Array antenna mutual coupling calibration method and system based on spatial multi-scale feature decoupling

Abstract

The application provides an array antenna mutual coupling calibration method and system based on spatial multi-scale feature decoupling, relates to the technical field of antenna testing, and comprises the following steps: placing a target antenna array in a full-wave anechoic chamber, connecting two ports to a vector network analyzer, and terminating the remaining ports with a matching load; measuring and obtaining the transmission coefficient between the two ports in a preset working frequency band to represent the coupling strength between the ports; repeatedly performing the measurement operation on multiple port pairs to obtain a scattering parameter matrix representing the mutual coupling effect; and comparing the scattering parameter with a preset technical index to determine whether the antenna isolation meets the requirements. The application can comprehensively and accurately represent the mutual coupling effect of a multi-channel antenna, avoid errors caused by reflection interference in the testing process, realize rapid determination and quality control of the isolation performance, and provide a reliable basis for the design optimization and engineering application of a base station antenna.

Description

Technical Field

[0001] This application relates to the field of antenna testing technology, and more specifically, to a method and system for mutual coupling calibration of array antennas based on spatial multi-scale feature decoupling. Background Technology

[0002] As a crucial component of mobile communication systems, base station antennas directly impact the coverage quality and transmission performance of wireless networks. Currently, base station antennas are evolving towards multi-channel, large-scale arrays, and electrically modulated designs to support wider operating bandwidths, higher antenna isolation, and more flexible beamforming capabilities. However, in multi-channel antenna arrays, due to the limited spatial distance between elements, complex feeding networks, and prominent edge effects, electromagnetic coupling can easily occur between different channel ports. This coupling effect not only leads to a decrease in antenna isolation but can also cause radiation pattern distortion, port impedance mismatch, and overall system performance degradation.

[0003] Therefore, how to provide a method that can efficiently and accurately test the mutual coupling effect of base station antennas under multi-channel conditions, and dynamically determine and compensate based on technical indicators, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application provides a method and system for mutual coupling calibration of array antennas based on spatial multi-scale feature decoupling.

[0005] In a first aspect, this application provides a method for mutual coupling calibration of array antennas based on spatial multi-scale feature decoupling, including:

[0006] The target antenna array is placed in a fully anechoic chamber, and the first and second ports of its multiple channel ports are connected to a vector network analyzer.

[0007] Terminate all remaining unconnected channel ports to a matching load;

[0008] The vector network analyzer measures and obtains the transmission coefficient between the first port and the second port within a preset operating frequency band to characterize the coupling strength between the two ports.

[0009] By iterating through multiple port pairs, repeatedly performing the steps of connecting the port pairs to the vector network analyzer, terminating the remaining unconnected channel ports with matched loads, and measuring the transmission coefficient, scattering parameters characterizing the antenna mutual coupling effect are obtained.

[0010] Optional, including:

[0011] The measured scattering parameters are compared with preset technical specifications to determine whether the antenna's isolation performance is up to standard.

[0012] Optional, including:

[0013] The scattering parameters of each element port of the target antenna array are measured at multiple scanning angles to obtain amplitude and phase test data characterizing the array mutual coupling effect.

[0014] A two-dimensional coupling feature map is constructed based on the amplitude and phase test data, wherein the pixel position of the two-dimensional coupling feature map corresponds one-to-one with the spatial position of each element of the target antenna array, and the pixel value corresponds to the scattering parameter measured at the corresponding element position.

[0015] Two-dimensional spatial multi-scale decomposition processing is performed on the two-dimensional coupling feature map to separate the low-frequency feature subset that characterizes the global slow-changing trend of the mutual coupling effect and the high-frequency feature subset that characterizes the local spatial perturbation of the mutual coupling effect.

[0016] Based on a parallel processing model including a first processing unit and a second processing unit, the low-frequency feature subset and the high-frequency feature subset are processed respectively. The first processing unit is used to calculate the low-frequency processing result characterizing the global gradual change trend, and the second processing unit is used to calculate the high-frequency processing result characterizing the local spatial perturbation. The processing results of the first processing unit and the second processing unit are subjected to feature mapping and weighted superposition to calculate the amplitude and phase calibration coefficient characterizing the amplitude and phase deviation of the current array mutual coupling effect, and the amplitude and phase calibration coefficient is output.

[0017] Based on the amplitude and phase calibration coefficients, a preset neural network quantization model is used to calculate and output test evaluation results that characterize the effectiveness of the amplitude and phase calibration coefficients in dynamically compensating for array mutual coupling effects.

[0018] Optionally, constructing a two-dimensional coupling feature map based on the amplitude and phase test data includes:

[0019] Based on the Cartesian coordinates of each element of the target antenna array in physical space, the amplitude and phase test data are mapped to the Cartesian coordinate system to obtain the initial physical domain coupling feature map.

[0020] Based on the beam scanning angle corresponding to the current measurement, an analysis domain coordinate system is constructed, wherein the first coordinate axis of the analysis domain coordinate system is parallel to the projection direction of the beam scanning direction in the target antenna array plane, and the second coordinate axis is orthogonal to the first axis;

[0021] An affine transformation operation is performed on the initial physical domain coupling feature map to map the amplitude and phase test data in the initial physical domain coupling feature map from the Cartesian coordinate system to the analysis domain coordinate system, thereby obtaining coupling feature data that is non-uniformly distributed in the analysis domain coordinate system.

[0022] The non-uniformly distributed coupling feature data is interpolated and resampled to obtain a two-dimensional coupling feature map with a uniform grid distribution in the coordinate system of the analysis domain.

