System and method for parallel polarization analysis
By integrating the grating of linear birefringent elements in the optical imaging system, parallel polarization analysis and full Stokes imaging are realized, solving the problems of grating design and system complexity in the prior art, and achieving compact and efficient polarization image acquisition.
Patent Information
- Application Number
- CN202080038933.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-04-15
- Filing Date
- 2020-04-14
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2040-04-14
AI Technical Summary
The prior art is difficult to achieve efficient parallel polarization analysis, and traditional grating design is complex, requiring multiple gratings to be interleaved or cascaded to complete full Stokes imaging, increasing the complexity and volume of the system.
Design an optical component including a substrate and a metasurface, using a linear birefringent element as a grating to achieve parallel polarization analysis and full Stokes imaging, integrating into the imaging system through a single grating, eliminating the dependence on moving parts and polarized optics.
A compact full Stokes imaging system is realized without additional polarization optics and moving parts, improving the simplicity and efficiency of the system, enabling effective analysis of multiple polarization steps and outputting high-quality polarization images.
Smart Images

Figure CN113906320B_ABST
Abstract
Description
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS
[0002] This application claims the benefit of and priority to U.S. Provisional Application No. 62 / 834,343, filed on April 15, 2019, which is incorporated herein by reference in its entirety.
[0003] STATEMENT OF FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT
[0004] This invention was made with Government support under National Science Foundation Grants 1144152 and 1541959. The Government has certain rights in this invention. Background Art
[0005] In this disclosure, a simple formalism for understanding polarization-dependent diffraction from polarization-sensitive gratings in the paraxial limit is presented. This formalization can be motivated and used to design gratings capable of parallel polarization analysis; in contrast to previous approaches, this is done here with a single grating, rather than using spatial interleaving or cascading of such gratings as is traditionally done. These gratings can be fabricated and experimentally characterized. It can also be shown how such gratings can be integrated into imaging systems, allowing for a compact full-Stokes imaging camera with no moving parts and no polarization optics (except the grating elements). Some of the imaging capabilities of this camera are demonstrated. This disclosure presents remarkable generalizations to polarization in diffraction, metasurfaces, and Fourier optics, suggesting new research directions in these areas and providing new perspectives on polarization technology. Summary of the invention
[0006] At least one aspect of the present disclosure relates to an optical component, comprising a substrate and a metasurface, the metasurface comprising one or more linear birefringent elements. The metasurface is disposed on the substrate. The one or more linear birefringent elements define a grating configured to perform parallel polarization analysis on multiple polarization orders of incident light of arbitrary polarization.
[0007] In some embodiments, the grating is configured to implement parallel polarization analysis without cascading incident light through bulk polarization optics. In some embodiments, the grating is not spatially interleaved with any other grating. In some embodiments, the metasurface is configured to implement full Stokes polarization measurement independent of any other optical components. In some embodiments, multiple polarization orders correspond to tetrahedrons inscribed in a Poincaré sphere. In some embodiments, the metasurface is configured to output light and analyze four different polarizations in parallel, none of which is orthogonal to any of the other polarizations. In some embodiments, one or more linear birefringent elements include an array of dielectric pillars having form birefringence. In some embodiments, multiple polarization orders include six polarization orders corresponding to octahedrons inscribed in a Poincaré sphere. In some embodiments, the metasurface implements a collection of analyzers for arbitrarily specified polarizations on the diffraction orders of the metasurface.
[0008] Another aspect of the present disclosure relates to an optical component, comprising a substrate and a metasurface, the metasurface comprising one or more linear birefringent elements. The metasurface is disposed on the substrate. The metasurface is configured to process incident light into a spatially varying Jones matrix.
[0009] In some embodiments, the metasurface includes a plurality of birefringent wave plates, and one or more of the angular orientation, absolute phase shift, and phase retardation of at least one of the plurality of birefringent wave plates is configured to process incident light as a spatially varying Jones matrix. In some embodiments, the spatially varying Jones matrix is linearly birefringent. In some embodiments, the spatially varying Jones matrix is unitary. In some embodiments, the spatially varying Jones matrix is an operator that operates in parallel on each of a set of diffraction orders of interest.
[0010] Another aspect of the present disclosure is directed to an imaging system including a metasurface, a lens, and an imaging sensor, wherein the metasurface includes one or more linear birefringent elements, and the metasurface is configured to provide the imaging sensor with multiple polarization orders sufficient for the imaging sensor to determine a full Stokes vector.
[0011] In some embodiments, the imaging sensor is configured to determine the full Stokes vector without using polarization orders other than those provided by the metasurface. In some embodiments, the metasurface is a surface of a substrate as a single substrate. In some embodiments, the metasurface defines a grating that is not spatially interlaced with any other grating, and the grating is configured to provide multiple polarization orders. In some embodiments, the metasurface defines a grating, and the grating is configured to provide multiple polarization orders without cascading incident light through multiple gratings. In some embodiments, the metasurface is configured to receive incident light and output multiple polarization orders without any moving parts. In some embodiments, the imaging sensor includes at least four regions, and the metasurface is configured to provide a corresponding one of a plurality of images corresponding to a corresponding one of at least four independent polarization orders to each region of the imaging sensor. In some embodiments, the imaging sensor is configured to simultaneously provide a corresponding one of a plurality of images to each region of the imaging sensor. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] To better understand the nature and purpose of some embodiments of the present disclosure, reference should be made to the following detailed description in conjunction with the accompanying drawings.
[0013] Figure 1 Schematically illustrates a layered view of polarization dependent paraxial diffraction according to some embodiments.
[0014] FIG. 2A to FIG. 2F A tensor grating for arbitrary parallel polarization analysis is shown in accordance with some embodiments.
[0015] Figure 3A A tensor grating integrated with an aspheric lens is shown to image four diffraction orders onto four quadrants of a CMOS imaging sensor in accordance with some embodiments.
[0016] Figure 3B It is shown that each copy of the image in each quadrant has been analyzed along a different polarization according to some embodiments.
[0017] Figure 3C According to some embodiments Figure 3A Side view of the ray trajectories in .
[0018] Figure 3D Optical microscope images of grating samples according to some embodiments are shown.
[0019] Figure 3E According to some embodiments Figure 3A The imaging system in can be packaged into a practical prototype with adjustable focal length.
[0020] Figure 4Indoor and outdoor photography with a camera is shown in accordance with some embodiments.
[0021] Figure 5 The transfer function required to produce a grating with exactly two non-zero diffraction orders according to some embodiments is shown.
[0022] FIG. 6A illustrates a schematic diagram of diffraction by a diffraction grating to a finite order of magnitude between k=m and k=n according to some embodiments.
[0023] 6B shows that each of equations S20 checks all pairs of orders separated by a given integer distance, starting with N in the first equation and proceeding downward, with the base case being the nearest neighbor diffraction orders, according to some embodiments.
[0024] Figure 7 Grating analysis of a tetrahedron (top) or octahedron (bottom) for polarization states is shown for each of two cases, in accordance with some embodiments.
[0025] Fig. 8A Parallel polarization analysis of a linear birefringent tensor grating is shown in accordance with some embodiments.
[0026] Figure 8B Some examples of analyzer configurations subject to symmetry conditions according to some embodiments are shown.
[0027] Fig. 9 It is shown how a contradiction appears at the midpoint of the diffraction pattern that prevents the solution of the equation.
[0028] Fig.10 The Fourier transform of a grating according to some embodiments is shown.
[0029] Fig.11 An overview of an optimization-based design flow for gratings of some embodiments is shown, according to some embodiments.
[0030] Fig.12 The calculated far field of an optimized tetrahedral grating according to some embodiments is shown.
[0031] Fig.13 Optimized tetrahedral and octahedral gratings are shown in accordance with some embodiments.
[0032] Fig.14 Scanning electron micrographs of octahedral and tetrahedral gratings are shown according to some embodiments.
[0033] Fig.15 Mueller matrix polarimetry is shown in accordance with some embodiments.
[0034] Fig.16The percentage of incident power directed into four or six diffraction orders of interest is shown according to some embodiments.
[0035] Fig.17 A side view schematic diagram of a camera is shown in accordance with some embodiments.
[0036] Fig.18 Angle-dependent calibration according to some embodiments is shown.
[0037] Fig.19 An image of a chessboard subjected to polarization analysis is shown in accordance with some embodiments.
[0038] Fig. 20 A synthesis of a full Stokes polarization image is shown in accordance with some embodiments.
[0039] Fig.21 Polarization processing of an image is shown in accordance with some embodiments. DETAILED DESCRIPTION
[0040] At least some embodiments of the present disclosure relate to metasurface gratings.
[0041] In the most general case of optical physics, Maxwell's equations are solved in their full complexity. However, in a wide variety of practical scenarios, simplification presents itself in the form of the scalar theory of diffraction, from which the concept of plane wave spread emerges. In a plot of the plane wave spread (or angular spectrum), an electromagnetic disturbance in a plane can be understood as being formed by the interference of many plane waves incident at different angles. These plane waves can then propagate forward individually to find the field anywhere else in space. This simple, intuitive understanding is the basis of the field broadly known as Fourier optics, and is fundamental to modern optical physics, especially holography and diffraction.
[0042] One can imagine scenarios where light (e.g., a plane wave at normal incidence, or some more complex intensity and phase distribution) encounters an obstacle that transforms the field. Plane wave expansion means that the Fourier transform of the field immediately after the obstacle provides information about the field propagating forward in space. In the traditional formulation of this problem, the field is assumed to be scalar valued. The question of how to extend this formalism to include the polarization state of the light and polarization-dependent obstacles naturally arises. This is particularly poignant in light of recent advances in micro- and nanofabrication and holography (over the past few decades), which have enabled the practical implementation of obstacles with customized, spatially varying polarization properties on a subwavelength scale at optical frequencies.
[0043] Strictly speaking, the traditional description of the polarization state of a plane wave is not included in the plane wave expansion diagram. As a two-dimensional description of the electric field path, the concept of polarization ellipse does not apply to the three-dimensional electric field generated when plane waves with different propagation directions interfere, as is inherent in plane wave expansion. However, in all plane waves In the nearly collinear paraxial (small angle) limit, the traditional 2D description of light polarization and plane wave spreading remains consistent. Here (whether explicitly stated or not), we can work in this paraxial limit.
[0044] In the following, three polarization-dependent diffraction regimes can be introduced, which are generally successively increased, such as Figure 1 outlined in .
[0045] The first mechanism that can be identified is that of scalar diffraction. In scalar diffraction, one can imagine light striking a planar surface, creating a complex-valued (amplitude and phase) field behind the obstacle. In Fourier (angular frequency) space, the grating is equivalently represented by the coefficients a k An infinite set of , some of which are evanescent, is described by The Fourier decomposition of is given by Figure 1 The top depiction. Figure 1 A layered view of polarization-dependent paraxial diffraction is shown. In the scalar regime (top), the periodic electric field distribution Generates a scalar weight In the vector regime (middle), the full polarization state It is allowed to vary periodically with space and produce a characteristic Jones vector (polarization) {|j k >}. Finally, in the tensor mechanism, the Jones matrix of the grating can be made spatially variable as - In this case, the Fourier coefficients {J k} are themselves operators that encode polarization-dependent behavior. If a specific polarization |λ〉 is incident, then the vector grating can be recovered from the tensor grating, whereas if the analysis is performed along the polarization |ξ〉, then the scalar grating can be recovered from the vector grating.
[0046] Scalar mechanisms may not admit polarization. However, scalar mechanisms can be the base case of a hierarchy that one can seek to build upon. Strategies for understanding polarization-dependent diffractive optics - particularly in the field of metasurfaces - can include a reduction to two such scalar diffraction problems. Polarization-dependent elements can be designed so that when a given polarization is incident, the output polarization state is uniform as a function of space, while the amplitude and phase vary. When an orthogonal polarization is incident, the output polarization is uniform again and undergoes a second scalar transfer function. In this way, optical elements with different functionalities can be implemented for selected linear, circular, and most recently, arbitrary elliptical polarization bases. It may be noted that what is broadly known as the "geometric" or "Pancharatnam-Berry" phase is a subcase of this scalar approach for creating scalar phase profiles for circularly polarized light (although the geometric phase itself is more general). Applying this scalar picture to polarization-dependent problems is a significant simplification and conceptual aid that allows the full complexity of polarization to be ignored by assuming a field with uniform polarization.
[0047] Increasing this complexity by one degree means a field where the Jones vector, the polarization state, varies with space:
[0048]
[0049] Here, we can use the brake symbol for the Jones vector, where is the complex-valued amplitude of the spatial variation, and χ and is the spatially varying parameter of the polarization ellipse. This can be called the vector mechanism of diffraction (not to be confused with the full vector diffraction in Maxwell's equations). Mathematically, the polarization distribution can be transformed into Projecting onto the grating steps. Increase to After that, the Fourier coefficients themselves become vectors, that is, they themselves are composed of the set {|j k >} gives the polarization ellipse. Equivalently, the polarization ellipse The spatial distribution of the different polarizations {|j k >} many beams (paraxial) interference. This Figure 1 This diagram of vector diffraction has been widely recognized before, with distinctions in the fields of polarization gratings, polarization holography, and more recently, metasurfaces.
[0050] Implicit in the scalar and vector mechanisms described above is the concept of an incident wave. The grating modifies this incident wave to generate a spatially varying complex field (in the former) or a distribution of polarization states (in the latter). In the most general case, a mechanism that can be referred to here as tensor diffraction, the behavior of the grating can be decoupled from the amplitude and polarization state of any particular illuminating plane wave. In fact, the most general way to view polarization-dependent gratings is that the polarization operator (the Jones matrix itself) is a spatially varying quantity:
[0051]
[0052] The concept of plane wave expansion again naturally extends to this tensor mechanism. However, the Fourier weights of the plane wave are now also operators, and the Jones matrix {J k When the spatially varying operator nature of the grating is acknowledged, the diffraction orders are independent of any light that may illuminate the grating, and instead behave in a sense that all possible incident plane wave polarizations are accounted for at once. k . Tensor Raster The weight J imposed on the spatial carrier can be k However, in contrast to the traditional wave interference pattern, it is the polarization operator that interferes, not the electric field. Figure 1 Bottom overview.
