An orbital angular momentum superposition state phase measurement system and method based on a diffractive optical neural network

Abstract

The application discloses an orbital angular momentum superposition state phase measurement system and method based on a diffractive optical neural network. The system comprises an optical field modulation unit, a multi-layer diffractive optical structure and a spatially distributed detection unit. The optical field modulation unit modulates the phase of the input optical field to generate an orbital angular momentum superposition state. The multi-layer diffractive optical structure comprises at least two phase modulation layers. The phase modulation layers are connected through optical field diffraction propagation. The phase distribution of the phase modulation layer is optimized and trained to extract the rotational symmetry feature of the input optical field and map it to the spatial angle of the output plane. The spatially distributed detection unit is arranged on the output plane to simultaneously collect the light intensity information of multiple spatial positions. The system obtains the phase information of the measured superposition state by identifying the spatial angle position of the light intensity peak. The application adopts a non-scanning and non-interference architecture, completes the calculation in the optical domain, realizes fast and real-time measurement, and has continuous phase recognition capability.

Description

Technical Field

[0001] This invention relates to the field of quantum optical measurement and intelligent optical computing, and in particular to a system and method for measuring the phase of orbital angular momentum superposition based on a diffraction optical neural network. Background Technology

[0002] Orbital angular momentum (OAM) is an important degree of freedom in the optical field. Beams carrying OAM exhibit a helical phase wavefront structure and have broad application prospects in quantum information processing, high-dimensional optical communication, and optical micromanipulation. Particularly in quantum communication, OAM superposition states can serve as encoding carriers for high-dimensional qubits, and accurate measurement of their phase information is crucial for applications such as quantum state tomography and quantum key distribution. With the rapid development of quantum technology, higher demands are being placed on the speed, accuracy, and real-time performance of OAM superposition state phase measurement.

[0003] Chinese patent CN120369108A ​​discloses a method for measuring the orbital angular momentum spectrum of a light beam based on a diffraction neural network. By constructing a multi-layer diffraction neural network model and utilizing the diffraction propagation characteristics of the light field, photons of different orbital angular momentum states contained in the beam under test are mapped to different spatial positions on the target surface of an array detector, thereby achieving the identification of OAM modes. This scheme can obtain the orbital angular momentum spectrum of the beam under test by analyzing the position and intensity information of bright spots on the detector target surface. However, this method mainly focuses on the pattern recognition of a single OAM eigenstate and does not address the measurement of the phase parameters of OAM superposition states, thus failing to meet the requirement for accurate measurement of the relative phase of superposition states in quantum information processing.

[0004] Chinese patent CN113340418B discloses a method and system for measuring the orbital angular momentum spectrum of a light beam based on a convolutional neural network. The method involves incident the light beam to be measured along the optical axis of the system. After diffraction by a specially designed pure-phase orbital angular momentum spectrum measurement grating, a planar array detector receives the far-field diffracted light field distribution. The diffraction image is then input into a trained convolutional neural network for processing to obtain the orbital angular momentum spectrum of the light beam. While this method offers high measurement accuracy, it requires converting the light field information into an electrical signal before processing it through an electronic neural network. This results in delays in photoelectric conversion and data processing, and the computational speed is limited by the processing power of electronic devices, making real-time measurement difficult. Furthermore, the system consumes relatively high power.

[0005] In summary, existing OAM measurement techniques have the following shortcomings in superposition phase measurement: First, most schemes are designed for the identification of a single OAM pattern, and their ability to measure the relative phase of superposition states is limited; second, existing technologies often lack effective interpolation inference capabilities for phase values ​​not appearing in the training set, and the resolution of phase measurement is limited by the training sample density. Therefore, a new technological solution is needed that can achieve fast and accurate measurement of OAM superposition phase and has continuous phase recognition capabilities. Summary of the Invention

