Intelligent metasurface four-corner sparse observation assisted signal source positioning method
By arranging RIS units at the four corners of a virtual grid plane, and combining implicit neural representation networks and physical regularization loss functions, the problems of high hardware cost and low positioning accuracy in existing signal source localization technologies are solved, achieving high-precision signal source localization with low complexity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAZHONG UNIV OF SCI & TECH
- Filing Date
- 2026-04-17
- Publication Date
- 2026-07-21
AI Technical Summary
Existing signal source localization technologies require large-aperture arrays and radio frequency receiving links, resulting in high hardware costs, large size, high power consumption, and high calibration complexity. Sparse sampling methods have low localization accuracy and high hardware costs.
A smart metasurface method with sparse observation at four corners is adopted. By arranging a small number of RIS units at the four corners of the virtual grid plane, the virtual full aperture signal is reconstructed using an implicit neural representation network. The localization is then performed by combining the physical regularization loss function, which reduces hardware complexity and improves localization accuracy.
It achieves the positioning accuracy of an equivalent full-aperture array with low hardware complexity, reduces hardware cost and power consumption, improves positioning stability and resolution, and has better generalization and robustness.
Smart Images
Figure CN122430784A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of wireless positioning technology, and more specifically, relates to a signal source positioning method assisted by intelligent metasurface four-corner sparse observation. Background Technology
[0002] Signal source localization typically relies on array receivers consisting of a large number of antenna elements, using array manifold matching, beamforming, or subspace algorithms to obtain azimuth and elevation angle information. To achieve high angular resolution and noise immunity, existing solutions often require large-aperture arrays and RF receiver links matching the number of array elements, leading to increased hardware cost, size, power consumption, and calibration complexity.
[0003] Intelligent metasurfaces (RIS), composed of numerous controllable elements, modulate electromagnetic wave propagation by loading different reflection configurations, and can be used for communication enhancement and electromagnetic environment shaping. By utilizing RIS for reception on the incident side or forming multiple snapshot observations on the reflection link, RIS can assist in reverse sensing of signal source orientation. However, many existing schemes rely on sampling the entire RIS array to obtain sufficient information in the spatial dimension, resulting in high hardware complexity. To reduce computational cost, some methods involve sparse sampling from the entire RIS array, using the sparsely sampled signal for signal source localization. Compared to full array sampling, this method has relatively lower localization accuracy and still requires significant hardware costs. Summary of the Invention
[0004] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a signal source localization method assisted by four-corner sparse observation of intelligent metasurface, the purpose of which is to obtain the positioning accuracy equivalent to the full aperture array by using only a small number of RIS units at the four corners.
[0005] To achieve the above objectives, this invention provides a method for locating signal sources with four-corner sparse observation assistance from intelligent metasurfaces, comprising: Under the codebook configuration of the RIS unit, the observed samples are obtained after the transmitted signal of the signal source to be located is passively reflected by the RIS unit and received by the receiver. The codebook configuration is then changed to obtain an observation sequence consisting of T such observed samples. The RIS unit is only used in... The four corners of the virtual grid plane are arranged, symbol " " indicates transpose; The coordinates of the RIS units are normalized and then input into the implicit neural representation network. In this process, the complex field matrix of the virtual full aperture signal is obtained. Estimate , Represent the set of complex numbers; estimate the virtual full aperture complex field. Input the observation model to obtain the receiver output observation sequence. Estimate ; Including and Under a loss function including data consistency loss, the implicit neural representation network... During training, when the loss converges, the implicit neural representation network... Output the complex field matrix of the virtual full aperture signal The optimal estimate; The azimuth and elevation angles of the signal source to be located are determined based on the optimal estimate.
[0006] Furthermore, the loss function also includes physical regularization loss. The physical regularization loss The construction methods include: Construct the row prediction coefficient vector as follows The column-directed prediction coefficient vector is ;in, The preset physical regularization loss The maximum prediction order; The complex field matrix based on the virtual full aperture signal Estimate The row-direction prediction coefficient vector and the column-direction prediction coefficient vector are used to construct the prediction residuals:
[0007]
[0008] in, , These are the row-directed forecast residuals and the column-directed forecast residuals, respectively. The index of each grid within the virtual grid plane. , ; Construct the physical regularized loss using the predicted residuals. :
[0009] in, The implicit neural representation network Network parameters.
