs executed in the processor 51 (such as the training dataset generation program 61). The storage 57 is a non-volatile memory device such as a hard disk or SSD.

[0065] The program for implementing the functions of the training data generator 48 may be stored and distributed on a non-transient computer-readable medium 56 and installed on the storage 57. The program for implementing the functions of the training data generator 48 may be downloaded to the training data generator 48 via a communication network such as the Internet or an intranet. The program for implementing the functions of the training data generator 48 includes a training dataset generation program 61.

[0066] In this embodiment, an example is shown in which a general-purpose computer (processor 51) implements the functions of the training data generator 48 by executing a program. However, the embodiment is not limited to this, and all or part of the functions of the training data generator 48 may be implemented using an integrated circuit such as an ASIC or FPGA.

[0067] <How to generate training dataset 62> Refer to Figure 11 to explain how the training dataset 62 is generated.

[0068] The training data generator 48 generates wavefront aberration parameters 64 (step S12). The wavefront aberration parameters 64 are, for example, a set of coefficients of a Zernike polynomial (see Figure 12A). The training data generator 48 outputs the wavefront aberration parameters 64 to the controller 49 and the storage 57. The storage 57 stores the wavefront aberration parameters 64.

[0069] A two-dimensional phase distribution corresponding to the wavefront aberration parameter 64 is formed in the spatial light modulator 45 (step S13). Specifically, the controller 49 receives the wavefront aberration parameter 64 from the training data generator 48. The controller 49 generates a wavefront aberration signal corresponding to the wavefront aberration parameter 64 from the wavefront aberration parameter 64. The controller 49 outputs the wavefront aberration signal to the spatial light modulator 45. The spatial light modulator 45 receives the wavefront aberration signal from the controller 49 and forms a two-dimensional phase distribution (see Figure 12B). The wavefront aberration resulting from the two-dimensional phase distribution formed in the spatial light modulator 45 is the wavefront aberration expressed by the wavefront aberration parameter 64 generated by the training data generator 48.

[0070] A one-dimensional image 65 of the light 46 that has traveled through the spatial light modulator 45 is obtained (step S14). Specifically, the light source 41 emits light 42. Light 42 passes through the aperture 43 and collimator lens 44, travels through the spatial light modulator 45, and becomes light 46. The wavefront aberration caused by the two-dimensional phase distribution formed in the spatial light modulator 45 is the difference between the wavefront of light 42 and the wavefront of light 46. The two-dimensional light-gathering element 10 isotropically focuses the light 46 in a plane perpendicular to the optical path 47 of the light 46. The central light-shielding region 12 of the mask 11 blocks the image of the light source 41. Light 46 passes through the annular aperture 14 of the mask 11 and enters the linear light-gathering element 15. The linear light-gathering element 15 focuses the light 46 linearly. The line sensor 16 receives the light 46 focused linearly by the linear light-gathering element 15 and acquires a one-dimensional image 65 of the light 46 (see Figure 12C).

[0071] The line sensor 16 outputs a one-dimensional image 65 of the light 46 to the training data generator 48. The training data generator 48 receives the one-dimensional image 65 of the light 46 from the line sensor 16. The storage 57 of the training data generator 48 stores the one-dimensional image 65 of the light 46.

[0072] The training data generator 48 generates training data 63 (step S15). The training data generator 48 generates training data 63 by combining the wavefront aberration parameters 64 generated in step S12 and the one-dimensional image 65 acquired in step S14. The storage 57 of the training data generator 48 stores the training data 63.

[0073] The training data generator 48 determines whether the number of training data 63 has reached a predetermined number (step S16). If the training data generator 48 determines that the number of training data 63 has not reached the predetermined number, the training data generator 48 generates a different wavefront aberration parameter 64 in step S12 and repeats steps S12 to S16. If the training data generator 48 determines that the number of training data 63 has reached the predetermined number, the training data generator 48 terminates the generation of training data 63. In this way, steps S12 to S16 are repeatedly executed until the number of training data 63 reaches the predetermined number. The training data generator 48 generates a training dataset 62 consisting of multiple training data 63. The storage 57 of the training data generator 48 stores the training dataset 62.

[0074] The training dataset 62 may be generated by simulating the optical system of the training dataset generation device 40 (light source 41, aperture 43, two-dimensional light-gathering element 10, mask 11, linear light-gathering element 15, and line sensor 16) and the method of generating the training dataset 62 on a computer.