[0006] The purpose of this invention is to provide a system, method, and training method for measuring the phase of orbital angular momentum superposition state based on diffractive optical neural networks, in order to solve the problems of slow measurement speed, complex system structure, reliance on high-stability interferometers, inability to process in real time, and limited phase resolution in the prior art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: an orbital angular momentum superposition state phase measurement system based on diffractive optical neural network, comprising a light field modulation unit for generating the light field to be measured and a detection unit for detecting the light field, and further comprising: a multilayer diffractive optical structure, comprising at least two phase modulation layers, wherein each phase modulation layer is connected by light field diffraction propagation, and the phase distribution of the phase modulation layer is optimized and trained to extract the rotational symmetry features of the input light field and map them to the spatial angle of the output plane;

[0008] The detection unit is a spatially distributed detection unit, which is set on the output plane and is used to simultaneously collect light intensity information from multiple spatial locations;

[0009] The system obtains the phase information of the superposition state under test by identifying the spatial angular position of the light intensity peak.

[0010] Preferably, the system further includes a superposition state preparation module, the superposition state preparation module comprising:

[0011] Lasers are used to provide coherent light sources;

[0012] A beam expander system, optically connected to the laser, is used to adjust the beam size;

[0013] A polarization beam splitter, optically connected to the beam expander system, is used to filter polarization states;

[0014] A spatial light modulator, optically connected to the polarization beam splitter, constitutes the light field modulation unit, used to load a hologram to generate an orbital angular momentum superposition state;

[0015] An attenuator, optically connected to the spatial light modulator, is used to adjust the light intensity.

[0016] Preferably, it further includes an interpolation inference unit, which is used to calculate the energy ratio of adjacent detection areas by analyzing the light intensity distribution of adjacent spatial locations, and to calculate continuous phase values ​​that do not appear in the training set.

[0017] This invention also provides a method for measuring the phase of a superposition state of orbital angular momentum based on a diffractive optical neural network. The method, applied to the system, includes the step of generating the optical field of the superposition state of orbital angular momentum to be measured, comprising the following steps:

[0018] The light field to be measured is input into a multilayer diffractive optical structure, which includes at least two phase modulation layers, and propagates forward through diffraction propagation and phase modulation of the light field.

[0019] Acquire spatial distribution information of light intensity on the output plane;

[0020] Identify the spatial angular location of the light intensity peak and use this angle as the measured value of the superposition phase.

[0021] Preferably, the method further includes an interpolation inference step: for phase values ​​that do not appear in the training set, the energy ratio of adjacent detection areas is calculated by analyzing the light intensity distribution of adjacent spatial locations, and continuous phase values ​​are inferred.

[0022] Preferably, the orbital angular momentum superposition state is a superposition of orbital angular momentum eigenstates with topological charges of l1=1 and l2=2, respectively.

[0023] Preferably, the method is applied to quantum information processing, high-dimensional optical communication, or quantum state tomography.

[0024] The present invention also provides a method for training a diffractive optical neural network, including the steps of constructing a training dataset and adjusting network parameters through an optimization algorithm, the method being used to train a multilayer diffractive optical structure in the system;

[0025] The training dataset contains multiple orbital angular momentum superposition state samples with known phases;

[0026] The optimization algorithm adjusts the phase distribution of each phase modulation layer so that the output light intensity forms a peak at the target spatial position, which corresponds to the phase parameters of the input sample.

[0027] After training, a mapping relationship from phase parameters to spatial angles is established.

[0028] The beneficial effects of this invention are:

[0029] 1. The diffractive optical neural network architecture provided by the system of the present invention performs feature extraction and angle mapping on the input light field through at least two optimized and trained phase modulation layers, so that phase measurement can be completed in a single exposure. The input light field is propagated forward by diffraction and phase modulation through a multi-layer diffractive optical structure, forming an angle focusing distribution that corresponds one-to-one with the input phase at a specific circumferential position on the output plane. The photodetector array can acquire the complete spatial distribution of light intensity in a single acquisition, and the phase value can be directly calculated from the azimuth angle of the light intensity peak, without the need for a complex phase scanning process or a high-stability interferometer. The system structure is simplified and the robustness is significantly enhanced.