[0010] Furthermore, the observation model is as follows:
[0011]
[0012] in, , For the RIS unit Secondary codebook configuration. , , This is the set of indices for the RIS cells arranged at the four corners. Indicates taking the absolute value; Indicates vector diagonalization; The equivalent complex gain vector from the four corner RIS units to the receiver; The selection matrix for the RIS units arranged at the four corners; The complex field matrix of the virtual full aperture signal Vectorization, Indicates vectorization operation, It is noise.
[0013] Furthermore, the coordinates of the RIS units are normalized and then input into the implicit neural representation network. In this process, the complex field matrix of the virtual full aperture signal is obtained. Estimate ,include: The coordinates of the RIS units are normalized and then input into the implicit neural representation network. In the process, the complex field of the signal at the coordinates of each RIS unit is obtained; based on the complex field of the signal at the coordinates of each RIS unit, sampling is performed on the virtual mesh plane to obtain the complex field matrix of the virtual full aperture signal. Estimate .
[0014] Furthermore, the coordinates of the RIS units are normalized and then input into the implicit neural representation network. Previously, it also included: performing position encoding on the coordinates of the normalized RIS units to obtain encoded features; the implicit neural representation network The complex signal field at the coordinates of the RIS unit is obtained based on the encoded features.
[0015] Furthermore, the location encoding is as follows: Position encoding of sine and cosine.
[0016] Furthermore, the data consistency loss for:
[0017] in, For the first RIS unit The observation sequence output by the receiver under the secondary codebook configuration for The estimate, The implicit neural representation network Network parameters.
[0018] The present invention also provides a signal source localization system assisted by intelligent metasurface four-corner sparse observation, including a computer-readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read executable instructions stored in the computer-readable storage medium and execute the signal source localization method assisted by four-corner sparse observation of the intelligent metasurface described above.
[0019] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the signal source localization method assisted by four-corner sparse observation of intelligent metasurfaces as described in any of the preceding claims.
[0020] The present invention also provides a computer program product, including a computer program that, when the computer program is run on a computer, causes the computer to execute the signal source localization method assisted by four-corner sparse observation of the intelligent metasurface as described above.
[0021] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects: (1) This invention can achieve equivalent large aperture observation and improve positioning performance: When only a small number of RIS units at the four corners participate in signal observation, a virtual full aperture complex field is constructed to form a virtual large aperture array response, thereby providing a more sufficient observation dimension and angle spectrum construction basis for signal source positioning, and obtaining positioning accuracy equivalent to the full aperture array, thus improving positioning stability and resolution.
[0022] (2) The present invention can reduce hardware link and deployment costs: there is no need to densely deploy sensor links or multi-channel receivers on the RIS array. Observation and acquisition can be completed by simply arranging a small number of controllable RIS units and a single receiving channel (receiver) at the four corners. This significantly reduces the number of RF links, wiring and synchronous calibration complexity, while reducing power consumption and engineering deployment difficulty, and has better scalability.
[0023] (3) This invention can reduce the dependence on large-scale labeled data: During the reconstruction process, neither the observation data consistency loss nor the physical regularization loss depends on the pre-built large-scale labeled dataset. When system parameters or scenes change, adaptive updates can be made based on a small number of observations, which has better generalization when the propagation environment or scene distribution changes, and is convenient for engineering implementation and migration.
[0024] (4) The underdetermined extrapolation of the present invention is more stable and robust: Considering that the virtual full aperture is an underdetermined problem and is easily sensitive to noise, the present invention uses physical structure prior constraints (physical canonical loss) to constrain the spatial correlation of the virtual full aperture complex field, so that the extrapolation problem under sparse observation has better identifiability and convergence stability, reduces the reconstruction error caused by noise and model mismatch, and improves the reliability of the final positioning decision.