[0053] This tensor way of looking at paraxial diffraction has not been widely recognized or used in the literature of polarization and diffractive optics. In essence, it is nothing more than a direct extension of simple Fourier analysis to high-dimensional space. However, its real power lies in its inversion. Assuming there is a set of desired polarization devices to be implemented (e.g., polarizers, wave plates, optical active elements, or anything whose behavior can be described in a Jones matrix), represented by the operator {J k} and the corresponding set of diffraction orders of interest {l}. The operator has a spatially varying Jones matrix of the form:
[0054]
[0055] A single diffractive optical element will / can be constructed that implements all of these (potentially complex) polarization-dependent functions in parallel, and a mathematical prescription for its design and implementation is provided.
[0056] Finally, note that the three mechanisms of polarization-dependent diffraction—scalar, vector, and tensor—are actually closely related. If a plane wave with the Jones vector |λ> is incident on the tensor grating On, then the vector raster Produced, where the polarization weight is |j k >=J k|λ>. If the vector grating is projected (analyzed) along a specific polarization |ξ>, then the scalar grating Produces, where the scalar Fourier coefficient is After all, a scalar raster is a complex number that varies in space, a vector raster contains two, and a tensor raster contains four. This hierarchy is in Figure 1 Depicted in.
[0057] II. Parallel Polarization Analysis via Unitary Polarization Grating
[0058] A. Basic principles
[0059] The formalism presented in the previous section is general. Equation 3 provides a recipe for implementing diffractive elements with fairly arbitrary polarization behavior, but it is not specific to the nature of the desired optical function (as contained in {J k}) or suitable optical media to implement the synthesis
[0060] Here we can pursue more specificity, starting from the previous question. Of particular interest in polarization optics is the polarization analyzer, i.e., the polarizer. The polarizer is perhaps the most fundamental element in polarization optics and has practical importance in the measurement of polarization (polarimetry). Moreover, in previous studies of polarization-dependent diffraction optics, gratings that analyze (or "split") different polarization states at the order of the grating have been the focus of attention, especially for integrated polarimetry. In view of this, in some embodiments, we can focus on the order J of the grating. k A grating is implemented as a polarization analyzer (but this choice is not fundamental).
[0061] The element that acts as a polarizer can be described by a binary matrix, or in linear algebra, the outer product. Its Jones matrix can be expressed as
[0062]
[0063] Equation 4 describes the Jones matrix analyzed for a selected Jones vector q>: When q> is incident, the intensity transmission is maximum, but if instead we have<z|q> = 0, the output light goes out. The light emitted from J will be able to carry the polarization state |p>. The conventional polarizer familiar from laboratory experience has |p> = |q>, but this is not the core of the polarizer's Malus' law-like intensity transfer properties. Finally,
[0064] are complex (scalar-valued) weights.
[0065] Having decided k} should act as a polarization analyzer (Eq. 4), the desired tensor grating can be easily calculated by Eq. 3 But what physical optical elements can implement the results? Here, another specific choice can be made (again, this is not fundamental to the nature of tensor diffraction): a very advantageous implementation medium takes the form of a metasurface consisting of an array of dielectric pillars with formal birefringence. Locally, the Jones matrix of these metasurfaces can be well approximated by a linear birefringent waveplate, where the angular orientation, absolute phase shift, and phase retardation are all continuously adjustable and independent quantities, given by
[0066]
[0067] In other words, this metasurface can implement a sampled tensor grating where Locally, it has the form of Equation 5, where θ, and are all spatially varying quantities. By inspection, Equation 5 describes the Jones matrix, which 1) is unitary everywhere, i.e., for all where 1 is the 2×2 identity matrix, and more specifically, 2) linear birefringence, treating linear polarization as all The feature vector of .
[0068] From the general picture of tensor diffraction presented in Section I, two specific choices are made that the diffraction orders of the grating should behave like an analyzer and that its implementation must / can locally take the form of Eq. 5. But are these particularities mathematically consistent with each other? That is, for the implementation of a polarization analyzer (For a general choice of diffraction order {l}) the required equation 3 is For all Do they all take the form of a unitary, linearly birefringent waveplate (Eq. 5)?
[0069] In the Supplement, a rigorous investigation of this problem is given. There, it is shown that the linear birefringence requirement of Eq. 5 implies that J k Should be symmetrical. k This indicates that each should take the form That is, the output polarization of the analyzer should be the complex conjugate of the Jones vector being analyzed. Physically, this means that the analyzer does not change the shape of the polarization ellipse while reversing the handedness of its rotation.
[0070] However, even taking into account linear birefringence, the grating must / can also be unitary everywhere. It can be shown in the supplement that this restriction specifically implies that if one can stick to tensor gratings of exactly nth order, which are all polarization state analyzers, then the form of equation 5 can be matched everywhere (only) if n=2 and the analyzed polarizations are strictly orthogonal.
[0071] B. Design strategy and optimization
[0072] In other words, a tensor grating implemented by a metasurface that locally obeys Eq. 5 may not have arbitrarily many diffraction orders that act as polarization state analyzers, with no light leaking into other orders. But what if some light is allowed to leak into other diffraction orders? Then, the J of some orders k can behave like analyzers for arbitrary polarization states with a finite amount of leakage to other diffraction orders (implementing their own J k ), so that the overall grating formed in Equation 3 is still unitary everywhere. This motivates an interesting physical perspective on these tensor gratings as coupled systems, where adjacent diffraction orders compensate for the desired polarization-dependent behavior in a way that preserves unitarity.
[0073] How can we concentrate as much optical power as possible in the diffraction order of interest, while minimizing leakage? This is an optimization problem. Here, we may wish to implement a grating that locally takes the form of Equation 5 It has a set of diffraction orders {l}, each with J k , J k is the analyzer for any given set of polarization states {|q>}, as defined in Equation 4. Furthermore, one may wish to do this with as little light leaking as possible into diffraction orders other than {l}. The quantity can be defined as The power of the diffraction order k, and when the orthogonal polarization is incident The power of the diffraction order k. The way to express this optimization problem is to maximize In order to minimize leakage into undesired stages and maintain I q with I z The contrast between them should be as high as possible to constrain J k acts as an analyzer (see Eq. 4). This can be solved, for example, by gradient descent-based methods.
[0074] A detailed discussion of this optimization procedure is included in the supplement (where variational methods can also be used to describe the path to the solution in analytical form).
[0075] In the next section, these gratings can be designed and implemented using this approach.
[0076] III. Experiment
[0077] A. Case Study
[0078] In this paper, a conceptual framework and a scheme for designing tensor gratings that implement arbitrary, parallel polarization analysis are proposed. In the field of polarimetry or polarization state measurement, gratings that analyze incident light along different polarization states in parallel are of particular interest. In order to reconstruct the full Stokes vector of the incident beam, at least four projection measurements along different polarization states can be performed. In order to maximize the fidelity of this reconstruction, these states should be as different from each other as possible, which can be done by choosing polarization states corresponding to Platonic solids inscribed in the Poincaré sphere (polarization state space). To this end, the formalization of tensor diffraction and the above scheme can be applied to two case studies:
[0079] Tetrahedron: This grating analyzes four polarization states at the four innermost diffraction orders corresponding to its tetrahedron inscribed in the Poincaré sphere, one of which is circular.
[0080] Octahedral: This grating analyzes the six polarization states corresponding to an octahedron inscribed in the Poincaré sphere. These are |x>, |y>, |45>, |135>, |R>, and |L>, perhaps the most common polarization ellipse. These are analyzed at the six internal diffraction orders.
[0081] Some comparative work has addressed the problem of designing diffraction gratings that are able to separate and therefore analyze light based on its polarization state. However, in general, these works have adopted a scalar approach, attempting to impart opposite blazed grating phase profiles on orthogonal polarization states. Therefore, several individually designed gratings must / can be interleaved (often referred to as "spatial multiplexing") or cascaded in series to create a grating whose order analysis is for any more than two polarization states. The awareness of a tensor approach shows that this is not necessary - all functionality can be integrated into a single grating, and functions that are impossible to implement by interleaving individual beamsplitting gratings can be implemented. For example, the tetrahedron example above cannot be implemented in this way because none of its four polarizations is orthogonal to any of the others.
[0082] The grating described above is to be implemented as a square, two-dimensional grating (but this is not the primary option). FIG. 2A to FIG. 2F A tensor grating for arbitrary parallel polarization analysis is shown. Figure 2AShown using an optimized design scheme, gratings are designed to analyze four (top) and six (bottom) polarization states corresponding to the polarizations of the vertices of a tetrahedron (top) and octahedron (bottom) inscribed in a Poincaré sphere. On the left, the diffraction orders and polarizations they analyze are plotted in k-space. In the right two columns, an 11×11 element grating unit cell consisting of TiO2 rectangular pillars is shown, for which they perform this analysis at λ = approximately 532nm as designed (middle) and as fabricated (SEM, right). Figure 2B Schematic diagram showing Mueller matrix polarimetry: the grating is illuminated with variable polarization while the output on a single diffraction order is recorded using a full Stokes polarimeter to measure its Mueller matrix. Figure 2C The contrast (top row) and the intensity normalized to the incident power (bottom) of each grating order are shown for tetrahedral (left) and octahedral (right) gratings. For each order, the results of the analytical grating design (optimized), full-wave simulation of the grating, and experimental measurements are given. Each order is marked by the polarization ellipse it analyzes. Figure 2D The polarization analyzed by each grating from analysis grating design, simulations and experiments is shown. Figure 2E and Figure 2F The Poincaré sphere is shown for each order of a tetrahedral (E) or octahedral (F) grating, representing the output polarization at that grating order as a function of the input polarization. The wireframe shows how a point on the Poincaré sphere is represented by the experimentally measured Mueller matrix for each order. Map to new points. On each sphere, from Figure 2D In the figure, arrow "a" indicates the measured analyzer polarization |q (m,n) >, arrow "b" indicates the polarization of the equatorial mirror While arrow "c" represents the average output polarization of all points, according to the formalization here, it should overlap with the "b" arrow to a large extent, as shown. The darkened area or cap shows the average deviation from this average output polarization.
[0083] exist Figure 2A The leftmost column of These diffraction orders and the desired polarization state can be fed into the optimization mentioned above, yielding a local form of Eq. 5 (unitary linear birefringence). The locally required Jones matrix to the geometry of the actual structure can be mapped by reference to a library of such structures in the material platform / wavelength regime of interest. In some embodiments, that material platform is TiO2 columns. The high optical index of these structures allows operation at technologically important visible wavelengths, thereby facilitating the imaging applications discussed in Section IV. Other embodiments are possible, and neither TiO2 nor visible light wavelengths are necessary. Moreover, the grating proposed here contains 11 such elements on a side, separated by steps of approximately 420 nm. This choice is made to suit the imaging applications of Section IV while keeping the diffraction angle at λ = approximately 532 nm paraxial (θ D ~6.6°).
[0084] The grating unit cell for the tetrahedral and octahedral cases is Figure 2A (top and bottom, respectively), both shown in their designed (drawing) and manufactured (electron micrograph, right) states. Each is subdivided hundreds of times to create the grating.
[0085] B. Mueller Matrix Polarization Measurement
[0086] The tensor gratings described above are designed to implement specific polarization-dependent functions on their diffraction orders. To verify whether this is done in practice, the polarization-dependent behavior of each diffraction order of interest can be measured. In polarization optics, the Jones and Mueller matrices describe the polarization-dependent behavior. But of the two, the Mueller matrix is directly measurable, describing the light intensity rather than the electric field.
[0087] To this end, a Mueller matrix polarimetry can be performed, i.e., the 4x4 Mueller matrix is experimentally determined for each grating order (m,n) of interest. This is done by using a polarimeter with a known Stokes vector The grating is implemented by sequentially illuminating the grating with lasers of λ=about 532 nm of several (at least four) different input polarization states. For each input Stokes vector, a full Stokes polarimeter can be placed in the order of interest to record the corresponding output Stokes vector Will Link to The matrix This can then be determined numerically from the data. Figure 2B This is schematically outlined in and discussed in more detail in the Supplementary .
[0088] C. Results
[0089] The result of the design process is analysis It is possible to perform a numerical Fourier transform to obtain the Mueller matrix for each order, and it is also possible to simulate (using e.g. FDTD) the corresponding unit cell geometry, yielding a simulated Mueller matrix for each order (Supplement). Coupled with the experimentally determined Mueller matrix from fabricated gratings, there are thus three realms in which gratings are compared - digital, simulated, and experimental reality, with decreasing mathematical abstraction in each. If all three agree, then the general approach to tensor diffraction presented here can be said to be defensible in practice.
[0090] Direct comparison of matrix quantities is inherently difficult. Instead, since each diffraction order of interest is assumed to act as a polarization analyzer, it makes sense to ask the following questions: Do these orders act as analyzers? For which polarizations? And, how efficiently?
[0091] The present disclosure solves these problems at that stage. Figure 2C start. Figure 2C The left side of is dedicated to the tetrahedral case, while the right side is dedicated to the octahedral case. Figure 2C Each column of bars in represents a diffraction order designed to analyze the polarization ellipse plotted below (or above) it. Each group contains three bars, one for the numerical optimization results, one for the simulation, and one for the measurement. Figure 2C In the upper part of the , the contrast is plotted, i.e. the normalized difference between the intensity on that order when its preferred polarization is incident relative to the orthogonal polarization. In this respect, the optimized results are numerically predicted to be close to perfect (100%). From the full-wave simulations, the predicted contrast drops somewhat, and then drops again in the measured gratings. However, in all cases, the measured gratings' orders show a polarization contrast of more than 90%. From this, it is clear that all desired orders of both gratings act as polarization state analyzers as expected.
