A wavelength multiplexing focusing control device and its implementation method

Abstract

The present invention relates to a wavelength multiplexing focusing control device and its implementation method, the device includes a voltage-adjustable optical resonant cavity, a first dielectric metasurface layer and a second dielectric metasurface layer; wherein the voltage-adjustable optical resonant cavity includes a glass substrate, a metal metasurface, a liquid crystal layer, a metal metasurface, and a glass substrate arranged in sequence, and different coding modes are formed by applying voltages in different regions of the optical resonant cavity; the metasurface layer includes a silicon micron column structure and a dielectric substrate, and a diffraction neural network is used to simulate the diffraction process of multi-wavelength incident light, and the phase distribution of the double-layer metasurface layer is calculated; the overall amplitude distribution of the liquid crystal layer is changed by dynamically adjusting the external voltage to achieve continuous scanning of the focus point position of the light beam. The problem that traditional focusing control devices are insufficient in control accuracy and response speed is solved, and the device of the present invention has the characteristics of flexible design, small size, easy integration, easy processing, etc., and provides a more efficient and flexible beam control method.

Description

Technical Field

[0001] The present invention belongs to the field of terahertz science and technology, relates to beam scanning technology, is a wavelength multiplexing focusing control device and its implementation method, and is suitable for multi-band terahertz communication, imaging, radar detection and other fields. Background Art

[0002] The frequency range of the terahertz band is between 0.1 and 10 THz. It has ultra-wideband spectrum resources and excellent penetration performance, and is regarded as one of the important technologies of the 6G system. As wireless communication and imaging technologies develop towards flexibility and intelligence, the demand for active and rapid control of light beams is growing. Traditional focusing control devices can usually only work within a single wavelength range, making it difficult to achieve efficient wavelength multiplexing, limiting their prospects in multi-band applications. Existing equipment is difficult to effectively utilize multi-frequency signals, resulting in resource waste and performance bottlenecks. In addition, related technologies have deficiencies in control accuracy and response speed, making it difficult to meet the needs of modern communication and imaging systems for dynamic adjustment.

[0003] Wavelength multiplexing technology can effectively improve spectrum utilization by processing optical signals of multiple wavelengths on the same device at the same time, thereby enhancing the transmission efficiency and flexibility of the system. However, when achieving efficient deflection of multiple wavelengths, traditional methods often face the problems of complex phase control and slow real-time response. Liquid crystal materials have good dynamic response characteristics because they can change the molecular arrangement (director) under the action of electric or magnetic fields, thereby changing the refractive index. By applying an electric field, the dynamic response of liquid crystal materials in the terahertz frequency band can be used for optical control. Liquid crystal dynamic optical devices have the advantages of small size, light weight, low power consumption, fast response, and easy integration, which meet the requirements of modern communication equipment for thinness and low power consumption. At the same time, the rapid development of neural network technology has provided new possibilities for the optimization and control of terahertz optical deflection devices, and can efficiently optimize the phase and amplitude of terahertz waves in complex scenarios, thereby achieving more flexible multi-wavelength beam manipulation and dynamic real-time adjustment. Summary of the invention

[0004] In view of the fact that traditional focusing control devices are insufficient in control accuracy and response speed, and cannot meet the dynamic adjustment requirements of modern communication and imaging systems, a wavelength multiplexing focusing control device and its implementation method are proposed. It has the advantages of flexible design, small size, easy integration and easy processing, and is suitable for multi-band terahertz communication, imaging and radar detection.

[0005] The technical solution of the present invention is:

[0006] A wavelength multiplexing focusing control device comprises a voltage-adjustable optical resonant cavity, a first dielectric metasurface layer and a second dielectric metasurface layer; wherein the voltage-adjustable optical resonant cavity comprises a glass substrate, a metal metasurface, a liquid crystal layer, a metal metasurface and a glass substrate arranged in sequence, and voltage is applied to different regions of the optical resonant cavity to change the orientation of liquid crystal molecules in real time so as to form different coding modes; the metasurface layer comprises a silicon micron column structure and a dielectric substrate, and a diffraction neural network is used to simulate the diffraction process of multi-wavelength incident light, and gradient descent and back propagation algorithms are applied to calculate the phase distribution of the double-layer metasurface layer; the amplitude distribution of the liquid crystal layer is changed by dynamically adjusting the external voltage as a whole to achieve continuous scanning of the focus point position of the light beam.