[0075] <Step S20> Step S20 (see Figure 8) will be explained with reference to Figures 13 to 16. In step S20, the wavefront aberration parameter estimation model 67 is trained by machine learning using the wavefront aberration parameter estimation model generation device 60 (see Figure 13).

[0076] <Hardware configuration of wavefront aberration parameter estimation model generation device 60> Referring to Figure 13, the hardware configuration of the wavefront aberration parameter estimation model generation device 60 will be described. The wavefront aberration parameter estimation model generation device 60 includes an input device 50b, a processor 51b, memory 52b, a display 53b, a network controller 54b, a media drive 55b, and storage 57b.

[0077] The input device 50b accepts various input operations. The input device 50b is, for example, a keyboard, a mouse, or a touch panel.

[0078] The display 53b displays information necessary for processing in the wavefront aberration parameter estimation model generation device 60. The display 53b is, for example, an LCD or an organic EL display.

[0079] The processor 51b executes the processing necessary to realize the functions of the wavefront aberration parameter estimation model generation device 60 by running the wavefront aberration parameter estimation model generation program 66. The processor 51b is composed of, for example, a CPU or a GPU.

[0080] Memory 52b provides a storage area for temporarily storing program code or work memory when the processor 51b executes the wavefront aberration parameter estimation model generation program 66. Memory 52b is, for example, a volatile memory device such as DRAM or SRAM.

[0081] The network controller 54b receives, for example, a training dataset 62 (see Figure 10) from the training data generator 48 (see Figures 9 and 10) via the communication network. The network controller 54b transmits, for example, a wavefront aberration parameter estimation model 32 via the communication network to the wavefront aberration estimator 20 (see Figure 1), a display (not shown), or a storage device (not shown). The network controller 54b supports any communication method, such as Ethernet®, Wi-Fi, or Bluetooth®.

[0082] The media drive 55b is a device for reading programs or data stored on the computer-readable medium 56b. The media drive 55b may also be a device for writing programs or data to the computer-readable medium 56b. The computer-readable medium 56b is a non-transitory storage medium that stores programs or data non-volatilely. The computer-readable medium 56b may be, for example, an optical storage medium such as an optical disc (e.g., CD-ROM or DVD-ROM), a semiconductor storage medium such as flash memory or USB memory, a magnetic storage medium such as a hard disk, floppy disk or storage tape, or a magneto-optical storage medium such as an MO disk.

[0083] Storage 57b stores, for example, a training dataset 62, wavefront aberration parameter estimation models 32 and 67, and a wavefront aberration parameter estimation model generation program 66. Storage 57b is, for example, a non-volatile memory device such as a hard disk or SSD.

[0084] The program for realizing the functions of the wavefront aberration parameter estimation model generation device 60 may be stored and distributed on a non-transient computer-readable medium 56b and installed on storage 57b. The program for realizing the functions of the wavefront aberration parameter estimation model generation device 60 may be downloaded to the wavefront aberration parameter estimation model generation device 60 via a communication network such as the Internet or an intranet. The program for realizing the functions of the wavefront aberration parameter estimation model generation device 60 includes a wavefront aberration parameter estimation model generation program 66.

[0085] In this embodiment, an example is shown in which a general-purpose computer (processor 51) implements the functions of the wavefront aberration parameter estimation model generation device 60 by executing a program. However, the embodiment is not limited to this, and all or part of the functions of the wavefront aberration parameter estimation model generation device 60 may be implemented using an integrated circuit such as an ASIC or FPGA.

[0086] <Functional configuration of the wavefront aberration parameter estimation model generation device 60> Referring to Figures 14 and 15, the functional configuration of the wavefront aberration parameter estimation model generation device 60 will be described. The wavefront aberration parameter estimation model generation device 60 includes an image preprocessing unit 71 and a machine learning unit 72.