[0025] In summary, this invention proposes a signal source localization method assisted by four-corner sparse observations of intelligent metasurfaces. It is the first to utilize four-corner RIS subarray observations for signal source localization, retrieving the complex field response of the virtual full aperture using only the four-corner RIS subarray observations. This achieves localization accuracy close to that of full-aperture RIS with lower hardware complexity. Furthermore, the use of only four-corner RIS subarray observations results in low hardware complexity. Attached Figure Description
[0026] Figure 1 This is a schematic diagram of the observation model and the four-corner RIS arrangement in an embodiment of the present invention.
[0027] Figure 2 This is a flowchart illustrating the implementation of an embodiment of the present invention.
[0028] Figure 3 These are the results of the DoA experiment; among them, Figure 3 (a)-(c) in the figure are the simulation results of the four-corner RIS array scheme in the embodiments of the present invention, the DoA results of the 64×64 complete RIS array array and the 32×32 array RIS array array respectively. Detailed Implementation
[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.
[0030] Example 1 In the RIS-assisted reverse sensing scenario, the embodiments of the present invention arrange controllable RIS units only at the four corners of the plane, and do not arrange RIS units in other areas. Under the condition that the reflection link observation is collected by a single receiver, the equivalent virtual full aperture complex field is reconstructed, thereby locating one or more unknown signal sources.
[0031] This invention provides a method for locating signal sources with the assistance of four-corner sparse observation of intelligent metasurfaces, including the composition of a physical entity system and the method flow.
[0032] 1. Composition of physical entity system like Figure 1 As shown in the embodiments of the present invention, the physical entity system includes at least the following modules: (1) RIS array: geometrically defined The virtual grid coordinate system This represents the size of the planar RIS array. In practice, multiple RIS cells are placed only at the four corners of the RIS array (each corner RIS cell forms a RIS subarray, and the four corner RIS subarrays form the RIS array). The index set of the RIS cells placed at the four corners is... RIS elements are not placed in the remaining grid positions. The RIS array itself does not contain a receiver or sensor link, and signal sampling is not performed on the RIS array surface. The input of the RIS array is a pre-designed... The phase configuration control instructions (codebook) of the internal RIS unit output passive reflection modulation of the signal phase transmitted by the signal source to be located. In this embodiment of the invention, the length and width of the subarray formed by the RIS units arranged at each corner are greater than the number of signal sources to be located.
[0033] (2) The signal source Tx to be located, set in the far field of the RIS array, includes one or more signal sources. The input is a signal transmission command, and the output is a stable and continuous signal. In this embodiment of the invention, the signal emitted by the signal source to be located is an electromagnetic wave signal.
[0034] (3) Receiving module: includes a receiver Rx. The input is via... The link signal after passive reflection from the internal RIS unit outputs as a complex baseband sample sequence of T observations. ,symbol" " indicates transpose. After the internal RIS unit adjusts its state according to the selected codebook, the signal emitted by the signal source to be located is processed... After passive reflection by the internal RIS unit, the data is received by the receiver, which outputs a single observation complex baseband sample; Internal RIS unit Under the codebook configuration, the receiver outputs a sequence of T observations of complex baseband samples. .
[0035] Physical entities can be connected via wired or wireless communication.
[0036] 2. Algorithm modules exchange information through calls or data interfaces: (1) Quad-corner RIS subarray configuration module: determines the size of the quad-corner RIS subarray. (Including the length and width of the subarray) and the set of indices of the RIS cells in the RIS array. and generate Group configuration sequence , , that is Group codebook. Input is the layout parameters ( and ) and configuration strategy The output is Phase state of the internal RIS cell in each snapshot (single sample).
[0037] (2) Signal reconstruction module: using implicit neural representation network This represents the virtual full-aperture complex field. The inputs are the coordinates of the RIS cell, the receiver coordinates, and the observation sequence output by the receiver module. (i.e. The configuration sequence output by the four-corner RIS subarray configuration module. (Codebook), output is a virtual full-aperture complex field estimate. And the estimation of the observation sequence output by the receiver. .
[0038] (3) Prior constraint module: for Apply two-dimensional linear predictable physical structure regularization, and to and Apply regular expressions between the data. The input is... , as well as The output is the physical structure regularization loss. Data regularization loss And the updated structural parameters ( ).