[0007] Preferably, the incident light is linearly polarized light.

[0008] Preferably, the dielectric substrate is silicon.

[0009] Preferably, electromagnetic full-wave simulation software is used to calculate the performance of silicon micro-rod structures of different sizes under polarized light irradiation.

[0010] A method for realizing a wavelength multiplexing focusing control device is used for realizing the wavelength multiplexing focusing control device as described above, comprising the following steps: constructing an optical diffraction neural network, converting the control problem of a light beam into a matrix eigenvalue solution problem, adjusting the amplitude distribution of multi-wavelength incident light through an optical resonant cavity, applying gradient descent and back propagation algorithms to train the focusing position of the incident multi-wavelength incident light, optimizing the rotation angle of the silicon micron column in the metasurface in space, and finally realizing continuous and precise focusing control of the wavelength multiplexed light beam.

[0011] Furthermore, the diffractive neural network algorithm optimizes and calculates the phase distribution of the metasurface layer based on external inputs and settings, including the focus position, beam deflection direction, and amplitude distribution of the liquid crystal layer, and performs repeated training and adjustments;

[0012] The diffraction neural network training process uses a double-layer diffraction layer. The input layer is set close to the first diffraction layer. The distance between the two diffraction layers is 5000 microns. The distance from the second diffraction layer to the output layer is 7000 microns. The learning rate is set to 0.001, and the size of the neuron is 300 microns. In order to save training time, eight samples are taken as a batch and input into the neural network.

[0013] Furthermore, the specific operation of converting the light beam control problem into the matrix eigenvalue solution problem is:

[0014] Use E i(λ) (i = A, B, C, D) represents the wavelength-dependent independent output light field distribution of each phased array unit, α is the modulation coefficient of the phased array unit, representing the phase adjustment parameter used to form different coding modes, λ is the wavelength, M1, M2, M3, M4 are the preset output modes, that is, the total light field intensity under different modulation coefficient conditions, that is, under different input coding modes:

[0015]

[0016] Where α = 0, convert the above equation into matrix form:

[0017]

[0018] The conditions for the linear equations to be solved are: the number of rows of the matrix is ​​equal to the number of columns, and its determinant is not zero; if the rank of the matrix is ​​the same as the rank of the right-hand vector, then the equation has a solution; the light field distribution of the metasurface corresponding to a wide wavelength range under different input coding modes is respectively given by E A (λ), E B (λ), E C (λ), E D (λ) represents; these four parameters jointly determine the final phase distribution of the metasurface; thus, the problem of calculating the diffraction light field of the complex light beam is transformed into a problem of solving the matrix eigenvalue.

[0019] Furthermore, in the neural network framework, a series of known output modes M1, M2, M3, M4 and corresponding input coding data are constructed as training sets; input layer: constructed using multi-wavelength incident light, the different amplitude responses formed by the optical resonant cavity under the application of different voltages are encoded into a binary form, the transparent part is 1, and the opaque part is 0, and multiple coding modes are constructed as input; the four coding modes in the above matrix can be extended to more forms:

[0020]

[0021] Diffraction layer: Use two diffraction layers to apply Rayleigh-Sommerfeld diffraction to calculate the diffraction light field distribution of the incident light after passing through two dielectric metasurfaces; Output layer: Construct the distribution function of the focusing focus on the detection plane; Select the mean square error as the loss function to measure the gap between the predicted value and the true value; Use gradient descent or back propagation algorithm to train the network and adjust the network weights to minimize the loss function.

[0022] The beneficial effects of the present invention are:

[0023] 1. A wavelength multiplexing focusing control device and its implementation method disclosed in the present invention can realize continuous focusing control of wavelength multiplexing in the same device. By designing broadband geometric phase silicon micrometer columns and combining them with diffraction neural networks, the response of the device at different frequencies is optimized, so that the device can operate at multiple wavelengths / frequencies, thereby fully and flexibly utilizing terahertz spectrum resources.