[0092] Second, it must / can be verified that these orders not only act as analyzers, but also act as analyzers for the desired polarization states specified in the design. Figure 2D In , the polarization state with maximum output intensity for each grating order, i.e., the analyzer polarization, is plotted on a Poincaré sphere next to a Platonic solid representing the desired state. It can be seen that in the numerically optimized gratings, simulations, and measurements, the analyzer polarizations are close to their desired counterparts ( Figure 2D There is no dramatic confusion in the , where one analyzer polarization lands far away and appears closer to the other, falsely exaggerating the correspondence).
[0093] Both gratings show high contrast for the desired polarization state, demonstrating that the tensor approach to designing these gratings is effective. Next, we can turn to the question of efficiency. Being tensor, the diffraction efficiency of a grating cannot be quantified with a single number - it is polarization dependent. Figure 2C In the figure, for each order of the tetrahedral (left) and octahedral (right) gratings, the I q , when it comes from Figure 2D The maximum intensity that can be observed at an order occurs when the preferred polarization of I is incident (as a fraction of the power of the illumination beam). Note that even given a perfect grating that directs all light to the desired order, I q Nor can it be 100% unless all other orders of analysis are orthogonally polarized. Therefore, the symmetry constraints imposed by energy conservation (supplement) reduce the I of tetrahedral and octahedral gratings to q Restricted to one-half and one-third respectively. Figure 2C As can be seen in the optimization results and simulation predictions, something close to these limits is possible. In the measurements, it can be seen that I q is between one-half and two-thirds of what might be expected. This is a difference that is expected to be resolved by increasing control over the manufacturing process. The efficiency of both gratings is high enough to be of practical use. Averaged over all possible input polarizations, the diffraction efficiency in terms of power in the order of interest exceeds 50% of the incident power (Supplement). It can also be noted that in Figure 2C There are cases where measurements appear to outperform simulations, or simulations outperform numerical methods for a particular order. In isolation this may seem wrong, but gratings are complex, coupled systems, and imperfections that degrade the performance of a few grating orders may unexpectedly benefit another.
[0094] Finally, note that Figure 2C and 2D The data contained in does not actually require the determination of the full Mueller matrix, since the efficiency, analysis polarization state, and contrast can all be extracted from just the first row of the Mueller matrix, which determines the intensity of the outgoing beam. The remainder of the Mueller matrix controls the polarization state of the output beam, which may not be resolved. This is Figure 2E and 2Fwhere each sphere represents the behavior of one diffraction order of a tetrahedral or octahedral grating, respectively. The Poincaré sphere is uniformly sampled at 1600 points, which form the intersection points of the wireframe. All of these polarizations are numerically manipulated by the experimentally measured Mueller matrix for each order and the result is plotted as a new, distorted wireframe. If each order behaved like a perfect analyzer in the manner of Equation 4, then all points in the original sphere would map to a single point at the Stokes vector corresponding to |p> in the distortion map. Since this is not the case in experimental reality, Figure 2E and 2F The sphere in contains points distributed over the entire sphere, but the degree to which these points on the sphere are concentrated in one direction and very sparse elsewhere is This proves that each diffraction order does act as an analyzer, especially for tetrahedral gratings.
[0095] For each grating order, arrow "a" corresponds to the vector being analyzed (as previously described in Figure 2D ), that is, q in Eq. (m,n) >. Arrow "c" shows the average output polarization state across the grating levels, while the dark spherical cap shows the average deviation from this state. Finally, arrow "b" depicts It is relative to |q (m,n) > flips the chirality and is therefore mirrored about the equator of the sphere. According to the diagram presented above, the linear birefringence of the grating element means that the output takes in the form of. Figure 2E and 2F The close overlap of the “b” and “c” arrows in each of the figures supports this and provides further validity for the tensor approach.
[0096] IV. Full Stokes Polarization Imaging
[0097] These tensor gratings are now being applied to areas of practical interest. Both octahedral and tetrahedral gratings, as proposed above, can be used as compact and integrated full Stokes polarimeters (devices that can determine the polarization state of a beam) with no other optical components and no moving parts. Metasurface-based polarimeters have recently attracted great interest. However, for various academic and industrial fields, polarization cameras or imaging polarimeters are more practical. Polarization cameras capture all or part of the Stokes vector at each point in the image. In the case of four-element Stokes vectors, this involves four image acquisitions along independent polarization directions, which can be acquired in time sequence (limiting time resolution and often requiring moving parts), by composing a focal plane array with micropolarizers (requiring expensive manufacturing and usually not providing full Stokes vector determination) or by capturing images simultaneously along four paths, each with independent polarization optical elements (significantly increasing system volume and complexity).
[0098] Grating-based methods are used in some comparative applications. However, due to the limitations of those applications, multiple gratings - either cascaded in series or patterned next to each other - are used to implement measurements for full Stokes vector determination. In the following, it can be shown that the tensor method of some embodiments enables the development of flexible full Stokes imaging cameras without such complexity. The developed cameras and gratings can have a wide range of applications in machine vision and other fields.
[0099] A. Design of Imaging System
[0100] The task here is to integrate one of the above gratings into a photographic imaging system. A tetrahedral grating is chosen because it offers the possibility to determine the Stokes vector using (only) four measurements (the necessary minimum). An imaging system can be developed consisting of a tensor grating (implemented with a TiO2 metasurface with linear birefringent elements) followed by an aspherical lens (f=20 mm, the choice of which is discussed in the supplement) and finally a CMOS imaging sensor. This is done in Figure 3A Relative to Figure 1 , the grating is rotated 45° so that each of its orders forms an image of the scene on one quadrant of the imaging sensor. Each quadrant will / can then contain a version of the photograph analyzed along its characteristic polarization - these images can be acquired simultaneously and the Stokes vectors reconstructed pixel by pixel ( Figure 3B ), thereby forming a polarized image.
[0101] Simple ray tracing was used to optimize all aspects of the system: the separation of the grating and the aspheric surface, the separation of the aspheric surface and the imaging sensor, and finally the grating period (and therefore the diffraction angle θ D). The goal of the optimization is to acquire parallel ray beams over a ±5° field of view (FOV) - assuming they come from very distant objects - and to focus them as closely as possible within the quadrant of the image sensor used (defined in terms of the commonly used Strehl ratio). Furthermore, the tensor gratings come with an azimuthally symmetric (scalar) phase profile, which is added to the tensor grating during design and experienced by all diffraction orders. This produces a weak lensing effect that aids imaging. Although the design focuses only on a single grating order, by default all other grating orders will / can be imaged to their corresponding quadrants, since the imaging system is rotationally symmetric about its optical axis.
[0102] Figure 3A A tensor grating is shown integrated with an aspheric lens to image the four diffraction orders onto the four quadrants of a CMOS imaging sensor. A ray trace for one diffraction order is shown. The camera is designed to image distant objects, so each color corresponds to a different parallel beam incident on the grating over a ±5° field of view (FOV). Scale bar is 6 mm. Not shown: 10 nm bandpass filter at approximately 532 nm and aperture in front of the grating to limit the FOV.
[0103] Figure 3B Each copy of the image in each quadrant has been analyzed along a different polarization. Pixel-level intensity differences at different points in the scene can be used to synthesize a single polarization image of the scene, where the intensity difference at each point is is known.
[0104] Figure 3C Shows Figure 3A A clear side view of the ray trace in. The top, middle, and bottom correspond to ray beams incident at +5°, 0°, and -5°, respectively.
[0105] 3D shows an optical microscope image of a grating sample with a diameter of 1.5 mm and showing the Fresnel zone due to the weak lensing effect imposed on top of the tensor grating. Scale bar 250 μm.
[0106] Figure 3E Shows Figure 3A The imaging system in can be packaged into a practical prototype with adjustable focus.
[0107] The actual ray traces of the imaging system are Figure 3A and a clear side view is shown in Figure 3C The different ray colors correspond to parallel ray bundles incident on the tensor grating at different angles. The optimized grating has N = 11 elements with a pitch of about 420 nm, producing θ at λ = about 532 nm. D =6.6° (this is the reason why the grating unit cell size in Section III is also 11×11). Figure 3D A microscope image of a 1.5 mm sample is shown in . The sample is surrounded by metal to block stray light and shows the Fresnel zone, which arises from the additional lens phase profile imposed on top of the grating, the periodicity of which is too small to discern at this magnification.
[0108] Finally, the entire imaging system can be packaged into a prototype for practical use in polarization photography. Figure 3E , and allows focus adjustment (at the expense of this freedom, its size can be greatly reduced). An aperture in front of the camera allows control of the FOV and a 10 nm bandpass filter at approximately 532 nm prevents the dispersive nature of the grating from interfering with the imaging. Further details are deferred to the Supplementary Information.
[0109] B. Polarization Imaging
[0110] The grating-based camera can then be used for practical full Stokes polarization imaging. The raw sensor acquires approximately Figure 3B In order to merge the four images into one polarization image, the four images must / can be aligned (registered) with each other, forming a vector at each pixel of the image If the camera is calibrated, that is, if the polarization analyzed by each diffraction order is precisely known, then the Stokes vector at each pixel can be calculated as Here A is a matrix whose rows are the analyzer Stokes vectors. Image registration and polarization calibration are solved simultaneously by the angular correlation method described in the Supplement.
[0111] In the polarization image, is known at every pixel, but the entire Stokes vector is not easy to visualize. Instead, an image can be formed from scalars derived from the Stokes vector. In the literature on polarization imaging, two parameters of the polarization ellipse are commonly used and visualized. The first is the azimuth, given by the inverse tangent (S2 / S1), which yields the physical orientation of the polarization ellipse. The second is the degree of polarization, given by Given, it quantifies the degree to which the light is (or is not) fully polarized. Finally, an image can be formed from S0 alone, which is the intensity of the light, producing an image that would be acquired by a conventional non-polarization sensitive camera.
[0112] exist Figure 4 , example photos captured by a polarization imaging system are shown. For each image acquired raw on the sensor, S0 (standard monochrome photo), azimuth and DOP are shown. In all these images, the illumination light is unpolarized and diffuse, i.e., incident from all directions. Figure 4The indoor images in were taken under diffuse LED lighting, while the outdoor images were taken in broad daylight. Each example is described below in turn.
[0113] Figure 4 Parts (a) to (e) of FIG. 3 show indoor (parts (a)-(c)) and outdoor (parts (d)-(e)) photography with the camera depicted in FIG. 3. In each case, the unprocessed raw exposure, S0 (a conventional monochromatic intensity image), the azimuth of the polarization ellipse (in degrees, given by arctan(S2 / S1)), and the degree of polarization (DOP, given by See Section IV for details on each case.
[0114] Figure 4 (a) A simple test is constructed. The target is a paper frame with eight sheets of polarizing film, with their axes aligned radially outward - allowing image-forming light to pass through the sheets. A conventional photograph shows no difference between the two sheets (S0), but the azimuth image accurately reveals their angular orientation. Furthermore, the DOP image shows that light passing through the sheets is polarized relative to the surrounding surface.
[0115] Figure 4 (Part be) examines the polarization dependence of specular reflection. When unpolarized light is specularly reflected, it becomes partially polarized; in contrast to diffusely reflected light, which remains unpolarized, the direction of this partial polarization is perpendicular to the plane of incidence. Figure 4 In part (b) of Figure 1, a conventional soda bottle is imaged head-on. From the intensity image (S0) alone, the shape of the bottle cannot be determined, but the azimuth image shows a smooth, continuous change around the top of the bottle. This demonstrates the conical shape of the top of the soda bottle and can be used to characterize its 3D shape; in fact, the polarization properties of specular reflection have been proposed as a means of depth imaging. Figure 4 Part (c) of illustrates the same concept for a more complex object, a human face. The 3D nature of the face is not known a priori from the S0 image (as opposed to, say, just a printed photograph of the face), but the azimuth image tracks its 3D contours and can be used in 3D facial reconstruction.
[0116] at last, Figure 4 Parts (d) and (e) depict outdoor scenes. In (d), a bicycle is parked on the grass after it rained on the Harvard campus. In the foreground is the boundary between the grass and asphalt, as well as puddles. In the azimuth image, very clear contours can be seen between the wet road surface (where the polarization direction is clearly defined to be parallel to the ground) and the grass and puddles (where the azimuth is somewhat random because the reflection is diffuse). Moreover, the azimuth reveals the presence of an asphalt sidewalk behind the image, which was difficult to see in the intensity image. Figure 4In (e), a row of cars is imaged. Cars illuminated by sunlight are a favorite target of the polarization imaging literature because windshields tend to produce strong polarization signatures. This can be seen in all three cars in the azimuth image (where the windshield and body have a well-defined polarization azimuth) as well as in the DOP image (where the light from the windshield is highly polarized relative to the rest of the car).
[0117] Figure 4 The examples included in the paper demonstrate natural but powerful applications of this device for machine vision, remote sensing, and other fields. While these have been recognized before in machine vision and polarization imaging, the camera presented here does not require polarization optics beyond a single grating, can acquire data simultaneously (e.g., all data required for polarization analysis) without moving parts, can be used with simple imaging sensors (no micropolarizers), and can be made compact and mass-producible. Moreover, the design approach for the gratings - just a single grating with an imaging lens - shows that these types of gratings can be designed around existing imaging systems to create polarization-sensitive systems.
[0118] V. Conclusion
[0119] The present disclosure motivates a hierarchical picture of polarization-dependent diffraction in paraxial Fourier optics, which includes scalar, vector, and tensor mechanisms. The tensor picture of diffraction describes diffraction elements that are not orders with simple scalar weights, but can produce polarization-dependent behavior in their diffraction orders {J k}. The J of the polarization analyzer can be specified here k The authors present a novel approach to the problem of polarization in the presence of light and provide an optimization-based approach to its design. Two such gratings can be fabricated in the form of a dielectric metasurface and their performance fully characterized. Since these gratings analyze the incident polarization in parallel, they can be used to determine the polarization state of light. It can be shown how the gratings can be integrated into an imaging system to implement this in a spatially resolved manner, resulting in a polarization camera that does not require additional polarization optics, moving parts, or dedicated sensors. This camera is demonstrated in several photographic scenarios of practical interest.