[0030] 2. The multi-layer diffractive optical structure provided by the system of the present invention solidifies the phase and angle mapping relationship learned in the training stage into the physical structure of the phase modulation layer. In the recognition stage, the light field passes through the network at the speed of light without the need for back-end digital signal processing. The correspondence between the peak position of light intensity and the phase value is directly determined by optical propagation, realizing low-latency real-time measurement from light field input to measurement result output, and significantly reducing system power consumption.

[0031] 3. The multilayer diffractive optical structure provided by this invention establishes a continuous phase-to-spatial angle mapping relationship through training. Combined with the spatial distributed detection unit to collect the light intensity distribution of the output plane, for continuous phase values ​​that do not appear in the training set, the system output light intensity forms a transitional distribution between the target spatial positions corresponding to adjacent training phases. The interpolation inference unit can accurately infer continuous phase values ​​by analyzing the light intensity distribution of adjacent detection areas and calculating the energy ratio, thereby reducing inference errors.

[0032] 4. This invention achieves 100% accuracy in identifying trained discrete phase points, with a normalized peak light intensity greater than 0.9. The optimized phase modulation layer efficiently extracts and maps the rotational symmetry features of the input light field to a specific spatial angular position on the output plane, resulting in highly concentrated energy in the target sector, low crosstalk between adjacent sectors, and high light intensity contrast. This high-contrast angular focusing characteristic gives the system excellent anti-interference capabilities, enabling reliable phase identification even in real-world applications with environmental disturbances or detection noise, based on the precise spatial position of the light intensity peak. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the principle of the diffractive optical neural network for mapping the superposition state phase to spatial angle in an embodiment of the present invention;

[0034] Figure 2 This is a schematic diagram of the phase distribution of the two phase modulation layers of the diffractive optical neural network in an embodiment of the present invention;

[0035] Figure 3This is a schematic diagram of the device structure of the orbital angular momentum superposition state phase measurement system in an embodiment of the present invention;

[0036] Figure 4 This is a bar chart showing the light intensity distribution and normalized intensity histogram of the output plane when the training phase φ=0° and φ=60° in an embodiment of the present invention.

[0037] Figure 5 In this embodiment of the invention, the light intensity distribution and normalized intensity histogram of the output plane when the untrained phase φ=45° and φ=140° are shown.

[0038] Among them, 1. laser, 2. beam expander, 3. attenuator, 4. spatial light modulator, 5. multilayer diffractive optical structure, 6. spatial distributed detection unit, and 7. polarization beam splitter. Detailed Implementation

[0039] The specific embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that the terms used in the following embodiments are defined as follows: Orbital angular momentum (OAM) refers to the angular momentum carried by the light field related to the helical phase wavefront; topological charge number l refers to the quantum number of the OAM eigenstate, characterizing the number of rotations of the helical phase; superposition phase φ refers to the relative phase when two OAM eigenstates with different topological charges are superimposed; diffractive optical neural network (DONN) refers to a fully optical computing architecture composed of multiple phase modulation layers and free-space diffraction propagation; normalized light intensity refers to the relative intensity value after normalizing the light intensity distribution to its maximum value. The present invention provides a target recognition system and method based on a quantum correlation-based discrete diffraction neural network 7.

[0040] Example 1

[0041] A system and method for measuring the phase of orbital angular momentum (OAM) superposition states based on diffractive optical neural networks (DONN), such as Figure 3 As shown, this embodiment provides a superposition state phase measurement system for orbital angular momentum based on a diffractive optical neural network, including a superposition state preparation module and a diffractive optical neural network measurement module;