[0024] 2. The wavelength multiplexing focus control device and its implementation method disclosed in the present invention can greatly enhance the response of liquid crystal and improve the modulation depth by utilizing the structure of voltage-adjustable optical resonant cavity and the resonance effect. At the same time, the thickness of the voltage-adjustable optical resonant cavity structure is low, so that the modulation rate of liquid crystal can be enhanced to achieve efficient and high-speed focus control.

[0025] 3. The wavelength multiplexing focusing control device and its implementation method disclosed in the present invention use a neural network to train the diffraction process of the light beam, and use a metasurface to construct the corresponding diffraction process, so that the complex diffraction light field design is simplified to a matrix solution. Using this feature, the focus position can be switched and scanned only by amplitude modulation, which greatly simplifies the difficulty of device design.

[0026] 4. The wavelength multiplexing focusing control device based on neural network optimization disclosed in the present invention has a simple structure and is easy to manufacture. Its design method consumes little computing resources and can achieve dynamic high-precision control through calculation. It provides a new technical solution for multi-band optical communications, imaging, sensing and detection. It has high flexibility and scalability and can demonstrate superior performance in wireless communications, terahertz imaging, radar detection and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A three-dimensional schematic diagram of a wavelength multiplexing focusing control device based on neural network optimization according to an embodiment of the present invention;

[0028] Figure 2 A control principle diagram of a wavelength multiplexing focusing control device based on neural network optimization according to an embodiment of the present invention;

[0029] Figure 3 This is a design diagram of a voltage-adjustable optical resonant cavity of a wavelength multiplexing focusing control device based on neural network optimization of the present invention;

[0030] Figure 4 A design diagram of a supersurface layer of a supersurface device of the present invention, including a top view of a single periodic structure and a partial array structure;

[0031] Figure 5 Schematic diagram of wavelength multiplexing effect of the metasurface device of the present invention;

[0032] Figure 6The present invention is a flowchart of the neural network optimization of the method for realizing the metasurface device.

[0033] Figure ID:

[0034] 1. Glass substrate; 2. Metal supersurface; 3. Liquid crystal medium; 4. First medium supersurface layer; 5. Second medium supersurface layer; 6. Micron column structure; 7. Medium substrate. DETAILED DESCRIPTION

[0035] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0036] A wavelength multiplexing focus control device and its implementation method, the device uses liquid crystal as a controllable material, adjusts the orientation of liquid crystal molecules by applying an external voltage, changes the transmission intensity, and realizes the amplitude modulation and deflection of multi-wavelength light beams. Combined with the liquid crystal medium layer and the metasurface structure, the optimization ability of the neural network is used to optimize the phase and amplitude distribution of electromagnetic waves of different wavelengths to ensure efficient focus control.

[0037] Specifically, the wavelength multiplexing focusing control device includes a voltage-adjustable optical resonant cavity, a first dielectric metasurface layer and a second dielectric metasurface layer, wherein the voltage-adjustable optical resonant cavity includes a glass substrate, a metal metasurface, a liquid crystal layer, a metal metasurface, and a glass substrate arranged in sequence. By applying voltage to different regions of the optical resonant cavity, the orientation of the liquid crystal molecules is changed in real time to form different coding modes, thereby realizing the amplitude control of the spatial distribution of the incident terahertz beam. The metasurface layer includes a sophisticatedly designed silicon micron column structure array and a dielectric substrate to achieve diffraction control. Based on the Huygens principle, each metasurface unit can independently control the outgoing light field, and the light emitted from the metasurface unit can be equivalent to a secondary sub-wave source and continue to propagate. At a propagation distance r, the electric field generated by the secondary sub-wave source can be expressed as:

[0038]

[0039] Among them, silicon micrometer columns with half-wave plate properties are arranged at a certain rotation angle in space to form a metasurface. By modeling and constructing an optical diffraction neural network in Python language through the above diffraction algorithm, the problem of light beam control can be converted into a matrix eigenvalue solution problem. Furthermore, the amplitude distribution of the incident light is adjusted in real time through the optical resonant cavity, and the focusing position of the incident multi-wavelength terahertz beam is trained using the gradient descent and back propagation algorithms to optimize the rotation angle of the silicon micrometer columns in the metasurface in space, and finally the continuous and precise focusing control of the wavelength multiplexed light beam can be achieved.

