Broadband polarization multiplexing far-field imaging metasurface reversely designed based on Huygens principle

Abstract

The invention relates to a broadband polarization multiplexing imaging metasurface which combines a newest machine learning algorithm and an advanced optical simulation technology and aims at achieving broadband high-definition and polarization multiplexing imaging effects. Meanwhile, the complete process of the metasurface from concept proposing to final design is elaborated, and by accurately developing machine learning model training and performing optical simulation verification, the polarization multiplexing far-field imaging metasurface in the same wavelength range is heart-shaped under x polarization and round under y polarization, and the metasurface is in the shape of the heart. It is ensured that the metasurface has excellent imaging performance in different polarization states. The innovative design not only improves the imaging quality, but also expands the application potential in various optical systems.

Description

(1) Technical field

[0001] The present invention proposes a broadband polarization multiplexing far-field imaging metasurface, which belongs to the field of micro-nano optics and computational imaging technology. Specifically, it relates to a broadband polarization multiplexing far-field imaging technology based on a metasurface. Through the metasurface structural unit, multi-dimensional control of the polarization state, phase, and spectrum of the light wave is achieved. By leveraging its own characteristics of presenting different images at different polarization positions, it highlights great potential in fields such as optical imaging and brings new technological breakthroughs to related fields. The embodiments of the present invention will demonstrate how to overcome the limitations of existing technologies through an intelligent collaborative design framework and provide a new method for the design and manufacture of broadband polarization zoom metalenses. (2) Background technology

[0002] Metasurfaces, a novel optical element, have attracted widespread attention for their ability to precisely manipulate light waves on a subwavelength scale. Metasurface technology can manipulate fundamental parameters of light waves, such as polarization, amplitude, frequency, and phase, demonstrating enormous potential for applications in optical imaging, display technology, information storage, and optical encryption. In particular, the application of metasurfaces in far-field imaging can significantly improve imaging quality and efficiency, enabling more sophisticated and complex light field manipulation.

[0003] Metasurfaces offer a new approach to miniaturization and multifunctional integration of optical systems by flexibly controlling the phase, amplitude, and polarization state of the light waves in their structural units. Prior art metasurfaces based on geometric phase (Pancharatnam-Berry phase) have achieved polarization-multiplexed holographic imaging, but are limited by their narrow bandwidth characteristics (e.g., single-wavelength design) and are therefore incompatible with multispectral imaging requirements. Furthermore, conventional metasurfaces often employ a single control mechanism (e.g., phase or amplitude), which limits their multimodal imaging capabilities.

[0004] Traditional polarization imaging relies on a cascade of multilayer optical components (such as wave plates and polarizers), resulting in large system size, complex optical paths, and low energy efficiency. Especially in broadband applications, material dispersion effects can lead to phase mismatches between polarization control at different wavelengths, limiting the operating bandwidth and imaging quality. For example, metal wire-grid polarizers exhibit significant absorption losses in the visible to infrared range and cannot be dynamically controlled.

[0005] Existing metasurface technologies still face challenges in achieving broadband and polarization-multiplexed far-field imaging. On the one hand, broadband metasurfaces need to maintain high performance across a wide spectral range, which places higher demands on material selection and structural design. On the other hand, polarization-multiplexing technology requires independent control of light waves with different polarization states on the same metasurface to achieve richer imaging effects and higher information capacity. Therefore, developing a far-field imaging metasurface that can simultaneously meet the requirements of broadband and polarization-multiplexing is of great significance for promoting technological progress in related fields.

[0006] Patent CN113126465B proposes a three-primary color holographic metasurface based on dual-channel polarization multiplexing and its design method. Using only a single structural dimension and incorporating geometric phase principles, this method achieves a phase-based computer-generated holographic metasurface by controlling the phase of each periodic nanounit. This metasurface can produce a color hologram in the far field at normal incidence.

[0007] Patent CN118938359A provides a broadband polarization optical device and system, which solves the traditional detection technology It is necessary to combine a variety of complex optical devices such as polarizers, wave plates, lenses, etc. to modulate the polarization and wavefront of light to achieve 450-660nm broadband polarization lens. (3) Summary of the invention

[0008] In response to the shortcomings of the existing technology, the present invention provides a broadband polarization-multiplexed far-field imaging metasurface designed based on the inverse design of the Huygens principle, which can achieve high-definition and polarization-multiplexed imaging effects in a broadband area.

[0009] At the input level, two different light fields are obtained in two polarization states (x, y), which are the target light fields of this design. The light field is a description of a set of light rays in space, which contains the direction and position information of the light rays.