[0023] Optionally, obtaining a two-dimensional coupled feature map with a uniform grid distribution in the analysis domain coordinate system includes:

[0024] Based on the spatial location of the array elements in the target antenna array, the non-uniformly distributed coupling feature data is divided into regions to generate at least one central region subset corresponding to the array central region element and at least one edge region subset corresponding to the array edge region element.

[0025] Optionally, obtaining a two-dimensional coupled feature map with a uniform grid distribution in the analysis domain coordinate system further includes:

[0026] The configuration includes a multi-channel interpolation kernel function set including a first interpolation channel and a second interpolation channel, wherein the first interpolation channel adopts a first interpolation kernel function for maintaining the smooth continuity of the amplitude and phase signals, and the second interpolation channel adopts a second interpolation kernel function for maintaining the high-frequency detail features of the amplitude and phase signals;

[0027] Interpolation processing is performed using only the central region subset and the first interpolation channel to obtain the basic coupling field distribution map;

[0028] Interpolation processing is performed using only the subset of the edge regions and the second interpolation channel to obtain a detailed perturbation field distribution map;

[0029] The basic coupled field distribution map and the detailed perturbation field distribution map are spatially adaptively weighted and fused. For each target pixel in the uniform grid, the fusion weight is determined according to the spatial distance between the target pixel and the data points in the edge region subset, and the two-dimensional coupled feature map is obtained.

[0030] Optionally, performing two-dimensional spatial multi-scale decomposition processing on the two-dimensional coupling feature map to separate a low-frequency feature subset characterizing the global slowly varying trend of the mutual coupling effect and a high-frequency feature subset characterizing the local spatial perturbation of the mutual coupling effect includes:

[0031] Based on the beam scanning angle corresponding to the two-dimensional coupling feature map, the primary and secondary decomposition axes of the two-dimensional spatial multi-scale decomposition are determined. The primary decomposition axis is parallel to the projection direction of the beam scanning direction in the target antenna array plane, and the secondary decomposition axis is orthogonal to the primary decomposition axis.

[0032] Optionally, the two-dimensional spatial multi-scale decomposition process includes:

[0033] The first-stage one-dimensional multi-scale decomposition process is performed independently on each row of data in the two-dimensional coupled feature map along the main decomposition axis to obtain the main axis low-frequency component map and the main axis high-frequency component map.

[0034] A second-stage one-dimensional multi-scale decomposition process is performed on the principal axis low-frequency component map and the principal axis high-frequency component map along the secondary decomposition axis, wherein:

[0035] Each column of data in the primary axis low-frequency component plot is subjected to one-dimensional low-pass filtering along the secondary decomposition axis to generate the low-frequency feature subset.

[0036] The low-frequency component map of the main axis is subjected to a one-dimensional high-pass filter along the secondary decomposition axis to obtain the first high-frequency component.

[0037] The high-frequency component map of the main axis is subjected to one-dimensional low-pass filtering and one-dimensional high-pass filtering along the secondary decomposition axis to obtain the second high-frequency component and the third high-frequency component, respectively.

[0038] The first high-frequency component, the second high-frequency component, and the third high-frequency component are collectively used as the high-frequency feature subset, wherein the high-frequency feature subset characterizes the different spatial scale features of the local spatial perturbation of the mutual coupling effect.

[0039] Optionally, the processing of the low-frequency feature subset and the high-frequency feature subset based on the parallel processing model including the first processing unit and the second processing unit includes:

[0040] The first processing unit processes the low-frequency feature subset to calculate the low-frequency processing result;

[0041] The high-frequency feature subset is processed by the second processing unit, wherein the second processing unit includes multiple parallel dedicated processing channels, each of which corresponds to a first high-frequency component, a second high-frequency component, and a third high-frequency component in the high-frequency feature subset, and performs independent processing on the first high-frequency component, the second high-frequency component, and the third high-frequency component to obtain multiple high-frequency dedicated processing results.

[0042] A context-guided attention fusion method is used to fuse the multiple high-frequency specialized processing results, wherein:

[0043] The low-frequency processing results are used as global context guidance information;

[0044] The fusion weights of the multiple high-frequency dedicated processing results are dynamically calculated based on the global context guidance information through a pre-configured attention network.

[0045] The multiple high-frequency dedicated processing results are weighted and summed based on the fusion weights to obtain the final high-frequency processing result.

[0046] Secondly, this application provides an array antenna mutual coupling calibration system based on spatial multi-scale feature decoupling, comprising:

[0047] The acquisition module is used to measure the scattering parameters of each element port of the target antenna array at multiple scanning angles, and obtain amplitude and phase test data characterizing the mutual coupling effect of the array.

[0048] The construction module is used to construct a two-dimensional coupling feature map based on the amplitude and phase test data, wherein the pixel position of the two-dimensional coupling feature map corresponds one-to-one with the spatial position of each element of the target antenna array, and the pixel value corresponds to the scattering parameter measured at the corresponding element position.

[0049] The decomposition module is used to perform two-dimensional spatial multi-scale decomposition processing on the two-dimensional coupling feature map to separate the low-frequency feature subset that characterizes the global slow-changing trend of the mutual coupling effect and the high-frequency feature subset that characterizes the local spatial perturbation of the mutual coupling effect.

[0050] The processing module is used to process the low-frequency feature subset and the high-frequency feature subset respectively based on a parallel processing model including a first processing unit and a second processing unit. The first processing unit is used to calculate the low-frequency processing result characterizing the global gradual change trend, and the second processing unit is used to calculate the high-frequency processing result characterizing the local spatial perturbation. The processing results of the first processing unit and the second processing unit are subjected to feature mapping and weighted superposition to calculate the amplitude and phase calibration coefficient characterizing the amplitude and phase deviation of the current array mutual coupling effect, and the amplitude and phase calibration coefficient is output.