[0040] This method greatly simplifies the difficulty of beam control design. When the beam passes through the voltage-adjustable optical resonant cavity and the two-layer metasurface structure, a wavelength-dependent focused light spot can be formed on the detection plane. This method not only improves the focusing accuracy, but also can dynamically adapt to the needs of different focusing positions. It is widely used in broadband optical communications, imaging, detection and other scenarios.

[0041] The device structure and working principle of the present invention are:

[0042] 1. Device structure

[0043] The wavelength multiplexing focusing control device of the present invention includes a voltage-adjustable optical resonant cavity and a metasurface layer, which are specifically as follows:

[0044] Optical resonant cavity: It includes a glass substrate, a first metal metasurface, a liquid crystal layer (i.e., liquid crystal medium), a second metal metasurface, and a glass substrate arranged in sequence. The orientation of the liquid crystal molecules is adjusted by applying an external voltage, the dielectric constant of the liquid crystal material is changed, and the optical resonance effect is used to achieve real-time dynamic control of the beam amplitude within a wide frequency range.

[0045] The liquid crystal layer used is a liquid crystal medium material with anisotropy in the terahertz band and is anchored by an orientation layer.

[0046] The structure of the metal super surface layer is a grating structure with a thickness of 0.2 microns, a grating unit period of 150 microns, and a grating width of 110 microns. There is an orientation layer on the metal super surface for the orientation of liquid crystal molecules, and the orientation layer material is a polyimide film.

[0047] The optical resonant cavity can achieve amplitude modulation within a wide wavelength range, and the modulation depth can reach more than 90%.

[0048] Supersurface layer: including a well-designed silicon micron column structure array and a dielectric substrate, responsible for optimizing the effect of focusing control at different wavelengths. The silicon micron column structure is designed according to the principle of geometric phase, and the effect of terahertz phase modulation can be changed by rotating the angle.

[0049] The period of the metasurface unit structure is 300 microns, the number is 60*60, and the overall structure length and width dimensions are 18,000 microns and 18,000 microns.

[0050] Silicon microrods have the properties of a half-wave plate, rotating at different angles to form a calculated phase distribution; silicon microrods are designed based on the requirements of the half-wave plate properties, can operate over a wide frequency, and respond to wavelength multiplexing designs.

[0051] The dielectric substrate is silicon and its thickness is 600 microns.

[0052] 2. Working Principle

[0053] Amplitude modulation: By applying voltage to different areas of the liquid crystal, the transmittance is changed using the optical resonant structure to achieve different coding mode outputs.

[0054] Wavelength multiplexing control: Through the synergistic effect of the liquid crystal medium layer and the metasurface layer, light beams of multiple wavelengths can be focused and controlled at the same time.

[0055] Neural network optimization: Apply neural networks to learn and optimize the focusing process of light beams. By performing multi-task training on the diffraction of light beams of different wavelengths, the effect of wavelength multiplexing can be achieved, improving the efficiency and accuracy of focusing.

[0056] The incident light used was linearly polarized light.

[0057] The present invention proposes a wavelength multiplexing focusing control device based on neural network optimization. The device includes an optical resonant cavity and a metasurface layer. By applying voltage to different regions of the optical resonant cavity, the orientation of liquid crystal molecules is controlled to achieve amplitude modulation of light beams of different wavelengths. Figure 3 As shown in (b), the optical resonant cavity includes upper and lower glass substrates 1, upper and lower metal metasurfaces 2 and liquid crystal medium 3. (Note: The position expressions "upper" and "lower" in the present invention are only for the convenience of expression and understanding. The relative position from a certain perspective is not a limitation on the specific position) When the applied voltage changes, the molecular arrangement of the liquid crystal material changes accordingly, thereby adjusting the intensity of the transmitted light. The first dielectric metasurface layer 4 and the second dielectric metasurface layer 5 are both composed of a micron column structure 6 and a dielectric substrate 7. The direction and focus of the light beam are further controlled by geometric phase design. In this embodiment, different voltages are applied to the liquid crystal layer to encode different regions thereof into different states. This encoding form allows a multi-wavelength light beam to form a specific light intensity distribution after passing through the liquid crystal layer. After passing through the liquid crystal layer, the light beam further passes through two layers of optimized metasurface layers to form a focused light spot for multiple wavelengths on the rear detection plane. The focused light spots of different wavelengths can move continuously in space, and the position and movement of the light spots can be designed differently under different encoding states.