[0042] The superposition state preparation module includes a laser 1, a beam expander 2, a polarization beam splitter 7, a spatial light modulator 4, and an attenuator 3. In this embodiment, the laser 1 emits a Gaussian beam with a wavelength of 632.8 nm, a beam diameter of approximately 2 mm, and a power of approximately 10 mW. The beam expander 2 consists of two convex lenses, the first lens with a focal length of 50 mm and the second lens with a focal length of 200 mm. The distance between the two lenses is the sum of the focal lengths of the first and second lenses, i.e., 250 mm, achieving a 4-fold beam expansion, extending the beam diameter to approximately 8 mm. The expanded beam is incident on the polarization beam splitter 7, which is a cubic polarization beam splitter prism that transmits P-polarized light and reflects S-polarized light. In this embodiment, the transmitted P-polarized light is selected as the subsequent modulation light field. Transmitted light is incident on spatial light modulator 4, which is a pure phase-type liquid crystal spatial light modulator with a resolution of 1920×1080 pixels, a pixel pitch of 8μm, a phase modulation range of 0 to 2π, and a refresh rate of 60Hz. Spatial light modulator 4 loads a computer-generated hologram, identifies the rotational symmetry features inherent in the intensity distribution of the input OAM superposition state, and establishes a model based on the phase parameters. The end-to-end physical mapping relationship to the macroscopic spatial angular coordinate θ; the hologram is generated using a phase superposition method. and The topological charge numbers are respectively and The mathematical expression for the superposition state of the OAM eigenstates is:

[0043]

[0044] Where φ is the phase of the superposition state to be measured; the superposition state light field output by the spatial light modulator 4 passes through the attenuator 3, which is a neutral density filter combination composed of multiple filters, and the attenuation range can reach up to 10. -10 In this embodiment, the adjustment is made to approximately 10 times. -9 The attenuation is multiplied by a factor of 1, reducing the light intensity to approximately 0.01 nW, enabling single-photon level (approximately 10^6 nW) light emission. 7 Measurements at the level of one photon.

[0045] The diffraction optical neural network measurement module includes a multi-layer diffraction optical structure 5 and a spatially distributed detection unit 6.

[0046] like Figure 1As shown, in this embodiment, the multilayer diffractive optical structure 5 consists of two phase modulation layers. The distance between the first and second phase modulation layers is 50 mm, and the distance between the second phase modulation layer and the output plane is also 50 mm. Each phase modulation layer is a pure phase-type diffractive optical element, implemented using a spatial light modulator 4 or a metasurface structure, with a resolution of 256×256 pixels, a pixel pitch of 10 μm, a total size of 2.56 mm × 2.56 mm, and a phase modulation range of 0 to 2π. The two phase modulation layers are connected to each other and to the output plane via free-space light field diffraction propagation. The light field propagation follows Fresnel diffraction theory, and the complex amplitude distribution of the light field is transformed according to the angular spectrum propagation operator during propagation.

[0047] like Figure 2 As shown, the phase distribution of the two phase modulation layers is optimized through deep learning training. The phase distribution of the first phase modulation layer exhibits a complex diffraction pattern with eight-fold symmetry, with the phase value continuously varying within the range of 0 to 2π, exhibiting a distinct periodic structure with a period of approximately 320 μm. The phase distribution of the second phase modulation layer also exhibits a complex diffraction pattern, but its symmetry differs from the first layer, displaying a spiral-like phase gradient distribution. The direction of the phase gradient is related to the rotational symmetry of the input orbital angular momentum (OAM) superposition state. The two phase modulation layers work together: the first layer is used to extract the rotational symmetry features of the input light field, and the second layer is used to map the extracted features to a specific spatial angular position on the output plane.

[0048] The spatial distributed detection unit 6 is located on the output plane and specifically includes 12 photodetectors. These 12 photodetectors are evenly distributed on a circle with a radius of 3 mm, with an angular interval of 30° between adjacent detectors, corresponding to 12 angular sectors: 0°, 30°, 60°, 90°, 120°, 150°, 180°, 210°, 240°, 270°, 300°, and 330°. Each photodetector has an effective detection area of ​​0.5 mm × 0.5 mm, a response wavelength range of 400 nm to 1000 nm, and a response time of less than 1 μs. The 12 photodetectors simultaneously collect light intensity information from their respective locations. The collected light intensity signals are converted into digital signals by an analog-to-digital converter, and the digital signals are input to a computer for further processing.