[0051] The evaluation module is used to calculate, based on the amplitude and phase calibration coefficients, using a preset neural network quantization model, and generate and output test evaluation results characterizing the effectiveness of the amplitude and phase calibration coefficients in dynamically compensating for array mutual coupling effects.

[0052] Compared with existing technologies, this application constructs a two-dimensional coupling feature map by measuring the scattering parameters of each array element at multiple angles, effectively capturing the spatial distribution characteristics of the array mutual coupling effect. Furthermore, through two-dimensional spatial multi-scale decomposition technology, the mutual coupling effect is finely separated into a global slowly varying trend and local spatial perturbations. Parallel processing models of linear and nonlinear processing are used for targeted processing, effectively extracting the global spatial features and local perturbation details of the mutual coupling effect, and realizing high-precision calculation of the amplitude and phase calibration coefficients. At the same time, a neural network quantization model is used to realize the automated and objective evaluation of the effectiveness of dynamic compensation of the amplitude and phase calibration coefficients, avoiding the subjectivity and limitations of traditional manual experience judgment. Attached Figure Description

[0053] Figure 1 A flowchart of the array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling provided in this application;

[0054] Figure 2 A flowchart of a method for constructing a two-dimensional coupled feature map based on the amplitude and phase test data provided in this application;

[0055] Figure 3 A flowchart of a method for obtaining a two-dimensional coupled feature map with a uniform grid distribution in the coordinate system of the analysis domain, provided in this application;

[0056] Figure 4 A schematic diagram of the array antenna mutual coupling calibration system based on spatial multi-scale feature decoupling provided in this application;

[0057] Figure 5 A flowchart of a dynamic compensation testing method provided in this application. Detailed Implementation

[0058] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0059] See Figure 5 The flowchart of a dynamic compensation testing method provided in this application includes steps S401 to S404, wherein:

[0060] S401: Place the target antenna array in a fully anechoic chamber and connect the first and second ports of its multiple channel ports to a vector network analyzer.

[0061] S402: Terminate all remaining unconnected channel ports to a matching load;

[0062] S403: The vector network analyzer measures and obtains the transmission coefficient between the first port and the second port within a preset operating frequency band to characterize the coupling strength between the two ports.

[0063] S404: Traverse multiple port pairs, repeatedly execute the steps of connecting the port pairs to the vector network analyzer, terminating the remaining unconnected channel ports to the matched load, and measuring the transmission coefficient to obtain scattering parameters characterizing the antenna mutual coupling effect.

[0064] Optionally, it also includes: comparing the measured scattering parameters with preset technical specifications to determine whether the antenna's isolation specification is qualified.

[0065] In a specific implementation, first, place the target antenna array to be measured in an anechoic chamber to avoid the influence of external electromagnetic interference on the test results. The antenna array to be measured is a multi-channel electrically tunable structure, and each channel port has an independent feeding interface. Through a radio frequency coaxial cable, connect the first port and the second port in the antenna array to be measured to the test ports of a vector network analyzer respectively; for the remaining ports not participating in the test, terminate them with standard 50-ohm matching loads to eliminate unnecessary reflected signal interference.

[0066] Subsequently, use the vector network analyzer to perform S-parameter measurement operations within a preset operating frequency band. S-parameters are scattering parameters. Specifically, scan the transmission coefficient S21 between the first port and the second port to obtain the amplitude and phase responses at different frequency points, thereby quantifying the electromagnetic coupling strength between the two ports. By adjusting the scan step and bandwidth settings of the vector network analyzer, the measurement accuracy can be improved.

[0067] After completing the measurement of a pair of ports, switch the connection ports of the vector network analyzer, traverse all possible port pairs in the antenna array, and repeat the above steps. Apply a matching load to the remaining unmeasured ports for each measurement to ensure the consistency of the measurement conditions. By sequentially traversing all port pairs, a scattering parameter matrix covering the entire antenna array to be measured can be obtained. This matrix comprehensively characterizes the mutual coupling effect between the channels of the multi-channel electrically tunable base station antenna.

[0068] In another embodiment, it further includes: comparing the scattering parameter matrix with a preset technical index. The technical index is an isolation threshold set based on antenna design specifications and communication system performance requirements. By analyzing the amplitude values of the measured transmission coefficients, the isolation between each port pair can be calculated and compared with the isolation threshold one by one. When the isolation of all port pairs is greater than or equal to the isolation threshold, it is determined that the isolation performance of the antenna array is qualified; if the isolation of any port pair is less than the isolation threshold, it is determined that the isolation performance of the antenna is unqualified.

[0069] Furthermore, when the determination result is unqualified, combined with specific scattering parameter data, locate the channel pairs with strong mutual coupling effects and guide subsequent structural adjustment and compensation measures, such as optimizing the arrangement spacing of radiation units, introducing decoupling structures, or adding compensation circuits in the feeding network. Through the above measures, the overall isolation performance of the antenna array can be effectively improved.

[0070] This embodiment ensures the authenticity and accuracy of test data through per-port S-parameter measurement in an anechoic chamber and combined with the matching load termination method; through comparison with preset technical indexes, it realizes quantitative evaluation and determination of the mutual coupling effect of the antenna, and can provide a reliable basis for the design optimization and quality control of base station antennas.

[0071] Optional, see Figure 1 The diagram shows a flowchart of the array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling provided in this application, including steps S101 to S105, wherein:

[0072] S101: Measure the scattering parameters of each element port of the target antenna array at multiple scanning angles to obtain amplitude and phase test data characterizing the array mutual coupling effect;

[0073] S102: Construct a two-dimensional coupling feature map based on the amplitude and phase test data, wherein the pixel position of the two-dimensional coupling feature map corresponds one-to-one with the spatial position of each element of the target antenna array, and the pixel value corresponds to the scattering parameter measured at the corresponding element position.