[0058] To achieve this process, this embodiment uses an optical diffraction neural network to simulate the diffraction process of the light beam. The phase distribution of the metasurface layer is optimized using gradient descent and back propagation algorithms to ensure that the light beam can be efficiently and accurately dynamically regulated under different voltage conditions. The metasurface layer design is based on the geometric phase principle and includes a silicon micron column structure array with a half-wave plate function. Among them, the rotation angle of each silicon micron column in the metasurface is optimized by a neural network, which can effectively adjust the diffraction process of different wavelengths to achieve focusing control.

[0059] In this embodiment, a neural network algorithm is used to optimize and calculate the phase distribution of the metasurface layer based on external inputs and settings, such as the focus position, beam deflection direction, and amplitude distribution of the liquid crystal layer. Through repeated training and adjustment, the device can achieve high-precision control of the terahertz beam in complex environments, with fast response speed and high control efficiency, and is suitable for various real-time application scenarios.

[0060] The diffraction neural network training process uses a double-layer diffraction layer. The input layer is set close to the first diffraction layer. The distance between the two diffraction layers is 5000 microns. The distance from the second diffraction layer to the output layer is 7000 microns. The learning rate is set to 0.001, and the size of the neuron is 300 microns. In order to save training time, eight samples are taken as a batch and input into the neural network.

[0061] Specifically, a three-dimensional schematic diagram of a wavelength multiplexing focusing control device based on neural network optimization according to an embodiment of the present invention is as follows: Figure 1 As shown, the incident light is linearly polarized light, which is incident vertically into the optical resonant cavity. The liquid crystal molecules form an arrangement under the action of voltage (V). The resonance of the metal metasurface electrode changes the amplitude of the light under the irradiation of the terahertz beam with a wide range of wavelengths. Different voltage modes produce different incident light coding modes, and then pass through the double-layer metasurface structure. The metasurface is composed of a sub-wavelength-scale silicon micron column structure array, which can highly accurately control the light wave. Applying the geometric phase principle, the silicon micron column of the half-wave plate also responds within a wide wavelength, so that the light beam is focused to different positions at different wavelengths, forming continuous wavelength control. The phase distribution of the metasurface structure is calculated and optimized by a neural network.

[0062] like Figure 2 As shown, this embodiment specifically implements a wavelength multiplexing focusing control device based on neural network optimization, and achieves precise control of the light beam by flexibly adjusting the voltage of the liquid crystal layer and the design of the metasurface layer. i (λ) (i = A, B, C, D) represents the wavelength-dependent independent output light field distribution of each phased array unit, α is the modulation coefficient of the phased array unit, representing the phase adjustment parameter used to form different coding modes, λ is the wavelength, M1, M2, M3, and M4 are the preset output modes, that is, the total light field intensity under different modulation coefficient conditions, that is, under different input coding modes:

[0063]

[0064] Where α = 0, convert the above equation into matrix form:

[0065]

[0066] The linear equations have a solution only if the number of rows of the matrix is ​​equal to the number of columns and its determinant is not zero. If the rank of the matrix is ​​the same as the rank of the right-hand vector, then the equation has a solution. The light field distribution of the metasurface corresponding to a wide wavelength range under different input coding modes is respectively given by E A (λ), E B (λ), E C (λ), E D (λ). These four parameters together determine the final phase distribution of the metasurface. In this way, the problem of calculating the diffraction light field of the complex light beam is transformed into the problem of solving the matrix eigenvalue.