[0010] The polarization-differentiated imaging plane (x, y) is introduced into the input light path, and the four-dimensional parameterization characteristics of the light field (the joint distribution of spatial position (x, y) and propagation direction (u, v)) are used to generate a dual-target light field with spatial-angle information difference.

[0011] The light field is defined as the phase space description of the radiation field, which satisfies the formula:

[0012]

[0013] For the sake of clarity, Figure 1 As shown, the binary target light field is input into the neural network to output the phase and amplitude of the metasurface, and the predicted light field under x-pol and y-pol is calculated by Huygens. The error gradient of the target light field and the predicted light field is calculated by MSE and fed back to the neural network so that the weights can be updated for optimization iteration.

[0014] For the sake of clarity, Figure 2 As shown, under x and y polarization incidence, they are marked as heart and circle respectively, with a specific size of 10 mm, and each imaging is configured with 10,000 discrete points for image processing.

[0015] For the sake of clarity, Figure 3 As shown in the figure, the schematic diagram of the structure of the super atom has SiO2 as its base and Si square pillars on top.

[0016] The system uses image binarization technology as input, which is implemented through simulation software. This design improves imaging accuracy and processing efficiency.

[0017] Furthermore, the image binarization technique involves setting the grayscale values of pixels in an image to 0 or 1 to achieve a distinct black and white effect. Specifically, the method generates a binary image by binarizing the grayscale values of each pixel in the image so that each pixel has only two possible values: pure black or pure white.

[0018] Furthermore, the binarization processing method can highlight the key features of the image, improve the accuracy of image recognition and analysis, and also help to extract key information in subsequent design.

[0019] The method of using a neural network for data processing in the present invention refers to processing two light field signals through an activation function ReLU in a multi-layer neural network.

[0020] Furthermore, the activation function defines the functional relationship between the output of the upper layer neurons and the input of the lower layer neurons. Specifically, the activation function is a key component in implementing nonlinear transformations, which enables artificial neural networks to learn and simulate complex data patterns.

[0021] The ReLU function expression is:

[0022] f(x)=max(0,x)

[0023] The artificial neural network used in the present invention adopts an auto-encoder structure, which is an unsupervised learning model composed of a large number of interconnected processing units and can process information nonlinearly and adaptively.

[0024] The autoencoder described therein realizes effective encoding and decoding of input data through internal nonlinear mapping, thereby extracting key features of the data.

[0025] The optical imaging method based on the metasurface involved in the present invention calculates the complex amplitude distribution of discrete points on the metasurface through the Huygens-Fresnel principle to achieve precise control of the light field distribution on the imaging surface.

[0026] Using the Huygens principle, each discrete point is considered a secondary wave source, which generates secondary waves on the imaging plane at a distance of 10 mm. The Huygens principle states that the superposition of spherical waves emitted by each secondary wave source on the imaging plane will form the final wavefront, thereby determining the amplitude distribution on the imaging plane.

[0027] The amplitude of the spherical waves emitted by each discrete point on the imaging plane is superimposed and calculated. This step involves complex mathematical operations that need to take into account the amplitude and phase of each point, as well as their position relative to the imaging plane.

[0028] The above superposition calculation ultimately yields the amplitude distribution on the imaging surface. This distribution reflects the imaging properties of the metasurface under specific polarization conditions and has important guiding significance for the design and optimization of metasurfaces.

[0029] Each discrete point (x i ,y i ) is expressed by the spherical wavelet as:

[0030]

[0031] Construct the complex amplitude matrices of x polarization and y polarization respectively:

[0032] (IV) Description of the accompanying drawings

[0033] Figure 1 Flowchart of the framework for reverse engineering of metasurfaces based on autoencoder neural networks

[0034] Figure 2 Schematic diagram of the x, y polarization incident metasurface

[0035] Figure 3 Schematic diagram of the superatom structure

[0036] Figure 4 The light field diagram from 0.8THz to 0.95THz under x-pol incidence

[0037] Figure 5 The light field diagram from 0.8THz to 0.95THz under y-pol incidence (V) Specific implementation methods

[0038] The following text describes the technical concept of the present invention through specific examples, aiming to more clearly illustrate its core innovations and implementation paths. It should be noted that the embodiments described herein are merely a concrete expression of the technical solution of the present invention, and the combination of its technical features does not constitute a limitation on the scope of protection of the claims.

[0039] The upper-layer neurons perform a weighted sum operation, calculating the weighted sum of the input signal and the corresponding weight to obtain the output value of the upper-layer neurons.

[0040] The obtained output values of the upper-layer neurons are input into the activation function ReLU for processing. The activation function generates the input values of the next layer of neurons through calculation. The introduction of the activation function ReLU is of great significance. It can improve the nonlinear fitting ability of neural networks, allowing neural networks to more effectively process complex light field data and realize matrix-based processing of light field data.