[0074] S103: Perform two-dimensional spatial multi-scale decomposition processing on the two-dimensional coupling feature map to separate the low-frequency feature subset that characterizes the global slow-changing trend of the mutual coupling effect and the high-frequency feature subset that characterizes the local spatial perturbation of the mutual coupling effect.

[0075] S104: Based on a parallel processing model including a first processing unit and a second processing unit, the low-frequency feature subset and the high-frequency feature subset are processed respectively, wherein the first processing unit is used to calculate the low-frequency processing result characterizing the global gradual change trend, and the second processing unit is used to calculate the high-frequency processing result characterizing the local spatial disturbance; the processing results of the first processing unit and the second processing unit are subjected to feature mapping and weighted superposition to calculate the amplitude and phase calibration coefficient characterizing the amplitude and phase deviation of the current array mutual coupling effect, and the amplitude and phase calibration coefficient is output.

[0076] S105: Based on the amplitude and phase calibration coefficient, a preset neural network quantization model is used to calculate and output test evaluation results characterizing the effectiveness of the amplitude and phase calibration coefficient in dynamically compensating for array mutual coupling effects.

[0077] Regarding the above S101:

[0078] Element ports refer to the connection interfaces between each radiating element in the target antenna array and external test circuits or feed networks, used for inputting or outputting radio frequency signals. Array mutual coupling effect refers to the electromagnetic coupling between elements in the target antenna array; that is, the radiation or scattering behavior of one element affects the radiation characteristics of adjacent or other elements, manifested as the correlation or change in the amplitude and phase of scattering parameters between elements. Amplitude and phase test data are the scattering parameter data of the element ports obtained by measuring with a vector network analyzer, specifically including the amplitude and phase information of the signals at each port.

[0079] In practice, the first step is to build a test system for measuring the scattering parameters of the target antenna array. This test system may include a vector network analyzer, RF cables, a signal calibration kit, a turntable, and the target antenna array under test.

[0080] The target antenna array under test is mounted and fixed on a turntable. The turntable is used to precisely control the rotation of the array relative to the incident beam direction to achieve parameter measurement at different scanning angles. Before testing, the test channel consisting of the vector network analyzer and the cable is rigorously calibrated using a signal calibration kit, such as standard short-circuit, open-circuit, or load components, to eliminate measurement link errors.

[0081] After calibration, the vector network analyzer is connected sequentially to the ports of each element in the target antenna array, and the measurement program is initiated. Specifically, at each preset scanning angle, the vector network analyzer excites each element port of the array and measures the reflection parameters of each port and the transmission parameters between ports. During each measurement, the scattering parameter data measured at each element port at the corresponding scanning angle, including amplitude and phase information, is recorded to form a complete scattering parameter measurement dataset.

[0082] Subsequently, the turntable rotates the antenna array to the next preset scanning angle and repeats the above measurement process. The entire measurement process sequentially covers multiple preset beam scanning angles, for example, scanning step by step at 5° intervals within the range of -45° to +45°, to obtain complete array multi-angle amplitude and phase test data.

[0083] In this way, amplitude and phase test data that fully characterize the array mutual coupling effect can be obtained for subsequent processing steps.

[0084] Scattering parameters (S-parameters) refer to the port characteristic parameters of high-frequency electronic systems or devices. They are commonly used to characterize the reflection and transmission characteristics of signals between ports. For example, the reflection coefficient S11 represents the signal reflection at the input port, and the transmission coefficient S21 represents the signal transmission from port 1 to port 2.

[0085] In practical implementation, for an N-port antenna array, it is necessary to measure a complete N×N scattering parameter matrix.

[0086] The calibration kit is a dedicated device for calibrating test links for vector network analyzers. It can include standard short circuits, open circuits, and loads to eliminate measurement link errors and improve the accuracy of test data.

[0087] Regarding S102 above:

[0088] Based on the amplitude and phase test data of array mutual coupling effect, a two-dimensional coupling feature map that can intuitively characterize the mutual coupling characteristics between array elements is further constructed.

[0089] In practice, based on the specific physical layout of the target antenna array, the position of each element in the array is represented using a two-dimensional Cartesian coordinate system. For example, with the geometric center of the antenna array as the origin, mutually orthogonal two-dimensional coordinate axes are established, mapping the actual physical position of each element relative to the center to two-dimensional coordinate values, thereby obtaining the position coordinate information of each element in the array. A two-dimensional coupling feature map is constructed based on these element position coordinates. Specifically, this two-dimensional coupling feature map can be viewed as a spatial distribution map, where the position of each pixel strictly corresponds one-to-one with the position of each element in the target antenna array, and each pixel value is taken from the scattering parameter data measured at the corresponding element. The scattering parameter data can be processed by unifying the measured amplitude and phase, for example, by encoding them as complex numbers or other comprehensive representations, to fully express the electromagnetic coupling characteristics between the elements.

[0090] For example, for each scanning angle, a reflection coefficient feature map is constructed, and the complex value of each pixel is the reflection coefficient of the corresponding array element at that scanning angle.

[0091] For example, to characterize the overall coupling strength of each element, the pixel value can be a scalar obtained by calculating the sum or root mean square of the magnitudes of the coupling coefficients between that element and all other elements.