[0039] (4) Localization module: This module stores the output of the trained implicit neural network. As a virtual large-aperture array response, based on this virtual full-aperture complex field estimation Construct a two-dimensional steering vector and perform a 2D MUSIC equal angle spectrum search to output the azimuth and elevation angles of one or more signal sources (signal sources to be located).
[0040] 3. Method and Flow like Figure 2 As shown, the method of this embodiment includes the following steps, the order of which can be adjusted without contradiction: (S1) Array modeling and four-corner RIS arrangement definition: geometrically predefined A two-dimensional grid coordinate system is used to define the complex field matrix representing the virtual full-aperture signal. Define the size of the four corner RIS subarray as... The index set of the actual RIS units arranged in the four corner blocks Defined by the Cartesian product of row and column indices: ； in This indicates the actual location of the RIS cell in the RIS array that is configured to reflect the emitted signal from the signal source to be located; that is, the RIS cell number within the entire RIS array. No RIS cells are placed in other locations. This represents the Cartesian product.
[0041] (S2) Multi-configuration snapshot sampling: for Internal RIS unit loading Different phase configurations (Codebook), under each codebook configuration, the receiving module collects the signal after reflection by RIS, forming a T-time observation sequence. .
[0042] (S3) Observation model construction: Establish a linear mapping from the virtual full-aperture complex field to the receiver observation, and convert the scalar observation sequence under the T-order RIS configuration codebook. They are stacked into a unified observation equation.
[0043] Let the virtual full aperture complex field matrix be... ,symbol Let the set of complex numbers be represented by the following: Then the vectorized matrix is obtained , This represents the vectorization operation. The selection matrix corresponding to the four-corner controllable RIS unit is: (As known), 0 indicates that there is no RIS unit at the current position, and 1 indicates that there is a RIS unit at the current position. In this embodiment of the invention, RIS units are set at the four corners of the virtual mesh coordinate system, where , If we take the absolute value, then the complex field vector of the four corner elements can be written as: .
[0044] Define the equivalent complex gain vector from the four corner units to the receiver as follows: , Including propagation phase and amplitude fading, which can be obtained through geometric calibration / calibration. For the... The secondary RIS reflection configuration, its configuration codebook is as follows: Then, the scalar observations (observed complex baseband samples) of the receiver under this configuration are: ; in, For noise; This represents element-wise multiplication. express transpose, This represents vector diagonalization. (The following is a separate, unrelated sentence:) The secondary observations are stacked as follows and order , that is, The unified linear observation equation output by the receiver is obtained as follows: ; Among them, let ,but It also reflects the four-corner arrangement of the RIS unit, the propagation gain, and the phase configuration modulation.
[0045] (S4) Implicit Neural Representation (INR) Modeling: Constructing an INR Network The network input is normalized RIS cell coordinates. and location-coded The obtained features. The network output is the estimated virtual full-aperture complex field value. , for The coordinates of each grid in the virtual grid coordinate system.
[0046] coordinates Normalization to Interval, and using Sine and cosine position coding To enhance the network's ability to represent high-frequency phase changes, positional encoding can be expressed as: ; Position-encoded feature input INR network Output coordinates The real part of the signal complex field With the imaginary part : ; ; in, It represents the imaginary unit.
[0047] The virtual full-aperture complex field matrix is obtained by sampling the network at all grid points. And to the quantization matrix, obtain . Substituting into the linear observation equation in S3, we obtain the estimated observation sequence output by the receiver. .
[0048] (S5.1) Neural Network Training: Using data consistency as the primary loss, predictive observations are constructed based on the unified model in S3, and network parameters are optimized. .
[0049] Depend on Constructing predictive observations: ; Using data consistency as the loss function: ; Through iterative updates make convergence.
[0050] (S5.2) Physical structure prior regularization: Based on S5.1, the full aperture complex field of the INR output is regularized. A two-dimensional linear predictable structural prior is introduced, and regular weights are set to form the total loss, so as to improve the identifiability and reconstruction stability of underdetermined extrapolation.