[0049] The method for measuring the phase of the superposition state of orbital angular momentum in this embodiment includes the following steps:

[0050] Step 1: Generate the superposition state optical field of the orbital angular momentum to be measured. The Gaussian beam emitted from laser 1 is expanded by beam expander 2 and then filtered by polarization beam splitter 7 to become P-polarized light. The P-polarized light is incident on spatial light modulator 4, which loads a pre-calculated hologram. The hologram encodes the superposition information of orbital angular momentum eigenstates with topological charge numbers l1=1 and l2=2, and the phase φ of the superposition state is to be measured. Spatial light modulator 4 modulates the phase of the incident light field and outputs the OAM superposition state optical field with topological charge numbers l1=1 and l2=2. The intensity distribution of the superposition state optical field exhibits a specific rotational symmetry pattern. The rotation angle of the pattern is related to the phase φ of the superposition state. When φ changes from 0° to 360°, the intensity distribution pattern rotates by 360° accordingly. After the superposition state optical field is attenuated to an appropriate intensity by attenuator 3, it is input as the optical field to be measured to the diffraction optical neural network measurement module.

[0051] Step 2: Input the light field to be measured into the multilayer diffractive optical structure 5, and propagate forward through diffraction and phase modulation. The light field of the superposition state of orbital angular momentum to be measured is first incident on the first phase modulation layer. The first phase modulation layer delays the incident light field pixel by pixel. The amount of phase delay is determined by the phase distribution of the first phase modulation layer. After the first phase modulation, the complex amplitude distribution of the light field changes. The changed light field propagates 50 mm in free space and then propagates to the second phase modulation layer. The light field is then delayed pixel by pixel again. The amount of phase delay is determined by the phase distribution of the second phase modulation layer.

[0052] After the second phase modulation, the light field continues to propagate 50 mm in free space to the output plane; the entire propagation process is achieved through the cascade of two phase modulations and two diffraction propagations, and the rotational symmetry characteristics of the input light field are gradually extracted and mapped to the spatial angular distribution of the output plane.

[0053] Step 3: Acquire the spatial distribution information of light intensity on the output plane. After the light field on the output plane is transformed by two layers of diffractive optical structures, the light intensity distribution exhibits a highly concentrated characteristic, with energy mainly concentrated in a specific angular sector on the circumference. Twelve photodetectors simultaneously acquire the light intensity values ​​of their respective angular sectors at a time of 1 ms. The acquired light intensity signal is an analog voltage signal, with the voltage amplitude proportional to the incident light intensity and a proportionality coefficient of 1 V / mW. The analog voltage signal is converted into a digital signal by a 12-bit analog-to-digital converter at a conversion rate of 1 MHz and a conversion accuracy of 0.024% of full scale.

[0054] Step 4: Identify the spatial angular position of the light intensity peak, and use the angle as the measured value of the superposition phase. Normalize the light intensity values ​​of the 12 angular sectors by dividing each light intensity value by the maximum value among the 12 light intensity values ​​to obtain the normalized light intensity value. Identify the angular sector with the largest normalized light intensity value. The angle θ corresponding to the angular sector with the maximum value is the estimated value of the measured superposition phase φ.

[0055] like Figure 4 As shown, measurements were performed on two typical phases φ=0° and φ=60° in the training set. When the input phase φ=0°, the light intensity distribution on the output plane exhibited a distinct bright spot located in the 0° sector. The normalized light intensity value of the 0° sector was 1.0, while the normalized light intensity values ​​of the other 11 sectors were all less than 0.1, indicating that the light intensity was highly concentrated in the 0° sector. When the input phase φ=60°, the bright spot position of the light intensity distribution on the output plane shifted to the 60° sector. The normalized light intensity value of the 60° sector was 1.0, while the normalized light intensity values ​​of the other sectors were all less than 0.1. The normalized intensity histogram shows that the measurement results for the training phases exhibited extremely high peak contrast, with the intensity ratio of the peak sector to the non-peak sector greater than 10:1. This indicates that the diffractive optical neural network possesses accurate classification capabilities for the phases in the training set, achieving a measurement accuracy of 100%.