[0120] The present disclosure is an advance in the understanding and design of polarization-dependent diffractive optics, going beyond simple phase profiles. While polarization analyzers have been discussed herein, the tensor approach presented is quite general: gratings implementing a variety of polarization operators (e.g., wave plates and more exotic behaviors) can be implemented. Moreover, the constraints of linear birefringent elements are by no means fundamental, and through the degrees of freedom provided by lithographic fabrication, these behaviors can be implemented using elements with more complex polarization responses. Finally, gratings enabled by this tensor approach present a simple means of full-Stokes polarization imaging that can be extended to other imaging systems and wavelength regimes. Compact passive devices such as those presented herein enable widespread use and adoption of polarization imaging in machine vision and related fields.
[0121] Replenish
[0122] S1. Mathematical description
[0123] Here, an attempt could be made to provide a smoother picture of the logic that underpins some of the methodology and results that support the tensor diffraction approach presented in the main text, providing a clarity that spatial limitations would not allow. In the course of this development, several of the results presented in the main text can be proved. This section is organized to parallel the simpler case of a scalar diffraction grating, which simply directs light into different directions in a polarization-insensitive manner. Let's start with these.
[0124] A. Beam splitting by a diffraction grating
[0125] A diffraction grating is an obstacle that varies periodically in space. As a result of the periodicity, the light encountering the grating is split into a finite number of diffraction orders. If the transmission function with periodic variation If a diffraction grating is illuminated by a unit amplitude plane wave incident normally on the grating, then the grating order k will / can be propagation, where λ is the wavelength of the illumination light and d is the period of the grating. In addition, each diffraction order has a complex-valued scalar weight a k gives the amplitude of the electric field of that order, and thus the strength and efficiency of order k. This weight is given by the term e in the Fourier series t(x) -ikx The coefficients of give:
[0126]
[0127] Here, for convenience, d = 2π can be taken. A question that naturally arises is how to design a grating that splits an incident plane wave into a finite number of orders with equal intensity (or more generally, some specified weighting), or equivalently, that can combine a finite number of beams into one plane wave. This problem has practical implications for optical system design.
[0128] According to the Fourier transform shift theorem, a grating that maps an input plane wave to one diffraction order will / can always be given by a linear phase ramp, so this is a simple example.
[0129] The next simplest example will be to consider a grating where the light is split equally into two diffraction orders, say (for demonstration) the first and the zeroth. The desired transmission hologram is then given directly by the sum:
[0130]
[0131] By mathematical construction, if a grating with this transfer function is illuminated by a plane wave of unit amplitude at normal incidence, then exactly two orders of equal intensity will / can appear. However, it is worth noting that this will / may not map a unit amplitude plane wave to an output with uniform amplitude. In fact, given by The specified power transfer function is now spatially varying:
[0132]
[0133] Figure 5 shows the transfer function required for a grating that produces exactly two non-zero diffraction orders. The x-axis is in units of grating period d. T(x) is plotted at Figure 5 There is a symmetry that ensures that (global energy conservation). This is explained in equation S2, which has the factor However, at some points a transmission above 1 is required, while at other points a transmission below 1 is required. Although energy is conserved globally, it is not conserved point by point: local gains and losses may be required. That is, this requires the use of a grating that conserves energy globally, but requires redistribution of light between different parts of the grating unit cell.
[0134] At this point, one might wonder what would happen if only losses were implemented (it is actually easier to modulate local absorption). If equation S3 is scaled by a factor of 1 / 2, then it will / can reach a maximum of one (no gain), and will / can only require losses (which is also in Figure 5 (depicted by a dotted curve in the figure).
[0135] But this comes at a price: Now And half of the incident energy must / can be absorbed by the grating to split the light into only two orders. If generalized to the case of N equal intensity orders, then it is easy to see that the power scaling factor will / can become 1 / N to ensure that only amplitude modulation is lost. In order to send light in N directions with a grating, must / can an efficiency of η = 1 / N be accepted?
[0136] B. Beam splitting via a phase-only scalar grating
[0137] Next, a somewhat related question. Is it possible to split light into a finite number of diffraction orders with 100% efficiency and without loss or gain modulation? This would mean that the grating is phase-only,
[0138] It is possible to show that a phase-only grating can have (only) one or infinitely many diffraction orders, and therefore a phase-only grating can never be a solution for the general beam-splitting scenario described above. This can be proved here. The proof is by contradiction.
[0139] In the following, it may be assumed that the grating is illuminated by a unit amplitude plane wave, so that the field following the grating is also Let the grating order coefficients (Fourier coefficients) of the periodic grating be the set {a k} is given. In addition, suppose a k = 0 lies between k = m and k = n, where n>m. Assume nm = N. Then the phase-only grating can be written as the Fourier expansion
[0140]
[0141] Since only the phase is
[0142]
[0143] Obviously, the sum can have no amplitude modulation (it can be independent of x). All exponential terms with kl ≠ 0 must cancel. The system of equations can be written as:
[0144]
[0145] There are N+1 conditions that must / can be satisfied. The first equation ensures that the highest order exponent does not contribute to the sum (which would / can be the case as e iNx The next equation covers terms with exponents of order N-1, and so on. Finally, at the basis of the system, an equation for the "first order" exponent is obtained, where the "nearest neighbor" grating order coefficients are multiplied together. The last equation ensures that the DC term has unit magnitude; this can be interpreted as a statement of conservation of energy.
[0146] Figure 6A shows a schematic diagram of diffraction by a diffraction grating to a finite number of orders between k = m and k = n. Figure 6B shows that each equation in equation S20 checks all pairs of orders separated by a given integer distance, starting with N in the first equation and proceeding downward, with the base case being the nearest neighbor diffraction order.
[0147] The first equation implies that the weight a of the outermost diffraction order is m or a n One of must / can be zero. Suppose you want to select |a m |=0. The next equation implies |a m-1 |=0, next is |a m-2 |=0. As the equation is traced back, we can eventually conclude that except for a n All coefficients except for , must / can be zero. Therefore, for a phase-only grating (if a finite number of terms is assumed), only one can have an intensity different from 0 (there can also be an infinite number of diffraction orders). This is the common case for blazed gratings with a linear phase distribution in x. A phase-only grating can also have an infinite number of diffraction orders, with a k is not zero; this case is not covered by the above proof because a finite number of orders between k=n and k=m is assumed.
[0148] C. Optimization of Phase Grating Only
[0149] The above proof shows that beam splitting with a diffraction grating can never be 100% efficient. But it does not answer the question of whether it is possible to do better than 1 / N.
[0150] It stands to reason that since 100% of the incident light passes through the phase-only grating, most of the incident light can go into a limited set of diffraction orders, while some leaks out into all the other diffraction orders. What is the best thing a phase-only grating can do to focus as much light as possible into a limited set of desired diffraction orders while minimizing this leakage?
[0151] First, state the problem: given a set of diffraction orders whose indices are given in the set {l}, find the function with period d Make efficiency function maximize:
[0152]
[0153] In the coefficient a k The magnitude of and the powers of the diffraction orders are subject to the desired relative proportions. k}Persistence|a k |=cγ k . {a k} is given by equation S1.
[0154] This is essentially a problem of variational calculus. With the method of Lagrange multipliers (solving the constraints), it can be shown that the optimal phase function has the form
[0155]
[0156] (Table S1)
[0157]
[0158] Table S1: Diffraction efficiency (η) of gratings implementing the optimization defined by equations S8 and S9 for only phase gratings phase ) and implement only loss modulation (η) as a function of N loss =1 / N, such as Figure 5 ), where N is the number of diffraction orders into which the light is directed. For simplicity, this assumes that the intensities of these orders are equal.
[0159] This gives the optimal analytical form. The weight μ k and stage α k must / can be determined based on the constraints of the original problem. Romero & Dickey showed that this leads to a system of 2N-3 (nonlinear) equations, where N is the number of diffraction orders to be controlled, which can be solved numerically using e.g. Newton's method.
[0160] Equations S8 and S9 are noteworthy because they solve the problem of finding the optimal continuous grating phase function , which is an infinite-dimensional problem (the function is defined at every point in the unit cell), and reduces it to dimensions 2N-3, which is numerically tractable.
[0161] In Table S1, for the problem of equally splitting light into N diffraction orders, the efficiency η obtained by using only the phase grating optimized using this method can be reported as phase The efficiency η of the loss-only modulation loss = 1 / N, as discussed above. Even though there is no general evidence that η phase >η loss N, inspection of Table S1 also suggests that this is the case, and this is intuitively appealing because phase-only gratings are lossless. Compared to gratings designed with more direct methods, phase-only gratings appear to be more successful at splitting light. Moreover, the phase-only approach is more compatible with readily accessible grating implementations such as surface relief gratings and more recently metasurfaces.
[0162] D. Parallel Polarization Analysis via Tensor Gratings
[0163] Here, the diffraction gratings that implement polarization analyzers (polarizers) on diffraction orders can be examined using the tensor methods introduced in the main text. These gratings are the focus of the main text. This section will be closely related to the development of Section S1A. On the one hand, it is desired to split the light into (only) N diffraction orders. In this section, it is desired to do the same thing, but keeping the polarization in mind. In particular, the light of order k should have the intensity prescribed by Malus's law, as indicated by its preferred polarization |q k > prescribed.
[0164] More specifically, a set of diffraction orders {l} can be specified. For each order k∈{l}, its tensor Fourier coefficient J can be specified k The following format should be used in is a scalar weight, |q k > is the Jones vector of the polarization of the order analysis, and |p k > is the output polarization of the order, as described in the main text.
[0165] For any choice of these parameters, you can write directly according to the method in the text
[0166]
[0167] (To correspond to the main text, this disclosure projects this problem into 2D, but the same arguments will hold in 1D.) By mathematical construction, this grating will / can have exactly N orders, each of which behaves as a forbidden polarization state {|q k >}. At this point in Section S1A, we calculate t(x) by Required power. Matrix quantity What is the equivalent of? The equivalent transmission is given by
[0168]
[0169] Here, Tr is the matrix trace. This T is the power transfer averaged over all possible incident polarizations on the Poincaré sphere. In fact, T is an expression for the top-left element of the Mueller matrix M corresponding to J - which makes sense, since this element of the Mueller matrix controls the intensity transfer. It is also easy to see that if J is unitary (i.e., if ), then T = 1, which is consistent with the definition of a unitary matrix. Next, analogous to the two beam splitting examples in Section S1A, we can calculate for some specific examples using Formula S10 and examine the variation of T on the unit cell of the grating that it involves. As examples of these, two 2D examples can be chosen that are analogous to the gratings considered in the main text. In particular:
[0170] Ideal tetrahedral grating: This grating sends all light into four diffraction orders. Each diffraction order acts as an analyzer for one of the four Jones vectors, which corresponds to the Stokes polarization state of a tetrahedron inscribed in the Poincaré sphere, with order {l} = {(0,-1), (0,1), (1,0), (0,-1)}. This is described in the main text. Figure 2A is shown in the upper left corner and is Figure 7 Reappear in. Figure 7 The perfect 2D tensor grating analysis for a tetrahedron (top) or octahedron (bottom) for the polarization states is shown for each of the two cases in the main text and is synthesized with the aid of equation S10 - these polarization states and their corresponding diffraction orders are shown on the left, from the main text Figure 2A Each grating produces a periodic 2D unit cell This can be analyzed to produce the three graphs on the right. From left to right, they are T (polarized average intensity transmittance given by equation S11, with the flat surface showing unity), the desired attenuation, and the desired blocking. In a sense, this graph is similar to the matrix space in Figure 5 The orientation of the tetrahedron is not unique, but in the present disclosure, one vertex can be chosen to be at the north pole of the sphere, corresponding to circularly polarized light, and the other vertex is aligned along the S1 axis of the Poincaré sphere. These two choices are sufficient to constrain the exact polarization state of the tetrahedron.
[0171] Ideal octahedral grating: This grating sends all light into six diffraction orders. Each diffraction order acts as an analyzer for one of the six Jones vectors, corresponding to the Stokes polarization states of the octahedron inscribed in the Poincaré sphere, with the orders {l} = (0,-1), (0,1), (1,0), (0,-1), (1,1), (-1,-1). These polarization states are x>, |y>, |45>, |135>, |R>, and |L>|, perhaps the most familiar polarization ellipses. This is discussed in the text. Figure 2A is shown in the lower left corner and Figure 7 Reappear in.
[0172] In the following, one will / can assume that for each order of these gratings, in addition to flipping the chirality, The output polarization of the diffraction order is the same as the polarization being analyzed. This limitation is implicit in the implementation of by using a linear birefringent waveplate-like element. This will / can be demonstrated below, but it is possible to work with this assumption here already.
[0173] For each of the above two cases, a sum can be formed as suggested by the Fourier expansion of equation S10. The last piece of this sum is the weight coefficient As will be explained below, adherence to global energy conservation and equal intensity diffraction orders means that in the tetrahedral case all All should be In the case of an octahedron, it should be Next, these required spatially dependent transmissions can be examined (see Eq. S11). For the perfect tetrahedral and octahedral cases, this is already Figure 7 For each of these examples, the polarization-averaged transmittance is plotted. and represents a uniform flat surface. It can be shown that in both examples
[0174]
[0175] where Ω represents the unit cell of the grating, which means that energy is conserved overall. This is achieved by choosing to frame this problem, as discussed in the next section, much like the scalar case above, where overall energy conservation is ensured by energy conservation in the Fourier series. However, by checking Figure 7 From the curves in , we can see that T is not unity everywhere in the unit cell. Just as in the simple scalar grating described above, amplitude gain is required in some places and loss in others, in such a way that the overall intensity is preserved but requires a redistribution of the intensity.