[0092] The electromagnetic radiation or scattering behavior of each element in the target antenna array exhibits a significant spatial correlation. Due to the radiative coupling of electromagnetic waves, the elements influence each other in amplitude and phase characteristics. Constructing a two-dimensional coupling feature map based on the spatial location of the element scattering parameter data can effectively capture and reveal this spatial correlation, thus providing an intuitive and effective data foundation for subsequent analysis and compensation of mutual coupling effects.

[0093] The above process is repeated for each preset scanning angle, that is, a series of complete two-dimensional coupled feature maps are constructed at multiple different scanning angles.

[0094] Regarding the above S103:

[0095] In step S103, based on the two-dimensional coupling feature map constructed in step S102, a two-dimensional spatial multi-scale decomposition process is performed to effectively separate the global slow-varying trend of the mutual coupling effect, i.e., the low-frequency spatial features, and the local spatial perturbation, i.e., the high-frequency spatial features, from the two-dimensional coupling feature map.

[0096] In practice, based on the beam scanning angle of the target antenna array, the directional axes for multi-scale decomposition in two-dimensional space are determined, including a primary decomposition axis and a secondary decomposition axis. The primary decomposition axis is chosen to be parallel to the projection direction of the current scanning beam direction onto the antenna array plane, while the secondary decomposition axis is perpendicular to the primary decomposition axis, forming an orthogonal decomposition reference in two-dimensional space. The reason for choosing these decomposition directions is that the mutual coupling effect of the target antenna array typically exhibits a clear trend change along the beam scanning direction, while directions orthogonal to the beam scanning direction show obvious local variation characteristics. Therefore, choosing this orthogonal coordinate axis can effectively capture the directional characteristics of the mutual coupling feature.

[0097] Subsequently, the two-dimensional coupled feature map is decomposed into spatial frequency using two-dimensional spatial multi-scale decomposition technology. Two-dimensional discrete wavelet transform or similar spatial frequency analysis tools can be used to decompose the two-dimensional coupled feature map into feature subsets of different spatial scales step by step.

[0098] Specifically, a one-dimensional multi-scale decomposition operation is first performed along the principal decomposition axis. That is, each row of the two-dimensional coupled feature map (along the principal decomposition axis) is decomposed row by row, generating an initial low-frequency component map and an initial high-frequency component map along the principal decomposition axis. The initial low-frequency component map mainly represents the global gradual change trend information of the two-dimensional coupled feature map along the principal decomposition axis, while the initial high-frequency component map mainly represents the local change features along the principal decomposition axis.

[0099] After completing the one-dimensional decomposition along the primary decomposition axis, a second-stage one-dimensional multi-scale decomposition is performed on the initial low-frequency component map along the secondary decomposition axis. This involves decomposing each column of the initial low-frequency component map (along the secondary decomposition axis) to obtain a subset of low-frequency features. This subset of low-frequency features reflects the global gradual variation trend in the entire two-dimensional coupled feature map.

[0100] Simultaneously, one-dimensional multi-scale high-pass and low-pass filtering processes are performed on the initial low-frequency component map and the initial high-frequency component map along the secondary decomposition axis to obtain multiple high-frequency components. These high-frequency components together form a high-frequency feature subset, which is used to characterize the local spatial perturbation features in the two-dimensional coupling feature map, namely the local coupling interference and non-uniform variation characteristics between array elements.

[0101] The two-dimensional spatial multi-scale decomposition processing method described above can clearly separate feature information at different spatial scales from the original two-dimensional coupled feature map.

[0102] Regarding S104 above:

[0103] In specific implementation, the parallel processing model described in step S104 can be implemented as a Linear Non-linear Network (LNN). The LNN model includes a linear processing unit (first processing unit) and a non-linear processing unit (second processing unit), which process the low-frequency feature subset and high-frequency feature subset obtained in step S103 in parallel to obtain the low-frequency processing result and the high-frequency processing result, respectively.

[0104] In practical implementation, firstly, for the low-frequency feature subset, a spatial linear mapping process is performed by a linear processing unit. This can be achieved, for example, using a spatial linear filter, a spatial smoothing filter, or a low-order linear polynomial regression model, to capture the spatially global, gradually varying trend of array mutual coupling effects from the low-frequency feature subset. This linear processing effectively preserves the gradually varying trend information contained in the low-frequency feature subset and outputs low-frequency processing results that characterize the global mutual coupling trend.

[0105] Simultaneously, for the high-frequency feature subset, a nonlinear processing unit performs spatial nonlinear mapping processing, which can be implemented using a convolutional neural network (CNN) or other nonlinear feature extraction networks. Specifically, a feature extraction network composed of multiple convolutional layers and nonlinear activation functions, such as ReLU or Sigmoid, can be designed to effectively extract the local spatial perturbation patterns reflected in the high-frequency feature subset. This nonlinear processing can effectively capture and amplify the detailed features of the local perturbation region and output high-frequency processing results to characterize the local mutual coupling perturbations between array elements.

[0106] After the linear and nonlinear processing units have completed their parallel processing, the LNN fuses the outputs of the two units through a feature mapping and spatial weighted superposition module. The specific fusion process is as follows:

[0107] First, feature normalization is performed on the low-frequency processing results output by the linear processing unit and the high-frequency processing results output by the nonlinear processing unit to make them consistent in terms of numerical scale, feature space dimension, and spatial resolution.

[0108] Secondly, based on the spatial physical characteristics or amplitude and phase distribution of the array, a spatial weight matrix is ​​calculated. The values ​​of the elements of this matrix are related to the spatial position of each array element or the local amplitude information in the low-frequency processing results, and are used to dynamically adjust the spatial fusion ratio of the low-frequency processing results and the high-frequency processing results.