[0087] Referring to Figure 14, the image preprocessing unit 71 preprocesses the one-dimensional image 65 of the training data 63 to generate a preprocessed training image 65b. For example, the image preprocessing unit 71 applies preprocessing to the one-dimensional image 65 of the training data 63, such as white balance correction, contrast correction, resizing, normalization, noise reduction, or filtering. If the wavefront aberration parameter estimation models 32,67 include a convolutional neural network (CNN), the image preprocessing unit 71 may convert the one-dimensional image 65 of the training data 63 into a two-dimensional image and generate the two-dimensional image as the preprocessed training image 65b. For example, the one-dimensional image 65 of the training data 63 is converted into a two-dimensional image by rearranging (reshaping) the pixels of the one-dimensional image 65 in two dimensions. The image preprocessing unit 71 outputs the preprocessed training image 65b to the machine learning unit 72.

[0088] Referring to Figures 14 and 15, the machine learning unit 72 generates a wavefront aberration parameter estimation model 32 by training a wavefront aberration parameter estimation model 67 using machine learning with the preprocessed training image 65b and the wavefront aberration parameters 64 of the training data 63 corresponding to the preprocessed training image 65b.

[0089] The machine learning unit 72 includes a wavefront aberration parameter estimation model 67 and a parameter optimization module 73. The wavefront aberration parameter estimation model 67 includes a neural network 67N and parameters 67P. The neural network 67N is a neural network classified as a deep neural network (DNN). The neural network 67N may also include a convolutional neural network (CNN). The neural network 67N is the same as the neural network 32N. The neural network 67N is pre-built and stored in the storage 57 of the wavefront aberration parameter estimation model generation device 60. The parameter optimization module 73 is a program module for optimizing parameters 67P. The machine learning unit 72 updates the values ​​of parameters 67P of the wavefront aberration parameter estimation model 67 by machine learning using pre-processed training images 65b and wavefront aberration parameters 64.

[0090] Specifically, the machine learning unit 72 inputs the preprocessed training image 65b into the wavefront aberration parameter estimation model 67 and outputs the wavefront aberration parameter 64b to the wavefront aberration parameter estimation model 67. The parameter optimization module 73 calculates the error between the wavefront aberration parameter 64b output by the wavefront aberration parameter estimation model 67 and the wavefront aberration parameter 64 of the training data 63. The parameter optimization module 73 optimizes the parameter 67P of the wavefront aberration parameter estimation model 67 so that this error is minimized. Any optimization algorithm can be used to optimize the parameter 67P. For example, gradient methods such as SGD (Stochastic Gradient Descent), Momentum SGD (SGD with inertia term), AdaGrad, RMSprop, AdaDelta, or Adam (Adaptive moment estimation) can be used as optimization algorithms.

[0091] Similarly, the machine learning unit 72 iteratively optimizes the parameter 67P of the wavefront aberration parameter estimation model 67 using multiple training data 63 included in the training dataset 62. In this way, the machine learning unit 72 generates a trained wavefront aberration parameter estimation model 67, including the optimized parameter 67P, as the wavefront aberration parameter estimation model 32. The optimized parameter 67P is parameter 32P (see Figure 3). The machine learning unit 72 outputs the wavefront aberration parameter estimation model 32 to the storage 57 of the wavefront aberration parameter estimation model generation device 60.

[0092] <Training method for wavefront aberration parameter estimation model 32> Referring to Figure 16, the training method for the wavefront aberration parameter estimation model 32 using machine learning with the wavefront aberration parameter estimation model generation device 60 will be explained.

[0093] The image preprocessing unit 71 reads the training data 63 from the storage 57 of the wavefront aberration parameter estimation model generation device 60. The image preprocessing unit 71 preprocesses the one-dimensional image 65 of the training data 63 to generate a preprocessed training image 65b (step S21). For example, the image preprocessing unit 71 applies preprocessing to the one-dimensional image 65 of the training data 63, such as white balance correction, contrast correction, resizing, normalization, noise reduction, or filtering. The image preprocessing unit 71 outputs the preprocessed training image 65b to the machine learning unit 72.

[0094] If the wavefront aberration parameter estimation models 32 and 67 include a convolutional neural network (CNN), the image preprocessing unit 71 may convert the one-dimensional image 65 of the training data 63 into a two-dimensional image and generate the two-dimensional image as the preprocessed training image 65b. For example, the one-dimensional image 65 of the training data 63 may be converted into a two-dimensional image by rearranging (reshaping) the pixels of the one-dimensional image 33 in a two-dimensional manner.