[0051] Let the maximum prediction order of a two-dimensional linear predictable structure be... The row prediction coefficient vector is The column-directed prediction coefficient vector is Indexes at each grid point on the RIS array. , , Construct the predicted residuals: ; ; Constructing physical regularization terms using prediction residuals: ; Total loss: ; in, The regularization weights are determined empirically. A two-stage training approach is used to achieve stable convergence: Stage 1 minimizes only the... get Find feasible initial values; stage 2 minimizes these initial values. and jointly update When the loss converges, the trained INR network is obtained. The network output at this time This serves as the final required virtual full-aperture complex field estimate.
[0052] (S6) DoA Estimation Output: The virtual full aperture complex field estimate obtained in S5.2 is then used. The single-snapshot array response is considered as a virtual uniform planar array. Based on DoA estimation is performed to obtain the azimuth and elevation angles of one or more signal sources.
[0053] like Figure 3 As shown, Figure 3 (a) in the figure refers to the arrangement at the four corners in an embodiment of the present invention. RIS subarray ( That is, in a 64×64 array, only the four corners are arranged. The total number of RIS subarrays required is [number missing]. The reconstructed equivalent full-aperture complex field is used to obtain the DoA result. Figure 3 (b) shows the full-aperture complex field reconstructed on a 64×64 full array (requiring a total of 64×64 RIS elements), with the obtained DoA result. Figure 3 (c) in the context of using The DoA results obtained by splicing RIS units into a full array (the same number of RIS units as used in the embodiments of this invention, but with a different arrangement) show that, compared to Figure 3 (c) In this case, the RIS subarrays are directly spliced together into an array surface according to their physical geometric positions. In the embodiment of the invention, the four corners (complete array) are... By reconstructing the array surface using an equivalent full-aperture complex field under sparse observations, higher and more stable positioning accuracy can be obtained. Furthermore, under the same experimental conditions, the positioning accuracy of this embodiment of the invention is comparable to... Figure 3 The positioning results shown in (b) are basically consistent when all the real full-aperture RIS units are involved in the observation.
[0054] This invention provides an equivalent full-aperture complex field reconstruction mechanism for sparse observations: Modeling is performed using a passive reflection link receiving observations through a small number of controllable RIS (Reflection Link) units. A system of linear equations about the complex field to be estimated is formed by changing the RIS configuration and stacking multiple observations. The virtual full-aperture complex field is solved under the constraints of these equations, achieving an equivalent extension from a small observation aperture to a virtual large aperture.
[0055] This invention provides an implicit representation parameterization for complex fields: it uses an implicit function representation of coordinate-mapped complex fields to model the full aperture complex field continuously, transforming unknowns from high-dimensional discrete values into low-dimensional network parameters, improving solvability and generalization ability under underdetermined conditions, and reducing dependence on prior labeled data.
[0056] This invention constructs a physical structure regularization driving mechanism: a two-dimensional linear and predictable structural prior constraint is introduced into the virtual full aperture complex field of the INR output, and the structural relationship that conforms to the physical laws is used to suppress noise and inconsistencies, thereby enhancing the extrapolation stability and robustness of underdetermined problems.
[0057] Example 2 This invention provides a signal source localization system assisted by four-corner sparse observation of an intelligent metasurface, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the signal source localization method assisted by four-corner sparse observation of an intelligent metasurface in Embodiment 1 above.
[0058] The relevant technical solutions are the same as above, and will not be repeated here.
[0059] Example 3 This invention provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the intelligent metasurface four-corner sparse observation-assisted signal source localization method in Embodiment 1 above.
[0060] Specifically, the memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.
[0061] The relevant technical solutions are the same as above, and will not be repeated here.
[0062] Example 4 This invention provides a computer program product, including a computer program that, when run on a computer, causes the computer to perform the steps of the intelligent metasurface four-corner sparse observation-assisted signal source localization method in Embodiment 1 above.
[0063] The relevant technical solutions are the same as above, and will not be repeated here.
