Method for estimating direction of arrival based on artificial neural network and spatiotemporal coding metasurface

Abstract

The application discloses a wave direction estimation method based on an artificial neural network and a space-time coding metasurface, wherein the space-time coding metasurface modulates incident signals according to a pre-designed space-time coding matrix, so that the modulated signals contain multiple harmonic components, and the amplitudes of a part of the harmonic components are selected to construct a vector s. Signals of different incident angles are modulated by the space-time coding metasurface and present different spatial spectrum distributions, corresponding to different vectors s. Through theoretical calculation or a test platform, a corresponding vector s collection in a selected incident angle range is obtained, so as to construct a data set. The data set is used for training an artificial neural network model, and a mapping relationship between the vector s and the signal incident angle is fitted. After the artificial neural network model is trained, the vector s is taken as input, and an estimated value of the signal incident angle is output. The application realizes high-precision wave direction estimation with simple hardware structure and a small amount of computing resources, and has the advantages of low cost, low time delay and high robustness.

Description

Technical Field

[0001] This invention relates to a wave direction estimation method based on artificial neural networks and spatiotemporally coded metasurfaces, belonging to the fields of novel artificial electromagnetic materials and communication technology. Background Technology

[0002] Novel artificial electromagnetic materials are metamaterials composed of artificially designed subwavelength structures. Due to their ability to finely modulate electromagnetic waves, they have been extensively studied in both scientific and engineering communities. As a two-dimensional version of metamaterials, metasurfaces possess advantages such as ultrathin thickness, high integration, and lower insertion loss than metamaterials, thus enabling more flexible manipulation of electromagnetic waves. Digitally coded and programmable metasurfaces were proposed in 2014. They can digitally manipulate electromagnetic waves using only a limited number of unit types (e.g., only two in the case of 1 bit). Digitally coded metasurfaces can be integrated with active devices, where each unit can be independently controlled by a control module, thus creating programmable metasurfaces. Digitally programmable metasurfaces have become a powerful and versatile platform for realizing various functions. Introducing a time-based control dimension allows for digital manipulation of electromagnetic waves in the spatial and frequency domains, resulting in spatiotemporally coded metasurfaces, further expanding the capabilities of digitally coded metasurfaces to manipulate electromagnetic waves.

[0003] Traditional direction-of-arrival (AOA) estimation methods, such as Multi-Signal Classification (MUSIC) and Estimation of Signal Parameters via Rotation Invariance Techniques (ESPRIT), typically rely on array antennas and require eigenvalue decomposition of the covariance matrix of the received signal, consuming significant computational resources and resulting in complex hardware structures, high costs, high latency, and low efficiency. Utilizing spatiotemporally coded metasurfaces to modulate and receive signals can reduce hardware complexity and cost. Several metasurface-based OAA estimation methods have been publicly reported to address these issues. For example, programmable metasurfaces are used as physical layer sampling devices, and the OAA is estimated based on the characteristics of the scattered waves by switching between numerous metasurface scattering modes; other methods use time-modulated arrays or time-controlled metasurfaces to analytically calculate the OAA using the amplitude and phase of harmonics. However, these methods have stringent requirements for testing equipment and environments, or require substantial computational or storage resources, resulting in limited accuracy and applicability.

[0004] A wave arrival estimation method based on spatiotemporally coded metasurfaces and artificial neural network algorithms can accurately obtain the signal direction with only harmonic amplitude information. It requires simple hardware structure, does not require a large amount of computing resources, has good robustness and practicality, and has the advantages of low latency and high efficiency. Summary of the Invention

[0005] Technical problem: The purpose of this invention is to solve the problems of complex hardware structure, high cost and low efficiency of existing wave direction estimation methods, and to provide a wave direction estimation method based on artificial neural network and spatiotemporally coded metasurface.

[0006] Technical Solution: To achieve the above objectives, the present invention provides a wave direction estimation method based on artificial neural networks and spatiotemporally coded metasurfaces, employing the following technical solution:

[0007] 1) The spatiotemporal coding matrix is ​​designed by optimizing the algorithm. The control module provides control signals to the coding unit of the programmable metasurface according to the spatiotemporal coding matrix to realize the modulation of time and space dimensions. The signal with an incident angle of d is modulated by the spatiotemporal coding metasurface and is received in direction A. The modulated signal contains multiple harmonic components. The amplitudes of M harmonic components are selected to construct an M-dimensional vector s.