[0176] So far, this tensor diffraction example has been completely analogous to the simple scalar example of Section S1A. In that discussion, the solution proposed for this problem was to lose only the modulation. This concept can be extended to this case - if Scale down
[0177]
[0178] Each order will / can act as an analyzer for the preferred polarization specified in the design without requiring local gain, at the expense of overall efficiency scaling by the same factor.
[0179] However, these tensor examples are not similar in all respects to the simple scalar case of Section S1A. As a tensor, the full 2×2 matrix quantity must / can be implemented at each point of the raster unit cell, so the scalar quantity T does not paint the complete picture. In other words, the required The polarization transformation properties of the diffraction order can be customized to implement the desired function at each diffraction order.
[0180] This is also Figure 7 . For each grating, the desired polarization bidirectional attenuation (unitless ratio) and retardation (angle) are plotted at each point in the unit cell. Polarization bidirectional attenuation is the degree to which a polarization element acts like a polarizer. A polarization bidirectional attenuation of 1 means perfect transmission for one preferred polarization and perfect extinction for the other. Retardation is a waveplate-like characteristic of the Jones matrix that quantifies the amount of delay of one polarization relative to the other. A general Jones matrix operator can have both of these properties at the same time, and not necessarily along the same axis. These quantities vary significantly and rapidly from point to point across the unit cell. It can be seen that in order to directly implement the Jones matrix corresponding to Equation S10, a medium that can control the overall transmission amplitude as well as the precise polarization properties is strictly required at each point.
[0181] The perfection required by the expansion in equation S10 then requires flexible, complex control of polarization and amplitude in a manner that varies rapidly with space over a periodic unit cell. This presents challenges for practical implementation in real materials and real structures that are beyond the realm of mathematics. In the scalar example of Section S1A, the present disclosure turns at this point to phase-only gratings, which do not solve the problem perfectly (because some light necessarily leaks into adjacent diffraction orders) but can still be implemented with real diffraction gratings and can be shown to implement high efficiency in optimized cases.
[0182] Is there an equivalent in this tensor example, i.e., a corresponding to the raster that is easy to implement in practice Mathematical simplification?
[0183] E. Parallel Polarization Analysis via Linear Birefringence Tensor Gratings
[0184] In the next section, we consider tensor gratings, where Unitary matrices constrained to act as linear birefringence - that is, as simple wave plates - are ubiquitous. These kinds of structures can be simply implemented with metasurfaces consisting of dielectric pillars that exhibit formal birefringence. If the dielectric has a high refractive index contrast with the surrounding medium, then it can confine light inside it. If the transverse cross-section of the pillar has two perpendicular mirror symmetry axes (e.g., the rectangular case), then two modes will / can propagate in the pillar, and due to the anisotropy, these two modes will / can propagate with different indices. Therefore, the pillar can behave as a wave plate-like element, whose Jones matrix can be written as
[0185]
[0186] Here, R(θ) is a 2×2 rotation matrix, and the angle θ corresponds to the physical angle of rotation of the pillar element. The phase can be continuously and independently controlled by adjusting the lateral dimensions of the cross section of the pillar. and The column dimensions (w x ,w y ) and the phases they assign library, and the rotation angle θ can obviously take any value between 0° and 360°. Therefore, the present disclosure does not elaborate on this point here, but only says that if Everywhere the equations S14 are in the form of tensors, there are convenient, established media for their practical implementation.
[0187] Thus, in some embodiments, the Jones matrix of the tensor raster may be maintained should be of the form of Equation S14 everywhere. As described in the text, Equation S14 implies two specific restrictions:
[0188] must / can be unitary everywhere. This implies that the local Jones matrix can modify the polarization state, but not the intensity. A unitary matrix can retard the components of the Jones matrix relative to each other.
[0189] More specifically, Must / can be linearly birefringent. The eigenvectors of a unitary matrix are strictly orthogonal. However, these eigenvectors can be linear, circular, or elliptical - this requirement implies that they must / can always be linear.
[0190] 1. Linear birefringence hints
[0191] It is shown that equation S14 will / can always describe θ, and Any choice of a symmetric matrix is simple, since the eigenvalue basis of J will / can always be a rotated version of the x / y system. In other words, the fact that J is linearly birefringent means that it is also symmetric in the x / y polarization basis (the basis in which equation S14 is written). If Everywhere it is of the form of equation S14, It is also true that if you add together many matrices that are themselves symmetric, the resulting sum will / can be symmetric as well.
[0192] A grating formed by a spatially varying Jones matrix has diffraction orders whose behavior can be described by the Jones matrix, which is the number of diffraction orders on the grating unit cell. The Fourier transform of this is a weighted sum of symmetric matrices - if Symmetric everywhere, then all J k , the Jones matrix that governs the behavior of the diffraction orders must / can also be symmetric everywhere.
[0193] If J k One of the two takes the form of an analyzer, i.e. if in some embodiments, if the order acts like a polarizer, then it must / can be symmetric. Under what conditions is the analyzer matrix symmetric? The analyzer Jones matrix is given by Given. The output polarization is parameterized as |p> = [p1p2] T The polarization parameter to be analyzed is |q>=[q1q2] T , you can have
[0194]
[0195] By checking, if or if and - In short, if Then this J is symmetrical.
[0196] 2. Relationship with comparative work on “metasurface polarization optics”
[0197] On the Poincaré sphere, this means that k >The corresponding vector must / can be relative to |q k >Mirrored about the equator, or physically, its polarization ellipse must / can be the same shape as the electric field rotated in the opposite direction ("flipped handedness").
[0198] This conclusion has been identified in comparative work, where the goal was to impart two independent phase distributions on an arbitrary basis of polarization states. That work did not operate in a framework as general as this one, but rather, using a scalar approach, designed gratings where the output polarization is uniform when one of the polarization basis vectors is incident. That work can be performed in the tensor framework proposed here: if the polarization basis of interest is given by |λ + > and |λ - > given, <λ + |λ - >=0, then the spatially varying Jones matrices of those jobs are given by
[0199]
[0200] in and is the polarization basis λ + and λ -The two phase profiles of interest are experienced by the 0th order diffraction. Equation S16 shows that any polarization basis can experience independent phase profiles, which is a subcase of the tensor method of some embodiments. Therefore, it is proved that It is more general for the analyzer produced in the diffraction pattern of the linear birefringence grating given here. 3. Unitary linear birefringence grating for polarization analysis can have only two orders as analyzer
[0201] In the scalar regime, it was possible to demonstrate that phase-only gratings can have either one or infinite diffraction orders. For two specific examples (octahedral and tetrahedral gratings), complex amplitude and polarization control is used for the exact solution, packaged in Figure 7 Could a linear birefringent tensor grating that can be easily implemented in practice be a solution to the problem? Could a tensor grating direct all light to a finite number of orders, whose Fourier matrix coefficients J k Both behave as polarization state analyzers?
[0202] This continues similarly to Section S1A. Here, we assume a 1D tensor grating given by J(x) (this can be generalized to 2D). We assume that all non-zero diffraction orders of the grating lie between k=m and k=n, where n>m and nm=N. Each grating order k has the operator J k , you can stick with the analyzer form, so you can have
[0203]
[0204] Then, the scalar weight set The set of output polarization states {|p k >} and the analyzer polarization state set {|q k >} to define a tensor raster If we can hold that the grating is linearly birefringent, then Now, the Fourier series for J(x) can be written as
[0205]
[0206] This grating must / can be unitary everywhere, so the product can be formed
[0207]
[0208] Here 1 is the 2×2 identity matrix. As before, if the matrix product is independent of the spatial coordinate x, then all exponential terms with kl ≠ 0 must / can cancel. This allows the sum to be regrouped as
[0209]
[0210] Each equation in equation S20 examines all pairs of orders separated by a given integer distance, starting with N in the first equation and working downwards, with the base case being the nearest neighbor diffraction order. The gas-imposed equation S20 is really no more complex than equation S6 for a phase-only grating, except that it involves matrix quantities, adding a lot of mathematical notation. Each equation in equation S20 is a coefficient of an exponential in the expanded equation S19 and can again be understood with the aid of Figure 6. At the top level, one can consider the terms involving the farthest neighbors k=m and k=n, and then work downwards to consider the base case of the nearest neighbor terms due to the values. Fig. 8A This is illustrated graphically in the schematic diagram in , which shows the parallel polarization analysis of a linear birefringent tensor grating, using the same notation introduced in this section. Each order from k = m to k = n implements its own polarization analyzer matrix J k .
[0211] Before considering the full complexity of the system equation S20, it is worth mentioning the bottom line (reproduced here):
[0212]
[0213] This is the expanded "DC" term. Since this term has no spatial correlation, it must / can explain the entire sum. This equation is the matrix representation of Parseval's equation in Fourier theory and is a representation of energy conservation in tensor gratings. Outer Product |q k > k | can be rewritten as
[0214]
[0215] Here, is the Jones vector |q k >The corresponding Stokes state of polarization (SOP). The SOP is the last three elements of the Stokes vector and gives |q k >'The coordinates of the polarization state on the Poincaré sphere. The vector is a vector where σ is the k-th Pauli matrix. Assume that all |q k > is normalized so that From equations S21 and S22 we can write
[0216]
[0217] Its terms can be separated into equations
[0218]
[0219] and
[0220]
[0221] in is a 2×2 zero matrix
[0222] Equation S24 is a direct expression of the energy conservation of this problem. In the case where all orders have equal strength (i.e., all in is a constant), equation S24 implies
[0223]
[0224] where N is the number of diffraction orders acting as polarization analyzers, where For each order in this iso-intensity case, |a| is the percentage of incident power directed to that order when incident with its preferred polarization and provides the I defined in the text q It is pointed out that for tetrahedral gratings with N = 4 and octahedral gratings with N = 6, any order of I q The maximum values of are 1 / 2 and 1 / 3 respectively. This is derived from equation S24.
[0225] Turning next to equation S25, it can be further simplified to yield
[0226]
[0227] is the zero vector of the three elements. This is a symmetry condition. It says that by its peak intensity Weighted Analyzer Stokes Vector The "center of mass" of must / can be located at the center of the Poincaré sphere. In the case of equal strength, the vector Must / can be symmetrically distributed on the Poincaré sphere. Figure 8B The set of vectors that satisfy this condition is shown in Several examples showing some analyzer configurations that meet the symmetry condition equation S27 Examples include tetrahedrons of some embodiments. This makes intuitive sense: if incident light is to be directed into a finite number of orders, all of which behave as polarization state analyzers, then the states analyzed by each order must / can be compatible with each other. If for some incident polarizations one diffracted order is dark, then the |q k >It must / can be ensured that the missing light is directed there. These power and symmetry conditions equations S24 and S27 are prerequisites that must / can be met.
[0228] After examining only the DC term, we can return to the rest of equation S20. The top level equation N, It is responsible for the index e iN x, requirements (If orders m and n have non-zero weights Equivalently, m |q n >=0 and the polarization state analyzed by the outermost diffraction order |q m >and|q n > must / can be orthogonal.
[0229] The next equation, N-1, solves for the term e i(N-1)x Read
[0230]
[0231] The sum of the two matrix quantities is specified to be 0. In order to avoid requiring |a n-1 |=0 and |a m+1 There are two possibilities for implementing this when |=0:
[0232] 1. and means |q n-1 >=|q n >and|q m+1 >=|q m >(Because the previous equation requires m |q n >=0).
[0233] 2. Matrix |q m > n-1 |and|q m+1 > n | describes parallel matrices (i.e., scaled versions of each other so that they add up to 0 without being all zero). Since all {|q k >} is normalized, so this means |q n-1 >=|q n >and|q m+1 >=|q m >.
[0234] Both scenarios require |q n-1 >=|q n >and|q m+1 >=|q m >, so we can accept this as a solution condition. We can then proceed to the next equation, which weights the exponentials of order N–2:
[0235]
[0236] Due to |q n-1 >=|q n >and|q m+1 >=|q m > and m |q n >=0, so and the middle terms can be eliminated. Now, the equation is exactly the same as the previous one, and we can get |q n-2 >=|q n >and|q m+2 >=|q m >Conclusion.
[0237] The equations proceed in this way, with each equation successively requiring that the diffraction order increasing by one index from n inward must / can have its |q k >=|q n >. And from m to the inside, add another one that must / can have its |q k >=|q m >. To satisfy equation S20, the diffraction orders from the left must / can all have analyzer polarization |q m >, while the diffraction orders from the right must / can all have |q n >. Eventually, the left and right halves of the diffraction pattern meet.
[0238] Fig. 9 shows how a contradiction prevents the solution of equation S20 from occurring at the midpoint of the diffraction pattern. The first equation requires that the outermost diffraction orders be polarized with orthogonal analyzers |q n >and|q m >. Each successive equation requires that a diffraction order closer to the center from the left or right side present |q m > or |q n > (arrow, top). Where the equation S20 deals with terms halfway across the diffraction pattern, two terms remain due to the orthogonality of the analyzer vectors for the left and right halves of the diffraction pattern (arrow, center). In the case where N is odd, this equation requires the central order to have an analyzer polarization state |q that exhibits two states simultaneously cent >(contradiction). When N is even, the previous equation requires the inner two orders |q + >and|q - >Yes|q n >and|q m >, but the next equation will require the opposite (also a contradiction). m |q n >=0, this midpoint contradiction can be avoided.
[0239] In the case where N is odd, a unique “central” diffraction order is achieved, whose analyzer polarization state |q center > does not clearly belong to the right or left half of the diffraction pattern. By examining the portion of equation S20 involving diffraction orders that are (N-1) / 2 integers apart, we have the equation that simplifies to
[0240]
[0241] Since the right and left halves of the diffraction pattern ( Fig. 9 , left) with orthogonal polarization |q m >and|q n >, so these terms can be retained. By inspection, equation S30 requires a contradiction. Its solution would require |q center > while being equal to |q n >and|q m >, given that the first equation states n |q m >=0, this is impossible.