[0067] In the neural network framework, a series of known output modes M1, M2, M3, M4 and corresponding input coding data are constructed as training sets. Input layer: constructed using multi-wavelength incident light, the different amplitude responses formed by the optical resonant cavity under the application of different voltages are encoded into a binary form, the transparent part is 1, and the opaque part is 0, and multiple coding modes are constructed as input. The four coding modes in the matrix above can be extended to more forms:

[0068]

[0069] Diffraction layer: Use two diffraction layers to apply Rayleigh-Sommerfeld diffraction to calculate the diffraction light field distribution of the incident light after passing through two dielectric metasurfaces. Output layer: Construct the distribution function of the focus on the detection plane. Select the mean square error as the loss function to measure the difference between the predicted value and the true value. Use the gradient descent or back propagation algorithm to train the network and adjust the network weights to minimize the loss function. The wavelength multiplexing in each input mode is as follows: Figure 6 As shown, when the wavelength changes continuously, the focused light spot moves continuously, realizing the free design of the focused light beam deflection.

[0070] like Figure 3 As shown in FIG. 1 , a design diagram of a voltage-adjustable optical resonant cavity of a wavelength multiplexing focusing control device based on neural network optimization of the present invention is shown in FIG. Figure 3 As shown in (a), the metal metasurface is a grating array structure. After applying voltage, the liquid crystal molecules in it will be redirected under the action of the voltage and turn to be distributed perpendicular to the grating layer. After the voltage is removed, the polyimide orientation layer used for liquid crystal molecule orientation on the metal metasurface will restore the original distribution. The transmission spectrum and phase distribution of the structure with and without voltage are shown in Figure 1. Figure 3 As shown in (c) and (d), the horizontal axis is frequency. In the frequency range of 0.65-0.72 THz, the amplitude modulation depth is greater than 90%, which is used to modulate the amplitude of the incident light field.

[0071] Figure 4This is a design diagram of the super surface layer of the super surface device of the present invention. Figure 4 (a) and (b) are the top view and front view of a single periodic structure, respectively. Figure 4 (c) is a top view of the array structure. The thickness of the micron column is h1 = 400 μm, the length and width of the micron column are l1 = 280 μm and w1 = 150 μm respectively, the thickness of the dielectric substrate is h2 = 600 μm, the unit structure period size is p1 = 300 μm, and θ is the rotation angle of the geometric phase, which is twice the phase.

[0072] Figure 5 This is a schematic diagram of the effect of the wavelength multiplexing metasurface focusing and deflecting device of the present invention. For the four coding mode inputs in the example, the outgoing light fields at five wavelengths are respectively given, and the focusing spot gradually moves upward under the condition of gradually changing wavelength.

[0073] The training model software uses the TensorFlow framework, which is based on the Python programming language.

[0074] Electromagnetic full-wave simulation software can be used to calculate how silicon micro-pillar structures of different sizes behave when illuminated by polarized light.

[0075] The present invention proposes a dynamic control focusing control device based on neural network optimization and its implementation method, which uses the voltage coding characteristics of the liquid crystal layer combined with the amplitude control of the metasurface layer to achieve dynamic focusing and control of the terahertz beam. By introducing the neural network optimization algorithm, the efficiency and accuracy of beam control are greatly improved, which can be widely used in wireless communications, terahertz imaging and detection, and other fields, providing a new solution for the application and development of terahertz technology.

[0076] The above-mentioned embodiment only expresses one implementation mode of the present invention, and its description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A wavelength multiplexing focusing control device, characterized in that: It includes a voltage-adjustable optical resonant cavity, a first dielectric metasurface layer and a second dielectric metasurface layer, which are sequentially arranged along the propagation direction of multi-wavelength incident light; wherein the voltage-adjustable optical resonant cavity includes a glass substrate, a metal metasurface, a liquid crystal layer, a metal metasurface and a glass substrate, and voltage is applied to different regions of the optical resonant cavity to change the orientation of liquid crystal molecules in real time so as to form different coding modes; the first dielectric metasurface layer and the second dielectric metasurface layer constitute a double-layer metasurface layer, which is used to diffract and control the multi-wavelength incident light after being encoded by the optical resonant cavity, and each metasurface layer includes a silicon micron column structure and a dielectric substrate, and a diffraction neural network is used to simulate the diffraction process of multi-wavelength incident light, and gradient descent and back propagation algorithms are used to calculate the phase distribution of the double-layer metasurface layer; the amplitude distribution of the liquid crystal layer is changed by dynamically adjusting the external voltage as a whole to achieve continuous scanning of the focal point position of the light beam.