[0008] 2) Different incident angles d correspond to different vectors s. Select the range of incident angles d of the incoming wave direction to be predicted, obtain the corresponding set of vectors s through theoretical calculation or test platform, and build a dataset.

[0009] 3) Construct an artificial neural network model, using vector s as the input layer and incident angle d as the output. Train it using a computational or test dataset to fit the mapping relationship between vector s and signal incident angle d, and obtain a trained artificial neural network.

[0010] 4) When the signal illuminates the spatiotemporally encoded metasurface, the signal is modulated and received to obtain a vector s, which is used as the input to the trained artificial neural network model, and finally the estimated value of the signal incident angle can be output.

[0011] in,

[0012] In step 1), the programmable metasurface is reflective, transmissive, or active radiative. Each coding unit integrates one or more active devices. The coding unit exhibits different electromagnetic responses according to different control signals. The electromagnetic response is either phase or amplitude.

[0013] In step 1), the control module provides control signals to the coding units of the programmable metasurface according to the spatiotemporal coding matrix to achieve modulation of time and space dimensions; each coding unit periodically changes its electromagnetic response according to the corresponding time coding sequence, so that signals with different incident angles produce different spatial spectrum distributions after being modulated by the spatiotemporal coding metasurface.

[0014] In step 2), the range of the incoming wave direction d to be predicted is selected, and the corresponding set of vectors s is obtained through theoretical calculation or test platform to form a dataset. Depending on the application scenario, the incident angle d can be represented as a one-dimensional or two-dimensional vector. In order to avoid the influence of the signal power on the estimation accuracy of this method, the vector s is normalized or normalized before constructing the dataset.

[0015] In step 3), an artificial neural network model is constructed, with vector s as the input layer and incident angle d as the output. The model is trained using a computational or test dataset to fit the mapping relationship between vector s and signal incident angle d. This mapping relationship is determined by the spatiotemporal coding matrix. By optimizing the spatiotemporal coding matrix, it can be ensured that s corresponding to different incident angles d have good orthogonality. Moreover, within the selected incident angle estimation range, a signal with high power can be received in a specific direction A. This helps to improve the estimation accuracy and robustness of this method.

[0016] In step 4), when the signal illuminates the spatiotemporally encoded metasurface, the signal is modulated and received to obtain vector s, which is used as the input of the trained artificial neural network model. Finally, the estimated value of the signal incident angle can be output. According to the preprocessing method of vector s when building the dataset, vector s is normalized or normalized before being used as the input vector of the artificial neural network model, and the direction of the received signal is consistent with the direction A selected when building the dataset.

[0017] Beneficial effects: A well-constructed spatiotemporal coding matrix ensures an accurate and robust mapping between the harmonic amplitudes of the modulated signal and its azimuth. Signal modulation and acquisition can be achieved using only a relatively simple hardware structure, significantly reducing the hardware complexity and cost required for wave direction estimation methods.

[0018] By introducing an artificial neural network to extract the mapping relationship between the harmonic amplitude and the signal azimuth of the modulated signal, the computational complexity required by conventional wave direction estimation methods is greatly reduced. The signal azimuth information can be obtained through relatively basic calculations, reducing latency and improving efficiency with extremely low computational resources. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a reflective spatiotemporally encoded metasurface;

[0020] Figure 2 This is a schematic diagram of an example of a spatiotemporal coding matrix;

[0021] Figure 3 Yes, when the incident signal is located at and respectively, the reflective spatiotemporal coded metasurface is in Figure 2 The spatial scattering mode of harmonics under the modulation of the spatiotemporal coding matrix;

[0022] Figure 4 It is the normalized harmonic amplitude received by the receiving antenna located in the direction of the normal to the spatiotemporal encoded metasurface when the incident signal is located at and respectively, from the -5th to the +5th harmonic;

[0023] Figure 5 It is the harmonic amplitude received by the receiving antenna located in the direction of the normal to the spatiotemporal coded metasurface when the azimuth of the incident signal is within the range;

[0024] Figure 6 This is a schematic diagram of an example of an artificial neural network model;

[0025] Figure 7 This is a schematic diagram showing the accuracy and error of the artificial neural network model in estimating the direction of arrival for different signal-to-noise ratios after training on a dataset obtained from simulation calculations.