[0242] If instead N is even ( Fig. 9 , right), then there is no clear "central" order, but rather a pair of central orders. In between is |q k >=|q n >and|q k >=|q m > where the regions meet. Since there is no central order, there does not appear to be a contradiction in this case. However, by examining the portion of equation S20 involving orders separated by N / 2-1, an equation of the form
[0243]
[0244] Here, the diffraction orders - and + are the diffraction orders to the left and right of the center, respectively, as Fig. 9 (there is no central order because N is even). All other terms in equation S31 are 0 because m |q n >=0, and the left and right halves must / can use outer order |q m >and|q n > analyzer polarization. The same reasoning means that equation S31 contains a contradiction: working from the outside in, an equation with an exponential term that is one higher than the exponential term of equation 31 requires |q - >=|q m >and|q + >=|q n >, but equation S31 is just the opposite, |q - >=|q n >and|q+ >=|q m >. Once again, we encounter a contradiction at the midpoint intersection of the diffraction pattern.
[0245] If there are any more than two with non-zero order, then this midpoint crossing problem described in Figure S5 will / can be encountered. Essentially, this stems from the mathematical fact that only two different polarization states can be orthogonal to each other. However, if there are only two polarization states acting as |q m >and|q m The diffraction orders of the analyzer of the orthogonal basis of , where q n |q m 》=0, if these orders =1 / 2 (as required by the power and symmetry conditions Equations S24 and S27), then Equation S20 is always satisfied. It can be claimed that this derivation only considers the case where all diffraction orders in the set between k=m and k=n have non-zero intensity or Is it possible to selectively "skip" some diffraction orders (allowing some with |a k |=0) to avoid the contradiction that occurs at the midpoint? However, by skipping some orders, the position of the midpoint is simply moved, and the contradiction still occurs. If all intermediate orders are "skipped" and only two orders are left, then a solution can be obtained.
[0246] So we can show by contradiction that if the birefringence is linear (if the grating is not restricted to linear birefringence, but just unitary, then there is no The requirement of q is invalid and the argument presented here fails. There are multiple orders that can act as analyzers.) The unitary directs all incident light to a finite set of diffraction orders that act as polarization state analyzers, then there can be only two such orders. These orders must / can have equal intensity and must / can analyze orthogonal polarizations and make q n |q m 》=0.
[0247] S2. Optimization-based Design of Tensor Gratings for Parallel Polarization Analysis
[0248] The basic conclusion of Section S1 is that its straightforward implementation will involve complex spatially modulated amplitude and polarization behavior. Similar to the phase-only grating in the scalar case, it is easy to find a solution that is unitary everywhere and linearly birefringent, since this has a very simple, proven practical implementation in the form of a dielectric metasurface that implements the ubiquitous Jones matrix of the form Eq. S14. It is also demonstrated in Section S1 that a linearly birefringent, unitary grating will never perfectly be the sought solution - if all light is directed into a finite set of orders that behave as analyzers (polarizers), then there can be (only) two such orders and they must / can analyze orthogonal polarizations, the latter constraint coming from energy conservation.
[0249] Similarly, in the beam-splitting problem, a phase-only grating can be shown to - mathematically - produce (only) one or an infinite number of non-zero diffraction orders (see Section S1B). If one still accepts an imperfect solution, with some light leaking into unwanted orders, an optimized solution with high efficiency can still be obtained (Section S1C). The same logic is easily followed here. Can an optimized unitary and linear birefringent grating efficiently implement an analyzer for the desired polarization state on a finite set of diffraction orders, minimizing leakage into other diffraction orders?
[0250] In the phase-only case, this optimization can be constructed as an optimization under constraints and an analytical solution to the global optimization found using variational computations (Section S1C). In theory, a similar approach applied to the (more complex) tensor case at hand is also possible. Alternatively, a simple numerical scheme can be concocted to find the optimal solution. It is first acknowledged that the scheme used here is not ideal - it may (only) produce local maxima and does not necessarily produce the optimal solution in a computationally complex way. Instead, what is attempted to be shown here is that knowledge of the tensor image of the diffraction allows gratings to have almost arbitrarily specified functions, which allows straightforward design and implementation.
[0251] A. Development of optimization solutions
[0252] 1. Parameterization of grating unit cells
[0253] In some embodiments, rectangular unit cells may be of interest (indeed, squares in the experimentally presented gratings). These gratings will be implemented with a discrete grid of structures, whose dimensions are represented as A x B, where A and B are integers. The discrete elements (a, b) in the grating will / can have a local Jones matrix J in the form of Equation S14. The local J can then be defined by three parameters at each location: (the phase of light polarized along the local x-axis of the element), (the phase of light polarized along the element's local y-axis), and θ (the angular orientation of the element). Each of these can be represented by a continuously varying function on the grating unit cell: , are parameterized for optimization purposes and then sampled at the center of each position (a,b). Instead, the spatially discrete nature of the final metasurface implementation can be acknowledged by parameterizing these three angles with a single number at each position in the A x B unit cell. This means there will / can be a total of A x B x 3 parameters to optimize, three for each discrete polarization element of the grating.
[0254] At each discrete position (a, b) of the grating unit cell, the Jones matrix J (a,b) Given by the following formula
[0255]
[0256] Here, θ (a,b) , and is the grating parameter at the (a,b)th position in the grating unit cell, and R is the 2×2 rotation matrix. Then, the spatially dependent periodic Jones matrix can be written as
[0257]
[0258] where d a and d b The separation between elements along the x and y directions, respectively, and rect da,db (xa·d a ,yb·d b ) is around the point (a·d a ,b·d b ) a ×d b A function that is equal to 1 in the rectangle and equal to 0 elsewhere. By using this rect box function in the parameterized grating, it can be assumed that light above each polarization element experiences a uniform polarization element and that the transition between adjacent elements is instantaneous. A more realistic model can be developed taking into account the radiation pattern of actual metasurface polarization elements, which is not attempted here.
[0259] Symbol III A·d,B·d It means that A·d is the value on x and y respectively. a and B·d b The infinite 2D array of Dirac delta functions separated by the size of a unit cell. A·d a,B·db gives the convolution of the expression in brackets with respect to one unit cell of the grating (in terms of d) is a way of expressing the infinite periodicity of the grating in 2D.
[0260] Given by {θ (a,b)}、 and (the set of these parameters at each discrete location in the unit cell) a fully defined grating may be (m,n) Interested, among them J (m,n) is the (m,n)th coefficient in the Fourier series of the periodic function J(x,y). (m,n) The polarization-dependent behavior of the diffraction orders (m,n) is given. (m,n) is given by the projection of J(x,y) onto the (m,n)th grating order. Applying the convolution theorem to equation S33, this is given by
[0261]
[0262] Where sinc(x)=sin(πx) / (πx).
[0263] Given a set of parameters {θ (a,b)}、 and Then the Fourier Jones matrix coefficients J can be calculated for any (m,n) (m,n) This is Fig.10 As shown in Fig.10 As shown in , at each discrete position (a, b) in an A x B grating unit cell (here shown as an 11 x 11 square grating), it can be assumed that there are fixed values θ, and These are the parameters to be optimized. Together, these parameters define the Jones matrix on the grating and the Jones matrix of the diffraction order {J (m,n)}, that is, the Fourier coefficients.
[0264] 2. Optimization
[0265] Equation S34 allows for the J of the grating to be (m,n) We can perform analytical calculations and know the A x B x 3 parameters that define the grating at each point. Next, we can seek to solve the inverse problem, that is, find the solution that makes J (m,n) The grating parameters behave in some way. First, the optimization problem solved here can be stated:
[0266] Problem. Given a set of grating orders (m,n)∈{l} and a corresponding set of Jones vectors {|q m,n >}, these orders should act as polarization state analyzers for these vectors, finding the parameters {θ that maximize the energy transfer to order {l} at each position (a, b) of the A×B unitary linear birefringent grating (a,b)}、 and (minimizing leakage to all other orders), while enforcing the requirement that they act as analyzers for the desired polarization.
[0267] Fig.11 An overview of the optimization-based design flow for gratings for some embodiments described in Section S2A is shown.
[0268] First, in order to solve the above problems, we can define the quantity
[0269]
[0270] I q (m,n) In its preferred polarization state Gives the optical power of order (m,n) at the time of incidence (normalized to the incident power). You can also define the polarization state It is the polarization state orthogonal to the polarization state being analyzed, such that This allows to define the strength:
[0271]
[0272] I q,⊥ (m,n) Given The optical power at the incident time step (m,n) (normalized to the incident power); ideally, this will be 0. Next, we can define the contrast at the step (m,n) as
[0273]
[0274] Ideally, according to Malus's law, this will be 1 for each order of interest, so that the preferred polarization is perfectly transmitted and the orthogonal polarizations is eliminated. Finally, the merit function can be defined
[0275]
[0276] Gradient descent optimization is performed on the 3 parameters A x B x to maximize the merit function Equation S38. This is done under the constraints (using the method of Lagrange multipliers), namely
[0277]
[0278] and
[0279]
[0280] where σ represents the standard deviation operator. In other words, the optimization merit function equation S38 maximizes the total energy directed to the diffraction order of interest and minimizes the energy leaking to other orders. The first constraint equation S39 forces the J of the order of interest to be maximally proportional by maximizing the contrast of its desired polarization state. (m,n) Behavior of the polarization state analyzer with respect to the desired polarization state. The second constraint equation S40 enforces that a situation will / will not occur where one diffraction order is particularly strong while the other diffraction orders are very weak, thereby encouraging uniform energy distribution among the diffraction orders of interest.
[0281] Once the parameters of the unit cell have been optimized, J(x,y) can be constructed by equation S33 and the diffraction order J (m,n)The polarization behavior of is checked by equation S34. Note that this procedure is a purely numerical scheme - it is not specific to wavelength or any real optical material. This representation of the grating does not contain any physics other than the concept of plane wave expansion.
[0282] B. Results for an 11-element grating
[0283] 1. Progress of optimization
[0284] 2. Optimization results
[0285] Optimization produces {θ (a,b)}、 and These can be converted into the space-varying Jones matrix by equation S33 and the Fourier coefficient matrix J (m,n) can be found using Equation S34. This can be done for many grating orders at once. For a particular incident polarization |λ), the power of each order can be calculated as This allows plotting the diffraction pattern predicted for the incident polarization |λ〉. For the optimized tetrahedral grating, this has been Fig.12 The analysis is complete. Fig.12 The calculated far fields for a mathematically optimized 11 x 11 tetrahedral grating are shown when four polarizations of interest are incident. For each diffraction order, the fraction of the incident power directed there is given. It can be seen that in each case one order is very bright, with most of the remaining intensity distributed among the other three orders as desired. The figures show the innermost 64 orders. Fig.12 Each of the four graphs in {|q (m,n) >} corresponds to the diffraction patterns observed when each of the four preferred polarizations in the tetrahedron is incident (the inner 64 diffraction orders). It can be seen that in each case the corresponding diffraction order is very bright, with most of the remaining intensity equally distributed among the other three design orders. This distribution is expected for a tetrahedron because its symmetry means that the projection of one vertex onto any other vertex is the same. It can be seen that more than 40% of the incident light is directed to the diffraction order when it is incident with its preferred polarization (recall equation S26, which has a maximum value of 50% in the tetrahedron case). Leakage to orders beyond the central four is minimized.
[0286] If one wishes to assign a single efficiency figure to the optimized grating, this can be done by calculating the fraction of the incident power contained in the diffraction order of interest as a function of the incident polarization. This was done for the optimized 11-element tetrahedral and octahedral gratings, and the results are plotted as Fig.13 as a function of the angle of incidence on the Poincaré sphere. Fig.13It is shown that mathematically optimized 11x11 tetrahedral and octahedral gratings are used to calculate the fraction of power confined to the diffraction orders of interest, as a function of the incident polarization state. The results are shown on the Poincaré sphere (two views are shown for each case for clarity). In both cases, over 80% of the incident power is confined to the desired order for all possible incident polarization states. It can be seen that over 82% (or 87%) of the power will / can be directed into the order of interest in the tetrahedral (or octahedral) case for all possible incident polarizations. Such high predictions demonstrate the success of this very simple optimization scheme and show that simple linear birefringent metasurface gratings are well suited for this task, even if they cannot reach perfection.
[0287] C. Transition to metasurface design
[0288] Once the grating has been optimized via the scheme of Section S2A, for each position in the unit cell of the diffraction grating, we have the optimized set {θ (a,b)}、 and At each location, implement the desired and The geometry of the physical element can be located at the material platform and wavelength of interest (in some embodiments, TiO2 and approximately 532nm) and rotated by an angle θ. If many elements are simulated using full-wave numerical simulation (e.g., using RCWA or FDTD techniques), then a library of elements can be assembled, where and the overall transport along x and y (t x and t y ) is known for each element geometry. Given the required Yes, as a measure of the distance between the expected and actual component performance in complex space
[0289]
[0290] This can be calculated for each element in the library. The element that minimizes this distance is selected. This can be done for each site in the grating unit cell, and the grating unit cell can be subdivided multiple times in the x and y directions to create the final grating.
[0291] S3. Experimental Characterization of Gratings
[0292] An important part of some embodiments is the claim that the specific behavior encapsulated in the Jones matrix can be embedded into the orders of diffraction gratings, allowing for arbitrary polarization-dependent functionality in parallel. In some embodiments, gratings whose orders act like polarizers or polarization analyzers may be of interest. Section S1 examines the mathematical limits of this problem, and Section S2 discusses optimization-based design solutions for these gratings. Here, the disclosure discusses how these gratings, once implemented, can be tested to verify how well the polarization-dependent behavior of each diffraction order matches the design.
[0293] Strictly speaking, this would involve the measurement of the Jones matrix for each diffraction order. However, the Jones matrix (and the Jones vector) are direct descriptions of the electric field, which contain phase information of the signal as a function of the optical frequency. As such, they do not admit of direct measurement. As a proxy, the Mueller matrix for each diffraction order can be measured, since the Mueller formalism deals with directly measurable beams.