[0064] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for locating a signal source with four-corner sparse observation assistance from an intelligent metasurface, characterized in that, include: Under the codebook configuration of the RIS unit, the observed samples are obtained after the transmitted signal of the signal source to be located is passively reflected by the RIS unit and received by the receiver. The codebook configuration is then changed to obtain an observation sequence consisting of T such observed samples. The RIS unit is only used in... The four corners of the virtual grid plane are arranged, with the symbol " " indicates transpose; The coordinates of the RIS units are normalized and then input into the implicit neural representation network. In this process, the complex field matrix of the virtual full aperture signal is obtained. Estimate , Represent the set of complex numbers; estimate the virtual full aperture complex field. Input the observation model to obtain the receiver output observation sequence. Estimate ; Including and Under a loss function including data consistency loss, the implicit neural representation network... During training, when the loss converges, the implicit neural representation network... Output the complex field matrix of the virtual full aperture signal The optimal estimate; The azimuth and elevation angles of the signal source to be located are determined based on the optimal estimate.
2. The signal source localization method assisted by four-corner sparse observation of intelligent metasurfaces according to claim 1, characterized in that, The loss function also includes physical regularization loss. The physical regularization loss The construction methods include: Construct the row prediction coefficient vector as follows The column-directed prediction coefficient vector is ;in, The preset physical regularization loss The maximum prediction order; The complex field matrix based on the virtual full aperture signal Estimate The row-direction prediction coefficient vector and the column-direction prediction coefficient vector are used to construct the prediction residuals: in, , These are the row-directed forecast residuals and the column-directed forecast residuals, respectively. The index of each grid within the virtual grid plane. , ; Construct the physical regularized loss using the predicted residuals. : in, The implicit neural representation network Network parameters.
3. The method for locating a signal source with four-corner sparse observation assistance based on a smart metasurface according to claim 1 or 2, characterized in that, The observation model is as follows: in, , For the RIS unit Secondary codebook configuration. , , This is the set of indices for the RIS cells arranged at the four corners. Indicates taking the absolute value; Indicates vector diagonalization; The equivalent complex gain vector from the four corner RIS units to the receiver; The selection matrix for the RIS units arranged at the four corners; The complex field matrix of the virtual full aperture signal Vectorization, Indicates vectorization operation, It is noise.
4. The signal source localization method assisted by four-corner sparse observation of intelligent metasurfaces according to claim 1, characterized in that, The coordinates of the RIS units are normalized and then input into the implicit neural representation network. In this process, the complex field matrix of the virtual full aperture signal is obtained. Estimate ,include: The coordinates of the RIS units are normalized and then input into the implicit neural representation network. In the process, the complex field of the signal at the coordinates of each RIS unit is obtained; based on the complex field of the signal at the coordinates of each RIS unit, sampling is performed on the virtual mesh plane to obtain the complex field matrix of the virtual full aperture signal. Estimate .
5. The signal source localization method assisted by four-corner sparse observation of intelligent metasurfaces according to claim 4, characterized in that, The coordinates of the RIS units are normalized and then input into the implicit neural representation network. Previously, it also included: performing position encoding on the coordinates of the normalized RIS units to obtain encoded features; the implicit neural representation network The complex signal field at the coordinates of the RIS unit is obtained based on the encoded features.
6. The signal source localization method assisted by four-corner sparse observation of intelligent metasurfaces according to claim 5, characterized in that, The location code is Position encoding of sine and cosine.
7. The signal source localization method assisted by four-corner sparse observation of intelligent metasurfaces according to claim 1, characterized in that, The data consistency loss for: in, For the first RIS unit The observation sequence output by the receiver under the secondary codebook configuration for The estimate, The implicit neural representation network Network parameters.
8. A signal source localization system assisted by four-corner sparse observation of an intelligent metasurface, characterized in that, Includes computer-readable storage media and processors; The computer-readable storage medium is used to store executable instructions; The processor is used to read executable instructions stored in the computer-readable storage medium and execute the signal source localization method assisted by four-corner sparse observation of the intelligent metasurface as described in any one of claims 1-7.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the intelligent metasurface four-corner sparse observation-assisted signal source localization method as described in any one of claims 1-7.
10. A computer program product, characterized in that, Includes a computer program that, when run on a computer, causes the computer to execute the intelligent metasurface four-corner sparse observation-assisted signal source localization method according to any one of claims 1-7.