[0095] The machine learning unit 72 receives the preprocessed training image 65b from the image preprocessing unit 71. The machine learning unit 72 inputs the preprocessed training image 65b into the wavefront aberration parameter estimation model 67 and causes the wavefront aberration parameter 64b to be output by the wavefront aberration parameter estimation model 67 (step S22). The parameter optimization module 73 of the machine learning unit 72 calculates the error between the wavefront aberration parameter 64b output by the wavefront aberration parameter estimation model 67 and the wavefront aberration parameter 64 of the training data 63. The parameter optimization module 73 optimizes the parameter 67P of the wavefront aberration parameter estimation model 67 so that this error is minimized (step S23).

[0096] Similarly, the machine learning unit 72 iteratively optimizes the parameter 67P of the wavefront aberration parameter estimation model 67 using multiple training data 63 included in the training dataset 62. In this way, the machine learning unit 72 generates a trained wavefront aberration parameter estimation model 67, including the optimized parameter 67P, as the wavefront aberration parameter estimation model 32. The optimized parameter 67P is parameter 32P (see Figure 3). The machine learning unit 72 outputs the wavefront aberration parameter estimation model 32 to the storage 57 of the wavefront aberration parameter estimation model generation device 60. The storage 57 of the wavefront aberration parameter estimation model generation device 60 stores the wavefront aberration parameter estimation model 32.

[0097] Step S20 is performed by the processor 51 (see Figure 10) of the wavefront aberration parameter estimation model generation device 60 executing the wavefront aberration parameter estimation model generation program 66 (see Figure 10).

[0098] <Adaptive optics device 85> The adaptive optics apparatus 85 of this embodiment will be described with reference to Figure 17. The adaptive optics apparatus 85 can be applied, for example, to a receiving device 82 of an optical satellite communication device 80.

[0099] The optical satellite communication device 80 includes a transmitter 81 and a receiver 82. The transmitter 81 is, for example, mounted on a satellite. The transmitter 81 includes a light source 3. The light source 3 is, for example, mounted on a satellite. The light source 3 is, for example, a laser light source or a superluminescent diode (SLD). The receiver 82 is, for example, installed on the Earth's surface. The receiver 82 includes an imaging lens 83 and a photodetector 84. The photodetector 84 is, for example, an image sensor.

[0100] Light source 3 emits light 4. Light 4 travels through medium 6. Light 4 is affected by aberrations in medium 6 and becomes light 7 with a distorted wavefront. The wavefront aberration caused by medium 6 is the difference between the wavefront of light 4 and the wavefront of light 7. Imaging lens 83 images light 7 onto photodetector 84. Photodetector 84 receives light 7.

[0101] To eliminate the effects of aberrations in the medium 6, the receiving device 82 further includes an adaptive optics device 85. The adaptive optics device 85 includes a beam splitter 87, a wavefront sensor 1, a spatial light modulator 86, and a controller 88.

[0102] The beam splitter 87 splits the light 7 into a first beam that goes towards the photodetector 84 and a second beam that goes towards the wavefront sensor 1. The first beam is incident on the photodetector 84. The second beam is incident on the wavefront sensor 1. The wavefront sensor 1 acquires a one-dimensional image 33 of the light 7 (see Figure 3). From the one-dimensional image 33, the wavefront sensor 1 estimates the wavefront aberration caused by the medium 6. Specifically, the wavefront sensor 1 outputs a wavefront aberration parameter 34 (see Figure 3) that represents the wavefront aberration caused by the medium 6. The wavefront sensor 1 outputs the wavefront aberration parameter 34 to the controller 88.

[0103] The controller 88 is communicated to the spatial light modulator 86 and the wavefront sensor 1. The controller 88 receives the wavefront aberration parameter 34 (see Figure 3) from the wavefront sensor 1 and generates a compensated wavefront aberration signal corresponding to the wavefront aberration parameter 34. The controller 88 outputs the compensated wavefront aberration signal to the spatial light modulator 86.

[0104] The controller 88 is a microcomputer that includes, for example, a processor, RAM, and a memory device such as ROM. A CPU or GPU may be used as the processor. RAM functions as working memory for temporarily storing data processed by the processor. The memory device stores, for example, a program executed by the processor. In this embodiment, the controller 88 controls the spatial light modulator 86 by having the processor execute the program stored in the memory device. Instead of a microcomputer, an FPGA may be used as the controller 88. Various processes in the controller 88 are not limited to being performed by software, but may also be implemented by dedicated hardware (electronic circuits).