[0026] Figure 8 It is a comparison between the direction estimation result of the incoming wave signal with a signal-to-noise ratio of -12dB and the actual incident angle after the artificial neural network model has been trained using the dataset obtained by simulation calculation;

[0027] Figure 9 This is a comparison between the direction estimation result of an incoming wave signal with a transmission power of 16dBm and the actual incident angle after the artificial neural network model has been trained using a dataset obtained from simulation calculations.

[0028] Figure 10 This is the error distribution of the direction estimation results for incoming wave signals with different transmission powers after the artificial neural network model has been trained using the dataset obtained from the test platform. Detailed Implementation

[0029] The present invention provides a wave direction estimation method based on artificial neural networks and spatiotemporally coded metasurfaces, comprising the following steps:

[0030] 1) A spatiotemporal coding matrix is ​​designed through an optimized algorithm. The control module provides control signals to the coding units of the programmable metasurface based on the spatiotemporal coding matrix, realizing modulation in both time and space dimensions. The incident signal is modulated by the spatiotemporally coded metasurface and received in direction A. The modulated signal contains multiple harmonic components. The amplitudes of M harmonic components are selected to construct an M-dimensional vector s.

[0031] 2) Different incident angles d correspond to different vectors s. Select the range of the incoming wave direction d to be predicted, obtain the corresponding set of vectors s through theoretical calculation or test platform, and build a dataset.

[0032] 3) Construct an artificial neural network model, using vector s as the input layer and incident angle d as the output. Train the model using a computational or test dataset to fit the mapping relationship between vector s and signal incident angle d.

[0033] 4) When the signal is irradiated onto the spatiotemporally encoded metasurface, the signal is modulated and received to obtain a vector s, which is used as the input of the artificial neural network model to output the estimated incident angle of the signal.

[0034] In step 1), the programmable metasurface can be reflective, transmissive, active radiative, etc. Each coding unit integrates one or more active devices. The coding unit exhibits different electromagnetic responses according to different control signals. The electromagnetic response can be phase or amplitude.

[0035] In step 1), the control module provides control signals to the coding units of the programmable metasurface according to the spatiotemporal coding matrix, thereby achieving modulation in both time and space dimensions. Each coding unit periodically changes its electromagnetic response according to the corresponding time coding sequence, so that incident signals at different angles produce different spatial spectral distributions after being modulated by the spatiotemporal coded metasurface.

[0036] In step 2), the range of the incoming wave direction d to be predicted is selected, and the corresponding set of vectors s is obtained through theoretical calculation or a test platform to construct a dataset. Depending on the application scenario, the incident angle d can be represented as a one-dimensional or two-dimensional vector. To avoid the influence of signal power on the estimation accuracy of this method, the vectors s can be normalized or normalized before constructing the dataset.

[0037] In step 3), an artificial neural network model is constructed with vector s as the input layer and incident angle d as the output. The model is trained using a computational or test dataset to fit the mapping relationship between vector s and signal incident angle d. This mapping relationship is determined by the spatiotemporal coding matrix. By optimizing the spatiotemporal coding matrix, it can be ensured that s corresponding to different incident angles d have good orthogonality. Furthermore, within the selected incident angle estimation range, a signal with high power can be received in a specific direction A. This helps to improve the estimation accuracy and robustness of this method.

[0038] In step 4), when the signal illuminates the spatiotemporally encoded metasurface, the signal is modulated and received, resulting in a vector s. This vector s is used as the input to the artificial neural network model, and the estimated incident angle of the output signal is used as the output. Following the preprocessing method for vector s when constructing the dataset, vector s is normalized or normalized before being used as the input vector of the artificial neural network model, and the orientation of the received signal is consistent with the orientation A selected when constructing the dataset.

[0039] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0040] This invention provides a wave arrival direction estimation method based on artificial neural networks and spatiotemporally coded metasurfaces, addressing the problems of complex hardware structure, high cost, and low efficiency in existing wave arrival direction estimation methods. By introducing a rationally constructed spatiotemporally coded matrix, an accurate and robust mapping relationship exists between the harmonic amplitude and signal orientation of the modulated signal. An artificial neural network is introduced to extract the mapping relationship between the harmonic amplitude and signal orientation of the modulated signal. Signal orientation information can be obtained through relatively basic calculations, reducing latency and improving efficiency with minimal computational resources. The wave arrival direction estimation method based on artificial neural networks and spatiotemporally coded metasurfaces can precisely control the electromagnetic wave energy distribution in the spatial and frequency domains, providing greater degrees of freedom. The specific design steps are as follows:

[0041] Step 1: Design the coding units to be arranged periodically in space to form a programmable metasurface;

[0042] Each encoding unit integrates one or more active devices. The unit exhibits different electromagnetic responses according to different control signals. The electromagnetic response can be the phase or amplitude of the reflected wave or the transmitted wave. Taking the reflection phase as an example, in the case of 1 bit encoding, "0" and "1" correspond to a reflection phase of 0 degrees and 180 degrees, respectively. In the case of 2 bits encoding, "00", "01", "10" and "11" correspond to a reflection phase of 0 degrees, 90 degrees, 180 degrees and 270 degrees, respectively.

[0043] Step 2: The control module provides control signals to the coding units of the programmable metasurface based on the spatiotemporal coding matrix, thereby achieving modulation of the time and space dimensions.

[0044] Each unit periodically changes its electromagnetic response according to the corresponding time-coded sequence, so that incident signals at different angles produce different spatial spectrum distributions after being modulated by the spatiotemporally coded metasurface.

[0045] Step 3: Design a spatiotemporal coding matrix using an optimized algorithm. The incident signal is modulated by the spatiotemporally coded metasurface and received in direction A. The modulated signal contains multiple harmonic components. Select the amplitudes of a portion of the harmonic components to construct a vector s.

[0046] Without loss of generality, we consider a two-dimensional scenario, taking a reflective metasurface as an example, consisting of N columns of programmable cells forming a digitally programmable metasurface. Each column of cells is modulated by an independent set of time-coded sequences, manifested as a periodically varying reflection coefficient of the cell. The time-coded sequence corresponding to the nth column of programmable cells can be represented by a periodic function of time t:

[0047]

[0048] in It is the reflection coefficient of the nth programmable unit in the lth time slot, and the length of each time slot is τ. A pulse function with period T0 = L × τ can be expressed as:

[0049]

[0050] Time-coded sequence Γ n The length of (t) is L, and N sets of time-coded sequences together form a spatiotemporal coding matrix Z∈R. N×L The spatiotemporally encoded metasurface is incident at an angle θ. i When illuminated by a plane wave, the scattering mode of the spacetime encoded metasurface changes periodically over time, and can be expressed as a function of the azimuth angle θ and the incident angle θ. i , and a function of time t:

[0051]

[0052] Where E(θ) is the basic coding unit at frequency f c The scattering mode under the given condition is approximately cosθ, k c =2πf c / c is the wavenumber of the incident wave in free space, c is the speed of light, and d is the spatial period of the coding unit. The periodically changing time-coded sequence Γ n (t) can be extended to a Fourier series, and the v-th order Fourier coefficients are expressed as:

[0053]

[0054] Therefore, the vth harmonic (f c The scattering mode of +vf0) can be further expressed as

[0055]

[0056] Where k v =2π(f c +vf0) / c is the wave number of the vth harmonic. The vth harmonic received by the antenna in the direction normal to the spacetime metasurface can be simplified to...

[0057]

[0058] The amplitude of the vth harmonic received in the normal direction is defined as s. v (θ i )=|G v (θ=0°,θ i )|, while frequency f c The amplitudes of harmonics on +vf0 (v = ±1, ±2... ±M) can form a vector s(θ) i )∈R 2MAs can be seen from formula (4), Depending on the spatiotemporal coding matrix Z, it determines the sequence of elements from s(θ). i ) to θ i The mapping relationship.

[0059] Step 4: Select the range of incoming wave directions to be predicted, obtain the corresponding set of vectors s through theoretical calculations or testing platforms, and build a dataset;

[0060] Depending on the chosen application scenario, the incident angle is represented as a one-dimensional vector. To avoid the influence of signal power on the estimation accuracy of this method, the vector s is normalized before constructing the dataset.

[0061] Step 5: Construct an artificial neural network model, using vector s as the input layer and the incident angle as the output. Train the model using a computational or test dataset to fit the mapping relationship between vector s and the signal incident angle.

[0062] The mapping relationship is determined by the spatiotemporal coding matrix. By optimizing the spatiotemporal coding matrix, it can be ensured that there is good orthogonality between s corresponding to different incident angles d. Moreover, within the selected incident angle estimation range, a signal with high power can be received in a specific direction A, which helps to improve the estimation accuracy and robustness of this method.