[0109] Subsequently, the two processing results are weighted and superimposed using a spatial weight matrix. Specifically, a pixel-by-pixel weighted linear combination method can be used to complete the fusion, obtaining a complete amplitude and phase calibration coefficient. This amplitude and phase calibration coefficient simultaneously includes the linear component of the global trend of the array mutual coupling effect and the nonlinear component of the local spatial perturbation, which can more accurately characterize the amplitude and phase deviation characteristics caused by the array mutual coupling effect at the current scanning angle.

[0110] Finally, the calculated amplitude and phase calibration coefficients are used as the output of this step S104 for the next step S105 to perform dynamic compensation effectiveness testing and evaluation.

[0111] It should be noted that the linear processing branch can be constructed as a multi-layer linear convolutional structure, such as consisting of several linear convolutional layers without activation functions and spatial smoothing layers. Specifically, 2 to 4 linear convolutional layers can be set, and the kernel size of each layer can be selected as 3×3, 5×5, or 7×7 to adapt to different spatial scale requirements. The number of convolutional kernels is determined according to actual needs, such as 16 to 64 convolutional kernels, to represent the spatially gradual trend. Subsequently, after the convolutional output, a spatial smoothing layer or a low-order polynomial fitting layer can be added to further extract low-frequency spatial features.

[0112] The nonlinear processing branch can employ a convolutional neural network structure, consisting of several convolutional layers, nonlinear activation layers, and local feature enhancement modules. For example, 3 to 6 convolutional layers can be selected, with a nonlinear activation function following each convolutional layer. The kernel size can be 3×3 or 5×5, and the number of kernels increases progressively, for example, from 16 to 32 to 64, to achieve progressively deeper extraction of local spatial perturbation features.

[0113] The spatial weight matrix module can be implemented as an attention fusion network, for example, based on a spatial attention mechanism. The network input includes the outputs of linear and non-linear processing branches, such as 1 to 3 convolutional or fully connected layers for feature extraction. Then, the sigmoid function is used to generate a spatial attention weight distribution matrix, which dynamically determines the spatial fusion weights of the linear and non-linear processing results.

[0114] LNN models can be trained using supervised learning, which involves collecting a certain amount of array mutual coupling effect test data and corresponding ideal or expected amplitude-phase calibration coefficients as a training set, using the error backpropagation algorithm to train the network, and optimizing the network's weight parameters so that the network output gradually approaches the expected amplitude-phase calibration coefficients.

[0115] Thus, the low-frequency feature subset mainly reflects the amplitude and phase variation trends of the target antenna array as a whole or within a large spatial range. These trends typically exhibit a smooth and continuous spatial distribution with relatively stable variation patterns. This global, gradually varying trend is essentially closely related to the array's structure, layout, and the physical characteristics of electromagnetic wave propagation, exhibiting obvious linear or approximately linear spatial distribution characteristics. Therefore, using a linear processing unit to perform spatial linear fitting or spatial smoothing analysis on the low-frequency features can reliably extract the overall distribution trend of array mutual coupling effects with relatively low model complexity, providing a foundation for the global analysis of mutual coupling effects.

[0116] On the other hand, the high-frequency feature subset reflects the spatial local perturbations caused by the interaction between array elements in the array mutual coupling effect. These local perturbations usually have obvious nonlinear characteristics, manifested as rapid fluctuations or abrupt changes in amplitude and phase within local regions, which cannot be effectively captured by linear methods alone. Therefore, it is necessary to fully extract and express these local nonlinear perturbation patterns through nonlinear processing units, such as nonlinear feature extraction tools like convolutional neural networks.

[0117] However, since the mutual coupling effect in real-world target antenna arrays often exhibits both of the aforementioned characteristics—namely, a linearly varying trend and nonlinear local perturbations—it is impossible to effectively characterize the array mutual coupling effect using only linear or nonlinear processing methods. Therefore, by employing eigenmaps and spatial weighted superposition, the importance of both characteristics can be dynamically considered, resulting in amplitude and phase calibration coefficients that accurately reflect the true distribution characteristics of the array mutual coupling effect.

[0118] Regarding the above S105:

[0119] Since the compensation for the mutual coupling effect of the target antenna array in practical applications cannot usually be directly evaluated by the amplitude and phase calibration coefficients themselves, it is necessary to further construct an evaluation model to analyze and test the effectiveness of the above amplitude and phase calibration coefficients in the dynamic compensation of the actual array mutual coupling effect.

[0120] In practice, the preset neural network quantization model can be an evaluation model based on a deep neural network structure, such as a convolutional neural network, a deep feedforward network, or other similar neural network structures. The input of this neural network quantization model is the amplitude and phase calibration coefficients obtained in step S104, and the output is an evaluation index or evaluation result used to quantitatively characterize the effectiveness of the amplitude and phase calibration coefficients in compensating for mutual coupling effects, such as mutual coupling suppression degree, array element radiation efficiency improvement rate, or beamforming quality index.

[0121] Specifically, the following steps can be performed during model building and implementation:

[0122] First, the input of the neural network quantization model receives the amplitude and phase calibration coefficient data obtained in step S104. In actual implementation, the input amplitude and phase calibration coefficient data can be organized into a two-dimensional matrix or tensor, where each element of the matrix or tensor corresponds to the calibration amplitude and phase information of each element position in the target antenna array.

[0123] Secondly, the internal structure of a neural network quantization model may specifically include:

[0124] One or more feature extraction modules are used to extract the spatial correlation features implied by the amplitude and phase calibration coefficients. For example, a convolutional network consisting of multiple convolutional layers and nonlinear activation functions can be used, with the number of convolutional kernels increasing layer by layer from 16 to 64, and the kernel size can be set to 3×3 or 5×5 for example, to gradually extract deeper features.