[0056] Example 2: DONN Training Method

[0057] This embodiment provides a training method for a diffractive optical neural network, which is used to optimize the phase distribution of the two phase modulation layers in Embodiment 1, so that the diffractive optical neural network can accurately map the phase information of the superposition state of the input orbital angular momentum to the spatial angular position of the output plane.

[0058] The training method includes the following steps:

[0059] (1) Construct a training dataset containing 12 OAM superposition state samples with known phases. The phases of the 12 samples are φ=0°, 30°, 60°, 90°, 120°, 150°, 180°, 210°, 240°, 270°, 300°, and 330°, with a phase interval of 30°. Each sample is a superposition of orbital angular momentum eigenstates with topological charge numbers l1=1 and l2=2. The mathematical expression for the superposition state is:

[0060]

[0061] (2) Set the optimization objective. The optimization objective is to make the light intensity of the OAM superposition state sample with input phase φ=n×30° (n=0,1,2,...,11) reach the maximum value in the θ=n×30° angle sector on the output plane after propagation through two layers of diffraction optical structure, and the light intensity of the other angle sectors be as small as possible.

[0062] (3) Initialize the phase modulation layer parameters. The phase distribution of the two phase modulation layers is initialized to a random phase distribution, with the phase values ​​uniformly and randomly distributed in the range of 0 to 2π.

[0063] (4) Forward propagation calculation: For each sample in the training dataset, the intensity distribution of the sample is used as input to calculate the light intensity distribution of the input light field after passing through the first phase modulation layer, free space propagation, the second phase modulation layer, and free space propagation again to reach the output plane.

[0064] (5) Calculate the loss function. Based on the light intensity distribution of the output plane, calculate the light intensity values ​​of the 12 angular sectors. After normalizing the light intensity values, compare them with the real labels and calculate the cross-entropy loss function.

[0065] (6) Backpropagation optimization parameters: The backpropagation algorithm and gradient descent method in deep learning are used to optimize the phase distribution of the two phase modulation layers, and the learning rate is 0.01.

[0066] (7) Iterate until convergence, repeat steps 4 to 6, and perform 5000 iterations of training until the average loss function of the training set is less than 0.01.

[0067] like Figure 2 As shown, the phase distribution of the two phase modulation layers after training exhibits a complex diffraction pattern, and the average loss function of the training set decreases from the initial 2.48 to 0.008, indicating that the optimization process effectively establishes the mapping relationship from the phase parameter φ to the spatial angle θ.

[0068] Example 3: Verification of Interpolation Inference Ability

[0069] This embodiment, based on Embodiment 1, further verifies the interpolation inference capability of the diffractive optical neural network for phase values ​​that do not appear in the training set.

[0070] In this embodiment, two phase values ​​φ=45° and φ=140° that did not appear in the training set were selected as test samples. Following the method of Embodiment 1, the spatial light modulator 4 was used to generate OAM superposition light fields with phases φ=45° and φ=140°. The superposition light fields were input into the diffraction optical neural network trained in Embodiment 2, and the light intensity distribution of 12 angular sectors of the output plane was collected.

[0071] like Figure 5As shown, when the input phase φ = 45°, the light intensity distribution on the output plane exhibits a transitional characteristic. The light intensity is distributed in two adjacent sectors, 30° and 60°. The normalized light intensity value of the 30° sector is 0.52, and the normalized light intensity value of the 60° sector is 0.48. The light intensity values ​​of the two sectors are close and significantly higher than those of the other sectors. Using an interpolation inference algorithm, the energy ratio of adjacent sectors is calculated, and the inferred phase value is φ = (0.52 × 30° + 0.48 × 60°) / (0.52 + 0.48) = 44.4°, with an inference error of 0.6°.

[0072] When the input phase φ = 140°, the light intensity distribution on the output plane is biased towards the 150° sector. The normalized light intensity value of the 150° sector is 0.68, and the normalized light intensity value of the 120° sector is 0.32. Using the same interpolation inference algorithm, the inferred phase value is φ = (0.32 × 120° + 0.68 × 150°) / (0.32 + 0.68) = 140.4°, with an inference error of 0.4°.