[0294] A. Manufacturing
[0295] The gratings shown in some embodiments are based on a dielectric metasurface composed of rectangular TiO2 pillars that operate in the visible light range (certain embodiments disclosed herein implement λ = approximately 532nm). Here, the fabrication of these gratings can be described. However, note that some embodiments do not rely on any specific material, wavelength, or mode that is polarization-dependent. Any polarization-dependent obstacles may belong to the formalization of some embodiments. The gratings are first written into an electron beam resist using a 125keV electron beam lithography system. The resist is developed and the TiO2 is deposited using atomic layer deposition (ALD) to fill the gaps in the resist from the sidewalls. In addition to filling these gaps, this also produces an excess TiO2 layer on top of the pillars. The remaining oxide is etched away using a Cl2 / BCl3 chemistry. Finally, the excess electron beam resist is removed with a solvent, resulting in free-standing TiO2 pillars with a height of approximately 600nm.
[0296] exist Fig.14 In , additional SEM images of the two gratings considered in the main text are provided. Fig.14 Additional scanning electron micrographs of the octahedral (left) and tetrahedral (right) gratings discussed in the main text are shown.
[0297] B. Mueller Matrix Polarization Measurement
[0298] The process of measuring the Mueller matrix of a sample is called Mueller matrix polarimetry. For each grating order of interest (four in the case of tetrahedrons and six in the case of octahedrons), Mueller matrix polarimetry can be performed to measure the experimental Mueller matrix of order (m,n)
[0299] The process involves illuminating the grating with a laser beam in multiple polarization states and observing the polarization and intensity of the outgoing beam with a full Stokes polarimeter. If the polarization state of the incident light is known, the transformation characteristics of the grating order can be derived.
[0300] refer to Fig.15 The optical setup depicted in Fig.15 Mueller matrix polarimetry is shown: for measuring the Mueller matrix associated with the grating order (m,n) Optical setup. Fig.15 The inset shows that a full Stokes polarimeter must / can "wrap around" the sample on a spherical mount in order to see each grating order at normal incidence. A collimated shock beam (λ = approximately 532 nm) passes through a linear polarizer (LP) to set a well-defined starting polarization state. This LP is aligned with the preferential polarization of the laser mode to maximize the output power. Next, the beam encounters a half-wave plate and then a quarter-wave plate. The angle of the wave plates is oriented toward θ HWP and θ QWP is adjustable. K pairs of quarter-wave plate and half-wave plate angular orientations are selected. The wave plates access these angular orientations and the polarization state of the generated beam is recorded by a commercial full Stokes polarimeter (the Stokes vector of the beam generated in the kth orientation is expressed as This process is called calibration of the experiment, and the results are stored in the calibration matrix middle:
[0301]
[0302] Here, is a 4xK matrix. In some measurements, K=16, and thus 16 waveplate configurations are chosen so that the output beam covers the Poincaré sphere more or less uniformly.
[0303] Next, the beam is allowed to illuminate a sample of the metasurface grating. Although the grating is designed for perpendicular incident plane wave illumination, the beam first passes through a lens. Since the focal length of the lens is very long (200mm), the incident light is still a plane wave. However, the lens means that all the incident light is confined to the boundaries of the grating (at 500μm, 500μm is quite small), so a measurement of the efficiency of the grating can be made.
[0304] The full Stokes polarimeter is positioned so that it stares at the grating order (m,n). It is particularly important that the polarimeter stares at the diffraction order of interest to perform an accurate measurement. Constructing the mount (inset, Fig.15 ), which allows the polarimeter to move between diffraction orders while keeping the grating centered.
[0305] Next, the wave plate visits all K orientations again and the grating is recorded in The Stokes vector output in response to the kth input polarization at order (m,n) is recorded as And all these Stokes vectors are stored in the output matrix middle:
[0306]
[0307] again, is a 4x K matrix. Note that the Mueller matrix of the grating order The quantity you want to measure is the kth input polarization Mapping to recorded polarization
[0308]
[0309] This can be expressed for all grating orders at once in matrix form:
[0310]
[0311] Since the matrix is not square (unless K = 4), so it cannot be inverted. The (right) pseudo-inverse of represents the Hermitian conjugation. Then, one can write
[0312]
[0313] Equation S46 represents the equation contained in and The Mueller matrix of order (m,n) is determined in the least squares sense from the experimental data in .
[0314] 1. Comparison of optimized, simulated, and experimentally measured Mueller matrices
[0315] In the main text, the properties of each diffraction order are compared in three domains: from the optimized spatially varying J (obtained as described in Section S2), from full-wave simulations, and from experiments (as described above).
[0316] From the Fourier transform from the optimized analytical expression, the Jones matrix J for any desired diffraction order is obtained (m,n) .
[0317] The grating unit cell in some embodiments is simulated using a finite-difference time-domain (FDTD) simulation. A monitor is placed a few wavelengths above the grating structure. The Jones vectors for each physical diffraction order can be calculated from the far-field projection of this monitor. If two simulations are run - one for |x> polarized illumination and one for |y> polarized illumination - then the Jones matrix J can be determined (m,n) In the analog case, the Jones matrix is easy to reconstruct directly because the phase information is preserved in the numerical simulation.
[0318] However, according to the experiment, the Mueller matrix can be obtained For comparison, the Jones matrix obtained from numerical optimization and full-wave simulation can be transformed into the equivalent Mueller matrix using the constant equation
[0319]
[0320] Here, j and k are indices from 0 to 3 to cover all elements of the 4 x 4 Mueller matrix, and σ i is the ith Pauli matrix (σ0 is the 2x 2 identity matrix). Fig.16 The percentage of incident power directed into the four or six diffraction orders of interest is shown as a function of incident polarization on the Poincaré sphere of the gratings measured experimentally in some embodiments. For both gratings, this efficiency is above 45% for all possible incident polarizations. For clarity, two views are shown for each sphere.
[0321] C. Overall grating efficiency
[0322] The question of diffraction efficiency is a problem that naturally arises in some embodiments. Here the grating is polarization dependent, so "efficiency" cannot be captured by a single number, since the amount of incident light concentrated into the grating order of interest is polarization dependent.
[0323] The Mueller matrix can be obtained from the experiment The efficiency of diffraction into the order of interest can be determined by calculating the incident Stokes vector with unit power The power in order (m,n) is determined by performing the operation and taking the first element of the resulting output Stokes vector. These can be summed over the order of interest to obtain a measure of efficiency. This has been Fig.16 is done as a function of all possible incident polarizations on the Poincaré sphere. It can be seen that for both gratings the efficiency is approximately 50%.
[0324] S4. Polarization imaging
[0325] A. Optical Design of Camera System
[0326] To date, some embodiments have focused on the design of gratings that separate light based on its polarization. If the intensities of these diffraction orders are measured, the grating alone is sufficient to determine the polarization of the incident light, as a full Stokes polarimeter. This would allow the polarization state of a laser beam to be measured. For various practical applications - especially in remote sensing and machine vision - imaging polarimeters (i.e., polarization cameras) that allow the Stokes vector of the light forming each pixel of a photographic image to be characterized are of greater interest.
[0327] Conceptually, the idea here is simple. The grating presented here can split light based on its polarization state, so if the grating is incorporated into an imaging system, can four images of the scene be formed at its diffraction orders, which can then be combined pixel by pixel to find the Stokes-vector polarization image of the scene in post-processing? The question then becomes how to incorporate some embodiments of the tensor grating into an imaging system to do this.
[0328]
[0329] Table S2 shows Fig.17 The optimized values of the variable parameters of the camera are shown in . Here, ρ is the radial coordinate measured from the center of the lens in mm.
[0330] There is no one unique approach here. An effort has been made here to show that a simple full Stokes polarization imaging system can be implemented where all polarization optics are handled by a single diffraction grating, and to demonstrate that this system can be used as a compact sensor in practical applications.
[0331] To this end, a basic imaging architecture can be determined that includes only three elements - a grating sample, an off-the-shelf catalog lens, and a CMOS imaging sensor, in that order. In some embodiments, one may seek to design a camera that is infinitely conjugate, that is, it does not take an image at a given distance, but rather it takes an image of an object that is very far away (measured relative to the focal length of any optics in the imaging system), so that the beams of rays emanating from any point on the object can be considered parallel. Here, one may set a goal that the camera should image over a 10° (±5°) field of view (FOV), which is a modest goal that relieves some of the constraints while still allowing imaging of real objects. For this imaging system, the FOV is the range of angles of incidence of the beam that can be imaged onto the sensor.
[0332] The tetrahedral grating from the text is used in the camera. Its four diffraction orders implement an analyzer sufficient for full Stokes determination. Since the Stokes vector has four elements, at least four images are required for full Stokes measurement. When mounted in the camera, the grating is rotated 45° so that the orders are towards the corners of the imaging sensor, rather than deflected towards the top, bottom and sides.
[0333] Fig.17 A side view schematic of the camera considered here is shown, with all adjustable parameters labeled. The different colors here correspond to parallel beams incident on the grating sample at different angles, labeled in the key in the lower left corner. These are deflected from their original orientation by the grating before the sample (e.g. +5° becomes nearly horizontal). The grating vectors here are oriented at 45°, so that one grating step points to the corner of the imaging sensor. Not shown: a 10 nm bandpass filter at λ = approximately 532 nm in front of the imaging system.
[0334] During the optical design process, consider one grating order. If all aspects of the imaging system are symmetric about the optical axis, then the imaging system by default will / can work exactly the same for the other three orders with only a rotation of 90°, 180°, or 270°. The design goal is to take parallel beams from the entire FOV (±5°) and focus them as closely as possible to one quadrant of the imaging sensor - the end result, the camera presented in the main text, is Fig.17 As shown, there are four degrees of freedom that can be exploited in accomplishing this task (once an off-the-shelf lens is chosen). These are the distance between the grating sample and the lens, the distance between the lens backplane and the imaging sensor, the grating angle (and period) of the sample, and the overall azimuthally symmetric phase profile that can be imparted to the deployed polarization-dependent phase profile on top of the grating. This overall phase profile produces a weak lensing effect that aids imaging.
[0335] A systematic ray tracing can be performed, and these parameters varied by an optimization algorithm so that the -5°, 0°, and +5° beams (or fields, in ray tracing terms) are as focused as possible on the plane of the imaging sensor, subject to the constraint that they occupy the boundaries of one quadrant. The "goodness" of this focusing can be defined as how many rays in the beam fall within the Airy spot.
[0336] This optimization can be performed for many readily available lenses. The lens that is ultimately selected (and Fig.17) is ThorLabs part #AL2520. This is an aspheric lens with a focal length of f=20mm (listed at the time of this writing) made from high-index optical glass S-LAH64. This lens was found to produce high-quality focus throughout the entire FOV - perhaps not surprisingly, it was found that the aspheric lens had an advantage here. The complete optimized parameters for the system designed around the AL2520 aspheric surface are listed in Table S2. Curiously, although the focal length of the aspheric surface is 20mm, it can be placed only 3.9mm away from the imaging sensor in the final camera for imaging. This makes some intuitive sense: imaging does not occur at the center of the lens, but rather at a more eccentric location where the gradient is steeper, thereby reducing the apparent focal length.
[0337] Fig.17 Not shown is a 10 nm bandpass filter in front of the camera at λ = about 532 nm. In some embodiments, everything including this imaging is monochromatic. Gratings are inherently dispersive and including color complicates some embodiments. Note however that if an imaging sensor with color filters is used, then the methods presented here can be used for spectroscopic polarimetry or color polarimetric imaging (different colors will form different images due to the dispersive properties of the grating).
[0338] Fig.17 The second element not depicted in the figure is an aperture whose size can be adjusted, which is placed at a certain distance in front of the grating. This is used to control the FOV of the camera so that the images do not overlap on the sensor, and explains why the polarization images presented here are circular. Note that the use of this aperture is different from the aperture in traditional photographic imaging systems, as it is not in the Fourier plane and does not control the f-number of the system. The aperture here may cause some mechanical vignetting, which is not addressed in detail here.
[0339] A 12-bit CMOS sensor is used here to increase the dynamic range of the system, which is particularly helpful for indoor imaging.
[0340] The simplicity of a system implemented by such a grating and the practical results it produces (no traditional polarization optics) can be advantageous.
[0341] B. Angle-dependent calibration
[0342] The output of the camera is four copies of the image, each analyzed along a different polarization. What are these polarization states? These must / can be known to produce a pixel-level polarization image Each pixel of the polarization image is actually a superpixel of four pixels (how to group and identify these pixels is called image registration, and will / can be addressed in the next section). The intensity at a pixel can be written as the vector where I1, I2, I3, and I4 are the pixel intensities from each sub-image. These intensities are the polarization of the incident light. The result k projected onto the characteristic polarization of quadrant k is, These characteristic polarizations can be grouped into the matrix So we can write
[0343]
[0344] In polarization measurements, is called the instrument matrix. If known, then it is straightforward to write Confirm this This process is called polarization calibration.
[0345] The calibration of the polarization camera presented in some embodiments means determining the characteristic polarization of each quadrant. The camera presented here is infinitely conjugate (it images an imaginary object at infinite distance) and each pixel of the image corresponds to light incident from a given angular direction. As a diffractive object based on nanostructures, it is possible to imagine that the characteristic polarization is a function of the angle of incidence, so the instrument matrix itself must / can be a function of the angle This can be handled empirically by using a calibration procedure that can absorb angle-dependent effects.