[0125] One or more feature integration modules are used to globally integrate the extracted multi-scale spatial features to generate a compact feature representation. For example, a global pooling layer or a self-attention mechanism can be used to summarize the spatial dimensions of features in order to effectively extract global feature information.

[0126] One or more prediction output modules are used to output the final test evaluation results. For example, they may consist of one or two fully connected layers, wherein the number of neurons in the fully connected layers may be set between 32 and 128, and finally connected to a non-linear activation function, such as Sigmoid or Softmax, to generate the final numerical or categorical evaluation results.

[0127] For example, the numerical evaluation results output by the network can be the mutual coupling suppression degree, which assesses the attenuation of the array mutual coupling effect before and after compensation, or the element radiation efficiency improvement rate, which characterizes the proportion of actual radiation efficiency improvement of each element after applying the amplitude and phase calibration coefficient. If the network output is a categorical evaluation result, it can represent the probability or confidence level that the compensation effect reaches the preset standard, such as categories like "high-efficiency compensation", "medium-efficiency compensation", and "low-efficiency compensation".

[0128] Regarding model training methods, a large amount of actual mutual coupling effect data measured under different array configurations and scanning angles can be collected as a training sample set. During training, known real or expected mutual coupling compensation effectiveness indices are used as label data. A supervised training method is adopted, and the network is trained through the error backpropagation algorithm, so that the network gradually learns a stable mapping relationship between the input amplitude and phase calibration coefficients and the output effectiveness index.

[0129] For example, if the input amplitude and phase calibration coefficients can effectively eliminate or reduce beam distortion or gain loss caused by the interaction between array elements in the array, the output evaluation result after feature recognition and processing by the neural network quantization model will show a high mutual coupling suppression degree or array element radiation efficiency improvement index, thus intuitively reflecting the effectiveness of the current amplitude and phase calibration coefficients in compensating for the actual array mutual coupling effect.

[0130] In this way, the automated quantitative evaluation of the array mutual coupling effect compensation effect is effectively realized, eliminating the subjectivity and uncertainty that may exist in the traditional method of relying on manual experience judgment or simple amplitude-phase comparison; by utilizing the fitting ability of the neural network quantization model, the nonlinear and high-dimensional feature information hidden in the amplitude-phase calibration coefficient can be efficiently captured, significantly improving the accuracy and reliability of the evaluation of the effectiveness of dynamic compensation of array mutual coupling effect.

[0131] As an optional implementation method, see [link to implementation details]. Figure 2 The flowchart of a method for constructing a two-dimensional coupled feature map based on the amplitude and phase test data provided in this application includes steps S201 to S204, wherein:

[0132] S201: Based on the Cartesian coordinates of each element of the target antenna array in physical space, the amplitude and phase test data are mapped to the Cartesian coordinate system to obtain the initial physical domain coupling feature map;

[0133] S202: Based on the beam scanning angle corresponding to the current measurement, construct an analysis domain coordinate system, wherein the first coordinate axis of the analysis domain coordinate system is parallel to the projection direction of the beam scanning direction in the target antenna array plane, and the second coordinate axis is orthogonal to the first axis;

[0134] S203: Perform an affine transformation operation on the initial physical domain coupling feature map to map the amplitude and phase test data in the initial physical domain coupling feature map from the Cartesian coordinate system to the analysis domain coordinate system, thereby obtaining coupling feature data that is non-uniformly distributed in the analysis domain coordinate system;

[0135] S204: Perform interpolation and resampling processing on the non-uniformly distributed coupling feature data to obtain a two-dimensional coupling feature map with a uniform grid distribution in the coordinate system of the analysis domain.

[0136] Because the spatial orientation between the array radiation direction and the array plane changes with the beam scanning angle of a real antenna array, directly using the initial physical domain, i.e., a fixed Cartesian coordinate system, to analyze the array mutual coupling effect at different scanning angles may result in a spatially irregular or non-uniform distribution of the obtained two-dimensional coupling feature map, which is detrimental to subsequent spatial analysis and feature extraction. Therefore, an analysis domain coordinate system can be dynamically constructed based on the beam scanning angle. Through coordinate transformation and interpolation resampling, the final two-dimensional coupling feature map can exhibit a uniform and regular grid distribution in the analysis domain coordinate system, thereby improving the accuracy and stability of subsequent spatial frequency decomposition and feature processing.

[0137] In practice, when determining the coordinate system of the analysis domain, the first coordinate axis (main axis) of the analysis domain is defined as parallel to the projection direction of the beam scanning direction in the array plane, and the second coordinate axis (secondary axis) is orthogonal to the main axis, based on the current beam scanning angle.

[0138] The analysis domain coordinate system determined by the above method has a clear physical meaning. That is, the direction of the first coordinate axis is consistent with the propagation direction of the main radiation beam of the antenna array, so that the main spatial variation trend of the mutual coupling effect can be clearly reflected in the analysis domain. The second coordinate axis corresponds to the spatial variation direction orthogonal to the scanning beam direction, which is the direction of obvious local disturbance.

[0139] After determining the coordinate system of the analysis domain, an affine transformation operation is further performed on the initial physical domain coupling feature map. In specific implementation, with the geometric center of the antenna array as the origin, the array data (scattering parameter values) in the Cartesian coordinate system of the initial physical domain are mapped to the coordinate system of the analysis domain through rotation and translation transformations, so that the transformed data is distributed in the new coordinate system. However, the data distribution at this time usually exhibits non-uniform characteristics.