[0073] Test results show that the inference error for all untrained phases is less than 5°, the average inference error is 2.1°, and the relative error is less than 3%. This indicates that the diffractive optical neural network has excellent interpolation inference capabilities and can accurately infer continuous phase values ​​from the light intensity distribution of corresponding sectors of adjacent training phases.

[0074] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A phase measurement system for orbital angular momentum superposition state based on diffractive optical neural network, comprising an optical field modulation unit for generating the optical field to be measured and a detection unit for detecting the optical field, characterized in that, It includes a multi-layer diffractive optical structure (5), which contains at least two phase modulation layers. Each phase modulation layer is connected by optical field diffraction propagation. The phase distribution of the phase modulation layer is optimized and trained to extract the rotational symmetry features of the input optical field and map them to the spatial angle of the output plane. The detection unit is a spatially distributed detection unit (6), which is set on the output plane and is used to collect light intensity information from multiple spatial locations simultaneously; The system obtains the phase information of the superposition state under test by identifying the spatial angular position of the light intensity peak.

2. The orbital angular momentum superposition state phase measurement system based on diffractive optical neural network according to claim 1, characterized in that, It also includes a superposition state preparation module, which comprises: Laser (1) is used to provide a coherent light source; A beam expander system (2), optically connected to the laser (1), is used to adjust the beam size; A polarization beam splitter, optically connected to the beam expander system (2), is used to filter polarization states; The spatial light modulator (4) is optically connected to the polarization beam splitter to form the light field modulation unit, which is used to load the hologram to generate the orbital angular momentum superposition state. The attenuator (3) is optically connected to the spatial light modulator (4) for adjusting the light intensity.

3. The orbital angular momentum superposition phase measurement system based on diffractive optical neural network according to claim 1, characterized in that, It also includes an interpolation inference unit, which is used to calculate the energy ratio of adjacent detection areas by analyzing the light intensity distribution of adjacent spatial locations, and to calculate continuous phase values ​​that do not appear in the training set.

4. A method for measuring the phase of a superposition state of orbital angular momentum based on a diffractive optical neural network, the method being applied to the system described in claims 1-3, comprising the step of generating the optical field of the superposition state of orbital angular momentum to be measured, characterized in that, Includes the following steps: The light field to be measured is input into a multilayer diffractive optical structure (5), which contains at least two phase modulation layers and propagates forward through diffraction propagation and phase modulation of the light field. Acquire spatial distribution information of light intensity on the output plane; Identify the spatial angular location of the light intensity peak and use this angle as the measured value of the superposition phase.

5. The method for measuring the phase of orbital angular momentum superposition state based on diffraction optical neural network according to claim 4, characterized in that, It also includes an interpolation inference step: for phase values ​​that do not appear in the training set, the energy ratio of adjacent detection areas is calculated by analyzing the light intensity distribution of adjacent spatial locations to infer continuous phase values.

6. The method for measuring the phase of orbital angular momentum superposition state based on diffraction optical neural network according to claim 4, characterized in that, The orbital angular momentum superposition state is a superposition of orbital angular momentum eigenstates with topological charges of l1=1 and l2=2, respectively.

7. The method for measuring the phase of orbital angular momentum superposition state based on diffraction optical neural network according to claim 4, characterized in that, The method is applied to quantum information processing, high-dimensional optical communication, or quantum state tomography.

8. A method for training a diffractive optical neural network, comprising the steps of constructing a training dataset and adjusting network parameters through an optimization algorithm, characterized in that, The method is used to train the multilayer diffractive optical structure (5) in the system of claim 1. The training dataset contains multiple orbital angular momentum superposition state samples with known phases; The optimization algorithm adjusts the phase distribution of each phase modulation layer so that the output light intensity forms a peak at the target spatial position, which corresponds to the phase parameters of the input sample. After training, a mapping relationship from phase parameters to spatial angles is established.

Images