[0105] The spatial light modulator 86 is positioned on the optical path 8 of the light 7, on the incident side of the light 7 relative to the beam splitter 87. The spatial light modulator 86 is not particularly limited, but examples include an LCOS spatial light modulator, a digital micromirror device (DMD) including a plurality of MEMS mirrors arranged in two dimensions, or a deformable mirror utilizing a piezo actuator. The spatial light modulator 86 can be controlled by a controller 88. The spatial light modulator 86 receives a compensated wavefront aberration signal from the controller 88 and forms a two-dimensional phase distribution. The spatial light modulator 86 imparts the compensated wavefront aberration resulting from the two-dimensional phase distribution to the light 7. The compensated wavefront aberration cancels out the wavefront aberration caused by the medium 6. Therefore, the effect of the aberration of the medium 6 can be removed from the output signal of the photodetector 84.

[0106] The receiving device 82 of the optical satellite communication device 80, which includes the adaptive optics system 85, may be mounted on a satellite, and the transmitting device 81 of the optical satellite communication device, which includes the light source 3, may be installed on the Earth's surface.

[0107] The effects of the wavefront sensor 1 and adaptive optics device 85 of this embodiment will be explained. The wavefront sensor 1 of this embodiment comprises a linear focusing element 15, a line sensor 16, and a wavefront aberration estimator 20. The linear focusing element 15 focuses light 7 that has traveled through the medium 6 in a linear fashion. The line sensor 16 receives the light 7 focused linearly by the linear focusing element 15 and acquires a one-dimensional image 33 of the light 7. The wavefront aberration estimator 20 includes an image preprocessing unit 37 and a wavefront aberration parameter estimation unit 38. The image preprocessing unit 37 preprocesses the one-dimensional image 33 to generate a preprocessed image 33b. The wavefront aberration parameter estimation unit 38 inputs the preprocessed image 33b to a wavefront aberration parameter estimation model 32 trained by machine learning and causes the wavefront aberration parameter 34 representing the wavefront aberration caused by the medium 6 to output.

[0108] In comparative examples such as wavefront sensors using diffraction imaging, iterative calculations are required to obtain the amplitude and phase of light from a two-dimensional intensity image. In contrast, the wavefront sensor 1 of this embodiment uses a wavefront aberration parameter estimation model 32 trained by machine learning to obtain wavefront aberration parameters 34 that represent wavefront aberration caused by the medium 6. Therefore, the wavefront sensor 1 of this embodiment can obtain wavefront aberration parameters 34 without performing iterative calculations. The wavefront sensor 1 of this embodiment also comprises a line sensor 16 and a linear focusing element 15. The upper limit of the frame rate specified by the line sensor 16 is higher than the upper limit of the frame rate specified by the two-dimensional image sensor. Furthermore, the linear focusing element 15 can increase the light intensity per pixel of the line sensor 16 across the entire range of pixels of the line sensor 16 more effectively than a spherical lens. Therefore, the line sensor 16 can acquire a one-dimensional image 33 of light 7 at a higher frame rate. The wavefront sensor 1 can sense wavefront aberration caused by the medium 6 at a faster speed. For example, by applying wavefront sensor 1 to an optical satellite communication device, atmospheric fluctuations above 10 kHz can be prevented from adversely affecting communication quality. Wavefront sensor 1 enables terabits per second (Tbps) optical satellite communication.

[0109] In the wavefront sensor 1 of this embodiment, the frame rate of the line sensor 16 is 100 kHz or higher.

[0110] Therefore, the wavefront sensor 1 can sense wavefront aberrations caused by the medium 6 at a faster speed.

[0111] The wavefront sensor 1 of this embodiment further comprises a two-dimensional light-gathering element 10 and a mask 11. The linear light-gathering element 15 is positioned between the mask 11 and the line sensor 16 in the optical path 8 of the light 7. The mask 11 is positioned between the two-dimensional light-gathering element 10 and the linear light-gathering element 15 and on the focal plane of the two-dimensional light-gathering element 10 in the optical path 8 of the light 7. The mask 11 includes a central light-shielding region 12, a peripheral light-shielding region 13, and an annular aperture 14 formed between the central light-shielding region 12 and the peripheral light-shielding region 13.