2. The wavelength multiplexing focusing control device according to claim 1, characterized in that: The incident light used was linearly polarized light.

3. The wavelength multiplexing focusing control device according to claim 1, characterized in that: The dielectric substrate is silicon.

4. The wavelength multiplexing focusing control device according to claim 1, characterized in that: Electromagnetic full-wave simulation software is used to calculate the performance of silicon micro-pillar structures of different sizes under polarized light.

5. A method for implementing a wavelength multiplexing focusing control device, characterized in that: The implementation of a wavelength multiplexing focusing control device as described in any one of claims 1 to 4 comprises the following steps: constructing an optical diffraction neural network, converting the light beam control problem into a matrix eigenvalue solution problem, adjusting the amplitude distribution of multi-wavelength incident light through an optical resonant cavity, training the focusing position of the incident multi-wavelength incident light using gradient descent and back propagation algorithms, optimizing the rotation angle of the silicon micron column in the metasurface in space, and finally realizing continuous and precise focusing control of wavelength multiplexed light beams.

6. The method for implementing a wavelength multiplexing focusing control device according to claim 5, characterized in that: The diffraction neural network algorithm optimizes and calculates the phase distribution of the metasurface layer based on external inputs and settings, including the focus position, beam deflection direction, and amplitude distribution of the liquid crystal layer, and performs repeated training and adjustments; The diffraction neural network training process uses a double-layer diffraction layer. The input layer is set close to the first diffraction layer. The distance between the two diffraction layers is 5000 microns. The distance from the second diffraction layer to the output layer is 7000 microns. The learning rate is set to 0.001, and the size of the neuron is 300 microns. In order to save training time, eight samples are taken as a batch and input into the neural network.

7. The method for implementing a wavelength multiplexing focusing control device according to claim 5, characterized in that: The specific operation of converting the light beam control problem into the matrix eigenvalue solution problem is: Use E i (λ) (i = A, B, C, D) represents the wavelength-dependent independent output light field distribution of each phased array unit, α is the modulation coefficient of the phased array unit, representing the phase adjustment parameter used to form different coding modes, λ is the wavelength, M1, M2, M3, M4 are the preset output modes, that is, the total light field intensity under different modulation coefficient conditions, that is, under different input coding modes: Where α = 0, convert the above equation into matrix form: The conditions for the linear equations to be solved are: the number of rows of the matrix is ​​equal to the number of columns, and its determinant is not zero; if the rank of the matrix is ​​the same as the rank of the right-hand side vector, then the equation has a solution; the light field distribution of the metasurface corresponding to a wide wavelength range under different input coding modes is respectively given by E A (λ), E B (λ), E C (λ), E D (λ) represents; these four parameters jointly determine the final phase distribution of the metasurface; thus, the problem of calculating the diffraction light field of the complex light beam is transformed into a problem of solving the matrix eigenvalue.

8. The method for implementing a wavelength multiplexing focusing control device according to claim 7, characterized in that: In the neural network framework, a series of known output modes M1, M2, M3, M4 and corresponding input coding data are constructed as training sets; input layer: constructed using multi-wavelength incident light, the different amplitude responses formed by the optical resonant cavity under the application of different voltages are encoded into a binary form, the transparent part is 1, the opaque part is 0, and multiple coding modes are constructed as input; the four coding modes in the above matrix can be extended to more forms: Diffraction layer: Use two diffraction layers to apply Rayleigh-Sommerfeld diffraction to calculate the diffraction light field distribution of the incident light after passing through two dielectric metasurfaces; Output layer: Construct the distribution function of the focusing focus on the detection plane; Select the mean square error as the loss function to measure the gap between the predicted value and the true value; Use gradient descent or back propagation algorithm to train the network and adjust the network weights to minimize the loss function.

Images