[0063] Step 6: When the signal illuminates the spatiotemporally encoded metasurface, the signal is modulated and received, resulting in a vector s. This vector s is used as the input to the artificial neural network model, and the estimated incident angle of the output signal is used as the output.

[0064] Following the preprocessing method for vector s when constructing the dataset, vector s is normalized or normalized before being used as the input vector of the artificial neural network model, and the orientation of the received signal is consistent with the orientation A selected when constructing the dataset.

[0065] Figure 1 This is a schematic diagram of a reflective spatiotemporally coded metasurface. The coding units are arranged periodically in space to form a programmable metasurface. The control module provides control signals to each programmable unit to achieve modulation in the time dimension. Under the modulation of the spatiotemporally coded matrix, the incident signal is converted into multiple harmonics.

[0066] Figure 2 This is an example of a spatiotemporal coding matrix with dimensions (8,8,8), representing that the programmable metasurface contains 8 columns of basic units, with a spatial arrangement period of half a wavelength, and a corresponding temporal coding sequence length of 10. Different grayscale squares represent "0" and "1" codes, corresponding to different electromagnetic responses of the coding units; here, electromagnetic response refers to the reflection phase. Each column of units is controlled to cycle periodically according to the temporal coding sequence shown in the figure.

[0067] Figure 3 The values ​​for the incident signal at θ are given respectively. i =0° and θ i At 60°, the reflective spatiotemporal coding metasurface in Figure 2 Under the modulation of the spatiotemporal coding matrix, the incident signal is converted into multiple harmonics. Different incident angles and different harmonics exhibit different spatial scattering modes. The figure shows the scattering modes from the -10th to the +10th harmonics. It can be seen that the scattered power is mainly concentrated in the -4th to +4th harmonics.

[0068] Figure 4 The values ​​for the incident signal at θ are given respectively. i =0° and θ i At 60°, the normalized amplitude of the harmonics received by the receiving antenna located in the direction of the normal to the spatiotemporal coding metasurface ranges from the -5th to the +5th harmonic.

[0069] Figure 5 Give the position of the incident signal at θ i The harmonic amplitude received by the receiving antenna located in the normal direction of the spatiotemporal coding metasurface when the range is [-90°, 90°]. The amplitudes of the -5th to +5th harmonics form a vector, corresponding to the signal orientation, and constitute the dataset.

[0070] Figure 6 An example of an artificial neural network model is given, with the following structural features: a 10-dimensional input layer, two hidden layers with 32 and 16 neurons respectively, a 1-dimensional output layer, the ReLU function as the activation function, and the mean squared error as the loss function. The vector composed of harmonic amplitudes is divided by the maximum value in the vector, normalized, and then used as the input vector of the artificial neural network. After the model is trained, the signal's azimuth information can be obtained from the vector composed of harmonic amplitudes through some basic calculations.

[0071] Figure 7 Given Figure 5 The accuracy of the incoming wave direction estimation after the artificial neural network is trained is shown. In addition to the ideal noise-free condition, data with a signal-to-noise ratio (SNR) of [10, 4, 0, -2] dB are added to the dataset for training, which helps improve the generalization ability and robustness of the artificial neural network model. It can be seen that the mean absolute error and maximum absolute error increase as the SNR decreases, leading to a decrease in estimation accuracy.

[0072] In addition, a dataset with a signal-to-noise ratio of [-4, -6, -8, -10, -12] dB was also used as a test set to further evaluate the performance of the trained network. It can be seen that even under low signal-to-noise ratio conditions, the artificial neural network model can maintain high accuracy in estimating the direction of arrival for unseen data.

[0073] Figure 8The paper presents a comparison between the direction estimation results of the incoming wave signal with a signal-to-noise ratio of -12dB and the actual incident angle after the artificial neural network model has been trained using the dataset obtained from simulation calculations.

[0074] Figure 9 The paper presents a comparison between the direction estimation results of an incoming wave signal with a transmission power of 16dBm and the actual incident angle after the artificial neural network model has been trained using a dataset obtained from simulation calculations.

[0075] In addition to using simulation data to train artificial neural networks, datasets can also be obtained using test platforms. Incident signals with different transmission powers can form a larger dataset, improving the robustness and accuracy of incoming wave estimation.

[0076] Figure 10 The error distribution of the direction estimation results of incoming wave signals with different transmission powers is presented after the artificial neural network model is trained using the dataset obtained from the test platform.