[0112] Therefore, the central light-shielding region 12 blocks the image of the light source 3 of the light 7. The line sensor 16 detects the light 7, which has been affected by the aberrations of the medium 6, with high precision. The wavefront sensor 1 can sense the wavefront aberrations caused by the medium 6 with even higher precision.

[0113] In the wavefront sensor 1 of this embodiment, the image preprocessing unit 37 generates a two-dimensional image as a preprocessed image 33b from the one-dimensional image 33 by rearranging the pixels of the one-dimensional image 33. The wavefront aberration parameter estimation model 32 includes a convolutional neural network. The wavefront aberration parameter estimation unit 38 inputs the two-dimensional image to the wavefront aberration parameter estimation model 32 and outputs the wavefront aberration parameters 34 to the wavefront aberration parameter estimation model 32.

[0114] The wavefront aberration caused by the medium 6 is a two-dimensionally distributed aberration. Therefore, by converting the one-dimensional image 33 to a two-dimensional image and inputting the two-dimensional image into a wavefront aberration parameter estimation model 32 that includes a convolutional neural network, the wavefront aberration parameters 34 can be obtained more quickly. The wavefront sensor 1 can sense the wavefront aberration caused by the medium 6 more quickly.

[0115] In the wavefront sensor 1 of this embodiment, light 7 is emitted from a laser light source or a superluminescent diode.

[0116] Therefore, light 7 has higher coherence. Due to the coherence of light 7, speckles are formed on the line sensor 16. The speckles improve the contrast of light 7 on the line sensor 16. Therefore, the line sensor 16 can acquire a one-dimensional image 33 of light 7 at an even higher frame rate. The wavefront sensor 1 can sense wavefront aberrations caused by the medium 6 at a faster speed.

[0117] The adaptive optics apparatus 85 of this embodiment comprises a wavefront sensor 1, a spatial light modulator 86, and a controller 88. The controller 88 is communicatively connected to the spatial light modulator 86 and the wavefront sensor 1, and can control the spatial light modulator 86. The controller 88 receives wavefront aberration parameters 34 from the wavefront sensor 1 that represent wavefront aberration caused by the medium 6, and causes the spatial light modulator 86 to form a two-dimensional phase distribution. The compensated wavefront aberration caused by the two-dimensional phase distribution cancels out the wavefront aberration caused by the medium 6.

[0118] Because the adaptive optics system 85 is equipped with a wavefront sensor 1, it can compensate for wavefront aberrations caused by the medium 6 at a faster speed. For example, by applying the adaptive optics system 85 to the receiving device 82 of the optical satellite communication device 80, atmospheric fluctuations of 10 kHz or higher can be prevented from adversely affecting communication quality. The adaptive optics system 85 enables terabits per second (Tbps) optical satellite communication.

[0119] (Embodiment 2) <Wavefront Sensor 1> Referring to Figure 18, the wavefront sensor 1 of Embodiment 2 will be described. The wavefront sensor 1 of this embodiment has the same configuration as the wavefront sensor 1 of Embodiment 1, but differs from the wavefront sensor 1 of Embodiment 1 in that it further includes a light scattering plate 17.

[0120] The light scattering plate 17 is positioned between the linear focusing element 15 and the line sensor 16 in the optical path 8 of the light 7. The light scattering plate 17 scatters the light 7. The line sensor 16 acquires a one-dimensional image 33 (see Figure 2) of the light 7 scattered by the light scattering plate 17. Even if the wavefront aberration caused by the medium 6 is small, the photodetector 84 can acquire a one-dimensional image 33 in which the wavefront aberration caused by the medium 6 is amplified by the light scattering plate 17. Therefore, the wavefront sensor 1 can sense the wavefront aberration caused by the medium 6 at a faster speed and with higher accuracy.

[0121] Referring to Figure 19, the training dataset generation device 40 of this embodiment is configured similarly to the training dataset generation device 40 of Embodiment 1, but differs from the training dataset generation device 40 of Embodiment 1 in that it further includes a light scattering plate 17. The light scattering plate 17 of the training dataset generation device 40 of this embodiment is positioned between the linear light concentrator 15 and the line sensor 16, just like the light scattering plate 17 of the wavefront sensor 1 of this embodiment.