[0041] This method can accurately calculate the phase and amplitude of the light field. These calculated phase and amplitude correspond to the phase and amplitude on the metasurface, reflecting the characteristics of the light field on the metasurface.

[0042] Based on the calculated light field phase and amplitude, the method of the present invention can predict and control the phase and amplitude on the metasurface. By training and adjusting the neural network, it is possible to accurately predict the phase and amplitude changes of the light field on the metasurface and effectively control them according to actual needs.

[0043] After completing the neural network pre-training, a physics-inspired autoencoder is implemented for the achromatic polarization-multiplexed zoom metalens.

[0044] The autoencoder process starts with input circular and cardioid target images containing amplitude distributions at four specific frequencies (0.8 THz, 0.85 THz, 0.9 THz, 0.95 THz) in x and y polarizations.

[0045] The main neural network outputs a structure matrix with dimensions of 100 × 100 × 2. The 100 × 100 grid is used to define the spatial position of the super-atom, and the two channels of the matrix specify the length and width of the super-atom, respectively.

[0046] In the main neural network, the hidden layer uses the ReLU (Rectified Linear Unit) activation function for nonlinear feature extraction; the output layer uses the Sigmoid function to limit the size of the super-atom to the range of 16-144μm.

[0047] The obtained structural parameters were fed into a pre-trained network to predict the polarization- and frequency-resolved transmission amplitude (in the range of 90%-100%) and phase (in the range of 0-2π). The predictions were then propagated according to Huygens theory to calculate the amplitude distribution at 10 mm (x-pol) and 10 mm (y-pol).

[0048] The mean square error (MSE) loss between the predicted and target amplitudes is used as the driving force to perform gradient backpropagation to iteratively optimize the structural parameters for 200 times.

[0049] There is a theoretical upper limit on the amplitude (10,000), assuming an ideal 100% transmission. However, in practice, this upper limit cannot be achieved due to the inherent limitations of dielectric atoms (efficiency ranges from 80-95%) and the lack of a "perfect" geometry that can simultaneously achieve broadband operation and polarization multiplexing.

[0050] The present invention designs a metasurface that contains 10,000 discrete meta-atoms, each of which can independently control the amplitude and phase of light waves transmitted through it.

[0051] Furthermore, metaatoms are arranged in a specific geometry on the metasurface to achieve precise control of x- and y-polarized light. By precisely designing the geometric parameters of each metaatom, we can control the amplitude change and phase delay of light waves as they pass through the metasurface.

[0052] Using Huygens' principle, which states that every point source can act as the origin of a new spherical wave, the effects of these discrete point sources are superimposed to predict the amplitude distribution on the imaging surface at a distance of 10 mm.

[0053] By calculating the superposition effect of the spherical waves generated by each meta-atom on the imaging plane, the overall amplitude distribution pattern can be predicted.

[0054] The amplitude and phase of 10,000 discrete points on the metasurface under x and y polarization are calculated using the Huygens superposition principle, and the amplitude distribution of 10,000 discrete points on the imaging surface at 10 mm under x and y polarization is calculated.

[0055] The results are as follows Figure 4 as well as Figure 5 As shown, they are polarization images formed under x and y polarization after the incident frequency is 0.8THz-0.95THz, which are heart-shaped and circular, respectively.

Claims

1. A polarization multiplexing broadband metasurface, characterized in that: include: a. a substrate, which is SiO2; b. a metasurface structure unit array, which is arranged on the surface of the substrate, and the metasurface structure unit array is composed of a plurality of metasurface structure units arranged periodically; c. wherein the metasurface structure units are configured to be able to achieve different functional multiplexing of incident electromagnetic waves with different polarization states within a wide frequency range.

2. The polarization multiplexing broadband metasurface according to claim 1, characterized in that: The metasurface structure unit includes at least two substructures of different shapes or sizes, and the substructures are designed to have different electromagnetic response characteristics to electromagnetic waves in a specific polarization state to achieve polarization multiplexing function.

3. The polarization multiplexing broadband metasurface according to claim 1, characterized in that: The broadband range is 0.8-0.95 THz. Within this broadband range, for incident electromagnetic waves with two different polarization states, the metasurface structure unit can achieve broadband polarization multiplexing far-field imaging.

4. The polarization multiplexing broadband metasurface according to claim 1, characterized in that: The substrate is made of a low-loss dielectric material, the relative dielectric constant of the dielectric material is 1.56, and the loss is less than 10%.

5. The polarization multiplexing broadband metasurface according to claim 1, characterized in that: The arrangement period of the metasurface structure unit array is 160um, and the arrangement period is optimized to meet the performance requirements of broadband and polarization multiplexing.

Images