[0077] The design method provided by this invention aims to solve the problems of complex hardware structure, high cost, and low efficiency of existing wave direction estimation methods. It provides a wave direction estimation method based on artificial neural networks and spatiotemporally coded metasurfaces, which can realize signal modulation and acquisition functions with only a relatively simple hardware structure, greatly reducing the hardware complexity and cost required for wave direction estimation methods. It also reduces latency and improves efficiency with extremely low computing resources, and has great application potential in practice.

[0078] The above description is merely a preferred embodiment of the present invention. Because the design concept of the present invention is clear and its application prospects are broad, this design method can be applied to microwave, millimeter wave, terahertz, infrared, and visible light bands. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for estimating the direction of arrival of waves based on artificial neural networks and spatiotemporally coded metasurfaces, characterized in that: Includes the following steps: 1) A spatiotemporal coding matrix is ​​designed through an optimized algorithm. The control module provides control signals to the coding units of the programmable metasurface based on the spatiotemporal coding matrix, thereby achieving modulation of the temporal and spatial dimensions; the incident angle is... d The signal is modulated by a spatiotemporally coded metasurface and received in direction A; the modulated signal contains multiple harmonic components, and the amplitudes of M harmonic components are selected to construct an M-dimensional vector. s ; 2) Different incident angles d Corresponding to different vectors s Select the direction and angle of incidence of the incoming wave that needs to be predicted. d The range is determined by obtaining the corresponding vector through theoretical calculations or testing platforms. s A collection of data, used to construct a dataset; 3) Construct an artificial neural network model, using vectors s As the input layer, the incident angle d As output, it is trained using a computational or test dataset to obtain a fitted vector. s With the angle of incidence of the signal d The mapping relationship between them is used to obtain a trained artificial neural network; 4) When a signal illuminates the spatiotemporally coded metasurface, the signal is modulated and received to obtain a vector. s This is used as input to the trained artificial neural network model, and the estimated value of the signal incident angle can be output in the end. in, In step 3), an artificial neural network model is constructed using vectors. s As the input layer, the incident angle d As output, it is trained using a computational or test dataset to obtain a fitted vector. s With the angle of incidence of the signal d The mapping relationship between them is determined by the spatiotemporal coding matrix. By optimizing the spatiotemporal coding matrix, different incident angles can be guaranteed. d corresponding s They have good orthogonality, and within the selected incident angle estimation range, a signal with high power can be received in a specific direction A.

2. The wave direction estimation method based on artificial neural networks and spatiotemporally coded metasurfaces according to claim 1, characterized in that, In step 1), the programmable metasurface is reflective, transmissive, or active radiative. Each coding unit integrates one or more active devices. The coding unit exhibits different electromagnetic responses according to different control signals. The electromagnetic response is either phase or amplitude.

3. The wave direction estimation method based on artificial neural networks and spatiotemporally coded metasurfaces according to claim 1, characterized in that, In step 1), the control module provides control signals to the coding units of the programmable metasurface according to the spatiotemporal coding matrix to achieve modulation of time and space dimensions; each coding unit periodically changes its electromagnetic response according to the corresponding time coding sequence, so that signals with different incident angles produce different spatial spectrum distributions after being modulated by the spatiotemporal coding metasurface.

4. The wave direction estimation method based on artificial neural networks and spatiotemporally coded metasurfaces according to claim 1, characterized in that, In step 2), select the direction of incoming wave to be predicted. d The range is determined by obtaining the corresponding vector through theoretical calculations or testing platforms. s The set of data is used to construct a dataset; the angle of incidence is determined based on the application scenario. d Represented as a one-dimensional or two-dimensional vector, to avoid the influence of signal power magnitude on the estimation accuracy of this method, the vector is... s After normalization or normalization, the dataset is then constructed.

5. The wave direction estimation method based on artificial neural networks and spatiotemporally coded metasurfaces according to claim 1, characterized in that, In step 4), when the signal illuminates the spatiotemporally coded metasurface, the signal is modulated and received to obtain a vector. s This is used as input to the trained artificial neural network model, and the final output is an estimated value of the signal incident angle, based on the vector values ​​used when building the dataset. s Preprocessing methods, transforming vectors s The signal is normalized or normalized before being used as the input vector of the artificial neural network model, and the direction of the received signal is consistent with the direction A selected when constructing the dataset.

Images