[0346] Fig.18 Angle-dependent calibration is shown: the grating sample for imaging is placed on a double rotating stage in front of the shock beam so that its orientation relative to the beam is can be adjusted. Light from the four diffraction orders used for imaging is incident on four photodiodes that report the intensity of the order. The polarization of the shock beam is adjusted using a wave plate, and these photodiodes are monitored, allowing the polarization instrument matrix A to be determined. This process is configured at many different angles Repeat below to determine the instrument matrix associated with the angle
[0347] This angle-dependent calibration Fig.18 and is conceptually very similar to the grating measurements discussed in Section S3B and the main text. A collimated laser (λ = approximately 532 nm) beam is incident on the grating, and its polarization state The grating is mounted on two rotating stages arranged so that the sample is at the intersection of the axes of these stages. When rotating, it is as if the light rays are incident on the grating from different angular directions, just as in a photographic situation. The grating is placed in its mount in the same orientation as it is placed in the camera system. The polarization of the shock beam can be adjusted by the wave plate to access multiple fixed configurations, where is known (using a full Stokes polarimeter, as shown in Section S3B). The response of the imaging grating order to the incident polarization is measured using four silicon photodiodes. If this is done for many incident polarizations, then the configuration The instrument matrix can be found by inversion in a least squares sense (also in Section S3B).
[0348] At many angles This process is repeated and the results are interpolated to produce the angle-dependent instrument matrix In order to use the polarization image reconstruction pixel by pixel, the Must / can be known. This is described in the next section.
[0349] C. "Chessboard" image registration
[0350] Since the camera is infinitely conjugate, each pixel (in each of the four quadrants) measures the direction from a specific angle However, given the pixel coordinates (x, y) in the polarization image, how do we know the corresponding ? For that matter, since there are four copies of the image on the sensor, how do you align them with each other so you know which sets of four pixels on the sensor belong together? Once you answer these questions, you can form a polarization image as Knowing the angular coordinates of each point allows the use of the correct inverse angle-dependent instrument matrix And correct image registration allows to collect a vector I(x,y) at each point.
[0351] Can be used Fig.19 A procedure described in addresses both of these problems. When placed at a viewing distance of 14 inches from the specimen, a checkerboard pattern is prepared in which each division is related to θ or The center of the board is well marked. Fig.19 The checkerboard here can appear square and regular because it covers a narrow field of view (5°). At a wider field of view, it will appear more distorted. The checkerboard was printed, mounted 14 inches from the camera, and illuminated with a green LED flashlight. A raw image of the checkerboard is obtained, such as Fig.19 Shown on the left. Fig.19The checkerboard acquisition shown here was taken after an outdoor photography session - the optics in the camera that renders distant objects are configured differently than the checkerboard that is in focus (14" away). However, the checkerboard acquisition must / can reflect the camera configuration when the image of interest was taken to ensure proper image registration. This defocus is not a problem.
[0352] Fig.19 Figure 1 shows a checkerboard image registration where a checkerboard pattern, each divided at an angle of 1° in each direction, is used as a calibration target for the camera. Corners are identified and marked in the image, which is used to generate pixel locations (x,y) and the angular coordinates of the light incident on the pixel. Optimized function mapping between .
[0353] This image is adaptively thresholded to increase the contrast of the checkerboard. Next, the Harris corner detection algorithm is applied to the image and the coordinates of the locations where the corners fall are recorded and indexed (e.g. Fig.19 These positions are represented by angular coordinates The pixels were manually labeled, and a curve fitting algorithm was used to generate four functions - one for each quadrant of the imaging sensor - mapping from pixel location (x,y) to angular coordinates
[0354] These functions also contain the center position (x0, y0) of each quadrant, this information can be used to overlay the images on top of each other as a simple translation. In this way, the images are very simply registered to each other.
[0355] D. Polarization Image Synthesis
[0356] Angle-dependent calibration generation function The instrument matrix as a function of the angle of the incident beam. Checkerboard image registration produces a function θ(x,y) showing to which angle in the field of view each pixel coordinate (x,y) corresponds Together, this yields the instrument matrix as a function of pixel coordinates This is Fig. 20 The angle-dependent calibration and checkerboard image registration produce the instrument matrix as a function of the pixel position (x,y).
[0357] like Fig.21 As shown in , when the original acquisition is performed, it can be split into four sub-images and the Stokes vector is calculated as a function of position to form the final polarization image. When acquired raw, the four quadrants can be registered with each other (the centers of these images are also known from the checkerboard image registration). This segmented image produces a four-element intensity vector where each element is a pixel drawn from the same position on all four quadrants. Finally, using the results of the calibration, the polarization image is calculated as
[0358]
[0359] In some embodiments, a metasurface grating includes a substrate and a plurality of nanoscale elements disposed on the substrate. The nanoscale elements define an angle-dependent phase profile that imparts a wave vector that varies with incident angle.
[0360] In some embodiments, the cross-section of at least one nanoscale element is rectangular or other polygonal in shape. In some embodiments, the cross-section of at least one nanoscale element is elliptical or circular. In some embodiments, the cross-section of the nanoscale element can have 2-fold rotational symmetry, or more generally, n-fold rotational symmetry, where n is an integer of 2 or greater than 2.
[0361] In some embodiments, the nanoscale elements are made of semiconductors, oxides (e.g., metal or non-metal oxides), nitrides (e.g., metal or non-metal nitrides), sulfides (e.g., metal or non-metal sulfides), pure elements, or combinations of two or more thereof.
[0362] In some embodiments, the substrate is transparent in the visible spectrum, such as a polymer substrate, a glass substrate, or a substrate comprising fused silica. Suitable substrates transparent in the visible spectrum can have a transmittance of at least about 40%, at least about 50%, at least about 60%, at least about 70%, at least about 80%, at least about 85%, at least about 90%, or at least about 95%, exceeding the visible spectrum or the design or operating wavelength in the visible spectrum.
[0363] In some embodiments, the substrate is curved or flexible, which provides alternative functions, such as adjusting the image distance to the eye or focusing light.
[0364] In some embodiments, the nanoscale elements include dielectric materials. Examples of suitable dielectric materials include metal and non-metal oxides (such as aluminum oxides (e.g., Al2O3), silicon oxides (e.g., SiO2), hafnium oxides (e.g., HfO2), zinc oxides (e.g., ZnO), magnesium oxides (e.g., MgO), or titanium oxides (e.g., TiO2)), metal and non-metal nitrides (such as silicon nitrides (e.g., Si3N4), boron nitrides (e.g., BN), or tungsten nitrides (e.g., WN)), metal and non-metal sulfides, and pure elements (e.g., silicon operating at near-infrared and mid-infrared wavelengths).
[0365] In some embodiments, the nanoscale elements have an aspect ratio (e.g., height / width) greater than about one, such as at least about 1.5: 1, at least about 2: 1, at least about 3: 1, at least about 4: 1, or at least about 5: 1, and up to about 10: 1 or more, or up to about 20: 1 or more. In some embodiments, the geometric dimensions (e.g., height / width / length or diameter / height) of the nanoscale elements are sub-wavelength, such as about 800 nm or less, about 700 nm or less, or about 600 nm or less.
[0366] In some embodiments, the nanoscale elements are tilted nanopillars having a non-zero tilt angle relative to the surface normal of the metasurface grating. In some embodiments, the non-zero tilt angle is about 1 degree or more, about 2 degrees or more, about 5 degrees or more, or about 10 degrees or more.
[0367] As used herein, the singular terms "a," "an," and "the" may include plural referents unless the context clearly dictates otherwise.
[0368] Unless otherwise indicated, spatial descriptions (such as "above", "below", "up", "left", "right", "lower", "top", "bottom", "vertical", "horizontal", "side", "higher", "lower", "up", "upper", "lower", etc.) are indicated relative to the orientation shown in the figure. It should be understood that the spatial descriptions used herein are for illustrative purposes only, and actual implementations of the structures described herein may be spatially arranged in any orientation or manner, provided that the advantages of the embodiments of the present disclosure are not deviated by such an arrangement.
[0369] As used herein, the terms "approximately," "substantially," "substantially," and "about" are used to describe and illustrate small variations. When used in conjunction with an event or situation, these terms may refer to situations where the event or situation occurs exactly as well as situations where the event or situation occurs very approximately. For example, when used in conjunction with a numerical value, these terms may refer to a range of variation of less than or equal to ±10% of that numerical value, such as less than or equal to ±5%, less than or equal to ±4%, less than or equal to ±3%, less than or equal to ±2%, less than or equal to ±1%, less than or equal to ±0.5%, less than or equal to ±0.1%, or less than or equal to ±0.05%. For example, if the difference between two numerical values is less than or equal to ±10% of the average value of the values, such as less than or equal to ±5%, less than or equal to ±4%, less than or equal to ±3%, less than or equal to ±2%, less than or equal to ±1%, less than or equal to ±0.5%, less than or equal to ±0.1%, or less than or equal to ±0.05%, then the two numerical values may be considered to be "substantially" the same.
[0370] In addition, quantities, ratios and other numerical values are sometimes presented herein in a range format. It should be understood that this range format is used for convenience and brevity, and should be flexibly interpreted to include the values explicitly specified as limits of the range, but also to include all individual values or sub-ranges encompassed within that range, as if each value and sub-range were explicitly specified.
[0371] Although the present disclosure has been described and illustrated with reference to the specific embodiments of the present disclosure, these descriptions and illustrations do not limit the present disclosure. It should be understood by those skilled in the art that various changes can be made and equivalents can be substituted without departing from the true spirit and scope of the present disclosure as defined by the appended claims. The illustrations may not necessarily be drawn to scale. Due to manufacturing processes and tolerances, there may be differences between the artistic reproduction in the present disclosure and the actual device. The present invention may also have other embodiments that are not specifically described. The specification and the drawings are considered to be illustrative rather than restrictive. Modifications may be made to adapt specific situations, materials, compositions of matter, methods or processes to the purpose, spirit and scope of the present disclosure. All these modifications are intended to be within the scope of the appended claims. Although the methods disclosed herein have been described with reference to specific operations performed in a specific order, it should be understood that these operations can be combined, subdivided or reordered to form equivalent methods without departing from the teachings of the present disclosure. Therefore, unless otherwise specified herein, the order and grouping of operations are not limitations on the present disclosure.
Claims
1. An optical component, comprising: substrate; as well as A metasurface disposed on a substrate, the metasurface comprising a plurality of repeated and adjacent subdivided grating unit cells, wherein each subdivided grating unit cell comprises a grid of linear birefringent elements, wherein each linear birefringent element of the subdivided grating unit diffracts at least four different polarizations of incident light into four different directions, and Wherein, in each subdivided grating unit cell, at least one element of the linear birefringent element has an angular orientation different from another element of the linear birefringent element.
2. The optical component of claim 1, wherein the metasurface is configured to perform parallel polarization analysis without cascading incident light through bulk polarization optics.
3. An optical component as described in claim 1, wherein the metasurface is not spatially interleaved with any other gratings.
4. The optical component of claim 1, wherein the metasurface is configured to perform full Stokes polarization measurements independent of any other optical components.
5. The optical component of claim 1, wherein the at least four different polarizations correspond to a tetrahedron inscribed in a Poincaré sphere.
6. The optical component of claim 5, wherein the metasurface is configured to output light and analyze four different polarizations in parallel, wherein none of the four different polarizations is orthogonal to any of the other polarizations.
7. The optical component of claim 1, wherein the one or more linear birefringent elements comprise an array of dielectric pillars possessing form birefringence.
8. The optical component of claim 1, wherein the at least four different polarizations of the incident light include six polarization orders corresponding to an octahedron inscribed in a Poincaré sphere.
9. The optical component of claim 1, wherein the metasurface implements a set of analyzers for arbitrarily specified polarizations at the diffraction orders of the metasurface.
10. The optical component of claim 1, wherein the metasurface is configured to process incident light into a spatially varying Jones matrix.
11. An optical component as described in claim 10, wherein the metasurface includes a plurality of birefringent wave plates, and wherein one or more of the angular orientation, absolute phase shift, and phase retardation of at least one of the plurality of birefringent wave plates is configured to process incident light into a spatially varying Jones matrix.
12. The optical component of claim 10, wherein the spatially varying Jones matrix is linearly birefringent.
13. The optical component of claim 10, wherein the spatially varying Jones matrix is unitary.
14. The optical component of claim 10, wherein the spatially varying Jones matrix is an operator that operates in parallel on each of the set of diffraction orders of interest.
15. The optical component of claim 1, further comprising: lens; as well as Imaging sensors, Wherein the metasurface is configured to provide a plurality of polarization orders to an imaging sensor sufficient to enable the imaging sensor to determine a full Stokes vector.
16. The optical component of claim 15, wherein the imaging sensor is configured to determine the full Stokes vector using no polarization orders other than those provided by the metasurface.
17. The optical component of claim 15, wherein the metasurface is a surface of a substrate, the substrate being a single substrate.
18. The optical component of claim 15, wherein the metasurface defines a grating that is not spatially interleaved with any other grating, and the grating is configured to provide the plurality of polarization orders.
19. The optical component of claim 15, wherein the metasurface defines a grating, and the grating is configured to provide the multiple polarization orders without cascading incident light through multiple gratings.
20. The optical component of claim 15, wherein the metasurface is configured to receive incident light and output the plurality of polarization orders without any moving parts.
21. The optical component of claim 1, wherein each linear birefringent element of each subdivided grating unit cell has a discrete element (a, b) having a local Jones matrix J defined by (a,b) : where θ (a,b) , and are the grating parameters at the (a, b)th position in the subdivided grating unit cell and R is the 2×2 rotation matrix.
22. The optical component of claim 21, wherein for each subdivided grating unit cell located at x, y, the metasurface comprises a periodic Jones matrix J(x,y) defined by: where the dimensions of each subdivided raster unit cell are A x B elements, where A and B are integers, and where d a and d b are the element separations along the x and y directions, respectively, where is around the point (a·d a ,b·d b ) a xd b A function that is equal to 1 in the rectangle and equal to 0 everywhere else, where the symbol It means that A·d is the value on x and y respectively. a and B·d b The size of one unit cell separates an infinite 2D array of Dirac delta functions, and where Represents the convolution of the expression within the brackets.
23. The optical component of claim 22, wherein the metasurface has a polarization order whose polarization behavior is described by a Jones matrix derived from the (m, n)th coefficient in the Fourier series of a periodic Jones matrix J(x, y) and defined by: in