[0122] Referring to Figure 11, the method for generating the training dataset 62 in this embodiment is the same as the method for generating the training dataset 62 in Embodiment 1. However, the method for generating the training dataset 62 in this embodiment differs from the method for generating the training dataset 62 in Embodiment 1 in that it uses the training dataset generation device 40 of this embodiment (see Figure 19) to generate the training dataset 62. Alternatively, the training dataset 62 may be generated by simulating the optical system of the training dataset generation device 40 of this embodiment (light source 41, aperture 43, two-dimensional light-gathering element 10, mask 11, linear light-gathering element 15, light-scattering plate 17, and line sensor 16) and the method for generating the training dataset 62 of this embodiment on a computer.

[0123] Referring to Figures 13 to 16, the training method for the wavefront aberration parameter estimation model 32 and the wavefront aberration parameter estimation model generation apparatus 60 of this embodiment are the same as the training method for the wavefront aberration parameter estimation model 32 and the wavefront aberration parameter estimation model generation apparatus 60 of Embodiment 1.

[0124] <Adaptive optics device 85> The adaptive optics apparatus 85 of this embodiment will be described with reference to Figure 17. The adaptive optics apparatus 85 of this embodiment has the same configuration as the adaptive optics apparatus 85 of Embodiment 1, but instead of the wavefront sensor 1 of Embodiment 1, it is equipped with the wavefront sensor 1 of this embodiment (see Figure 18). Therefore, the adaptive optics apparatus 85 can compensate for wavefront aberrations caused by the medium 6 at a faster speed and with higher accuracy.

[0125] The various aspects of this disclosure are summarized below as an appendix. (Note 1) A linear focusing element that focuses light traveling through a medium into a linear shape, Line sensor and, It is equipped with a wavefront aberration estimator, The line sensor receives the light that has been linearly focused by the linear light-gathering element and acquires a one-dimensional image of the light. The wavefront aberration estimator includes an image preprocessing unit and a wavefront aberration parameter estimation unit. The image preprocessing unit preprocesses the one-dimensional image to generate a preprocessed image. The wavefront sensor includes a wavefront aberration parameter estimation unit that inputs the preprocessed image into a wavefront aberration parameter estimation model trained by machine learning, and causes the wavefront aberration parameter that represents the wavefront aberration caused by the medium to be output by the wavefront aberration parameter estimation model. (Note 2) The wavefront sensor described in Appendix 1 has a frame rate of 10 kHz or higher. (Note 3) A two-dimensional light-gathering element, We also have masks, The linear light-gathering element is positioned between the mask and the line sensor in the optical path of the light. The mask is positioned in the optical path of the light between the two-dimensional focusing element and the linear focusing element and on the focal plane of the two-dimensional focusing element. The wavefront sensor according to Appendix 1 or Appendix 2, wherein the mask includes a central light-shielding region, a peripheral light-shielding region, and an annular opening formed between the central light-shielding region and the peripheral light-shielding region. (Note 4) The image preprocessing unit generates a two-dimensional image as the preprocessed image from the one-dimensional image by rearranging the pixels of the one-dimensional image. The wavefront aberration parameter estimation model includes a convolutional neural network, The wavefront sensor described in any of Appendix 1 to Appendix 3, wherein the wavefront aberration parameter estimation unit inputs the two-dimensional image to the wavefront aberration parameter estimation model and outputs the wavefront aberration parameters to the wavefront aberration parameter estimation model. (Note 5) The aforementioned light is emitted from a laser light source or a superluminescent diode, as described in any of Appendix 1 to Appendix 4 of the wavefront sensor. (Note 6) It is further equipped with a light scattering plate, The light scattering plate is a wavefront sensor according to any one of the appendices 1 to 5, disposed between the linear light concentrating element and the line sensor. (Note 7) The wavefront sensor described in any of Appendix 1 to Appendix 6, A spatial light modulator, The spatial light modulator and the wavefront sensor are communicateable to each other, and the system includes a controller capable of controlling the spatial light modulator. The controller receives the wavefront aberration parameter from the wavefront sensor, which represents the wavefront aberration caused by the medium, and causes the spatial light modulator to form a two-dimensional phase distribution. The adaptive optics device cancels out the wavefront aberration caused by the medium, which is caused by the compensated wavefront aberration caused by the two-dimensional phase distribution.

[0126] Embodiments 1 and 2 disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of this disclosure is indicated by the claims rather than the foregoing description and is intended to include all modifications in the sense and scope equivalent to the claims. [Explanation of Symbols]

[0127] 1 Wavefront sensor, 3,41 Light source, 4,7,42,46 Light, 6 Medium, 8,47 Optical path, 10 Two-dimensional focusing element, 11 Mask, 12 Central light-shielding region, 13 Peripheral light-shielding region, 14 Annular aperture, 15 Linear focusing element, 16 Line sensor, 17 Light scattering plate, 20 Wavefront aberration estimator, 23,50,50b Input device, 24,51,51b Processor, 25,52,52b Memory, 26,53,53b Display, 27,54,54b Network controller, 28,55,55b Media drive, 29,56,56b Computer-readable media, 30,57,57b Storage, 31 Wavefront aberration estimation program, 32,67 Wavefront aberration parameter estimation model, 32N Neural network, 32P Parameters, 33,65 One-dimensional image, 33b Preprocessed image, 34 Wavefront aberration parameters, 35 Image receiver, 36 Wavefront aberration estimation unit, 37 Image preprocessing unit, 38 Wavefront aberration parameter estimation unit, 40 Training dataset generator, 43 Aperture, 44 Collimator lens, 45 Spatial light modulator, 48 Training data generator, 49,88 Controller, 60 Wavefront aberration parameter estimation model generator, 61 Training dataset generation program, 62 Training dataset, 63 Training data, 64,64b Wavefront aberration parameters, 65b Preprocessed training image, 66 Wavefront aberration parameter estimation model generation program, 67N Neural network, 67P Parameters, 71 Image preprocessing unit, 72 Machine learning unit, 73 Parameter optimization module, 80 Optical satellite communication device, 81 Transmitter, 82 Receiver, 83 Imaging lens, 84 Photodetector, 85 Adaptive optics system, 86 spatial light modulator, 87 beam splitter.

Claims

1. A linear focusing element that focuses light traveling through a medium into a linear shape, Line sensor and, It is equipped with a wavefront aberration estimator, The line sensor receives the light that has been linearly focused by the linear light-gathering element and acquires a one-dimensional image of the light. The wavefront aberration estimator includes an image preprocessing unit and a wavefront aberration parameter estimation unit. The image preprocessing unit preprocesses the one-dimensional image to generate a preprocessed image. The wavefront sensor includes a wavefront aberration parameter estimation unit that inputs the preprocessed image into a wavefront aberration parameter estimation model trained by machine learning, and causes the wavefront aberration parameter that represents the wavefront aberration caused by the medium to be output by the wavefront aberration parameter estimation model.

2. The wavefront sensor according to claim 1, wherein the frame rate of the line sensor is 10 kHz or higher.

3. A two-dimensional light-gathering element, We also have masks, The linear light-gathering element is positioned between the mask and the line sensor in the optical path of the light. The mask is positioned in the optical path of the light between the two-dimensional focusing element and the linear focusing element and on the focal plane of the two-dimensional focusing element. The wavefront sensor according to claim 1, wherein the mask includes a central light-shielding region, a peripheral light-shielding region, and an annular opening formed between the central light-shielding region and the peripheral light-shielding region.

4. The image preprocessing unit generates a two-dimensional image as the preprocessed image from the one-dimensional image by rearranging the pixels of the one-dimensional image. The wavefront aberration parameter estimation model includes a convolutional neural network, The wavefront sensor according to claim 1, wherein the wavefront aberration parameter estimation unit inputs the two-dimensional image to the wavefront aberration parameter estimation model and outputs the wavefront aberration parameters to the wavefront aberration parameter estimation model.

5. The wavefront sensor according to claim 1, wherein the light is emitted from a laser light source or a superluminescent diode.

6. It is further equipped with a light scattering plate, The wavefront sensor according to claim 1, wherein the light scattering plate is disposed between the linear light concentrating element and the line sensor.

7. The wavefront sensor according to any one of claims 1 to 6, A spatial light modulator, The spatial light modulator and the wavefront sensor are communicateable to each other, and the system includes a controller capable of controlling the spatial light modulator. The controller receives the wavefront aberration parameter from the wavefront sensor, which represents the wavefront aberration caused by the medium, and causes the spatial light modulator to form a two-dimensional phase distribution. The adaptive optics device cancels out the wavefront aberration caused by the medium, which is caused by the compensated wavefront aberration caused by the two-dimensional phase distribution.

Images