[0140] For example, taking a target antenna array, in the initial physical domain coordinate system, the coordinates (x, y) of each array element and the corresponding amplitude and phase test data have been obtained. After the beam scanning angle is determined, the initial array element coordinates (x, y) are transformed into new coordinates (u, v) through a pre-calculated rotation angle. The new coordinates (u) are parallel to the projection of the beam scanning direction, and the coordinates (v) are perpendicular to it. After the affine transformation is completed, the scattering parameter data of the array elements are located in the (u, v) coordinates of the analysis domain coordinate system. However, since the initial array element layout is often a regular rectangular or approximately rectangular layout in the physical coordinate system, when the coordinate transformation occurs, these data usually cannot maintain the original regular grid distribution state, but become a scattered and irregular spatial distribution, which is difficult to use directly for subsequent spatial frequency analysis.

[0141] Therefore, further interpolation resampling is required to obtain a regular, uniform rasterized two-dimensional coupled feature map in the analysis domain coordinate system. Interpolation resampling involves calculating and filling new data points onto a regular grid based on existing discrete, non-uniform data points to obtain a uniform raster data distribution that meets the needs of subsequent analysis.

[0142] For example, a bicubic interpolation method is used to define a regular uniform grid (u', v') in the analysis domain coordinate system. Based on scattered (u, v) data points, the scattering parameter value at each (u', v') position is calculated using the interpolation method. After the above interpolation process, a two-dimensional coupled feature map with a regular spatial grid structure in the analysis domain coordinate system is finally obtained.

[0143] As an optional implementation, after mapping the amplitude and phase test data in the initial physical domain coupling feature map to the analysis domain coordinate system through affine transformation, the coupling feature data of the array elements generally exhibits a non-uniform distribution. This non-uniform distribution is mainly due to the rotation, translation, or scaling of the element positions caused by the affine transformation, resulting in significant differences in the distribution density of the array element data points in the analysis domain coordinate system.

[0144] For example, the array elements in the central region of an antenna array are often densely and regularly distributed, while the array elements in the edge region are relatively sparsely distributed or irregularly distributed. Therefore, if a uniform interpolation method is directly applied to all array data, it may result in the smoothing or loss of detailed information in the data of the array edge region, or excessive interpolation in the central region, causing unnecessary computational overhead and information redundancy.

[0145] To address the aforementioned issues, the non-uniformly distributed coupling feature data needs to be regionalized based on the spatial positions of each element in the target antenna array within the analysis domain coordinate system, forming at least one central region subset and at least one edge region subset. This regionalization aims to better achieve differentiated processing of the spatial data, employing different interpolation strategies for the central and edge regions to achieve more accurate and efficient construction of the two-dimensional coupling feature map.

[0146] In practical implementation, based on the actual layout of the target antenna array, the distance from the geometric center of the array to the center point is calculated using the array's geometric center as the reference origin. Specifically, the Euclidean or Manhattan distance between the element's position and the array's geometric center can be used to determine whether an element belongs to the central or peripheral region. A distance threshold can be set, classifying elements closer to the center point as a subset of the central region and those farther away as subsets of the peripheral region. This distance threshold can be flexibly set according to the specific array size, element distribution density, and actual application requirements. For example, elements within a certain distance from the center point, such as 30% to 50% of the array's overall diameter, can be classified as the central region, while elements exceeding this threshold can be classified as peripheral regions.

[0147] For example, taking a rectangular target antenna array with 64 elements, the position of each element is determined in the analysis domain coordinate system, using the array's geometric center as the reference point: if the distance of an element from the array's geometric center is less than a certain proportion of the array's maximum radius, it is classified as a central region subset; if it is greater than this proportion, it is classified as an edge region subset. After this region division, two parts of data are obtained: a central region subset and an edge region subset. The data point density in the central region subset is higher and the layout is more regular, while the data point density in the edge region subset is lower and the spatial layout is relatively irregular.

[0148] After the regions are divided, different interpolation strategies are applied to the data points of the central and peripheral subsets to obtain regular and uniform two-dimensional coupled feature maps in the analysis domain coordinate system. For example, since the data points in the central region are dense and have good regularity, a smooth interpolation strategy can be used to ensure the stability and continuity of the data in this region; while since the data in the peripheral region is sparse and irregular, an interpolation strategy that preserves local high-frequency details can be used.

[0149] High-frequency detail features refer to the rapidly changing local field disturbances caused by structural discontinuities at the antenna edge, which correspond to high-frequency components in space. To effectively preserve these features, interpolation kernel functions with high-pass or band-pass characteristics can be used for subsequent processing to avoid smoothing or losing these important local disturbance information during the interpolation process.

[0150] By employing the aforementioned region partitioning method, interpolation strategies can be optimized separately for the different data characteristics of the array center and edges. This effectively improves the overall accuracy and detail preservation of the two-dimensional coupled feature map, especially enhancing the local detail representation effect in the edge regions. Simultaneously, the differentiated processing approach avoids data processing redundancy in the central region, reduces unnecessary computational overhead, and improves the computational efficiency and accuracy of the entire feature map generation process.

[0151] As an optional implementation, see [link to implementation details]. Figure 3 The flowchart provided in this application describes a method for obtaining a two-dimensional coupled feature map with a uniform grid distribution in the coordinate system of the analysis domain, including steps S301 to S304, wherein:

[0152] S301: Configure a multi-channel interpolation kernel function set including a first interpolation channel and a second interpolation channel, wherein the first interpolation channel adopts a first interpolation kernel function for maintaining the smooth continuity of the amplitude and phase signals, and the second interpolation channel adopts a second interpolation kernel function for maintaining the high-frequency detail features of the amplitude and phase signals;

[0153] S302: Perform interpolation processing using only the central region subset and the first interpolation channel to obtain the basic coupling field distribution map;

[0154] S303: Perform interpolation processing using only the edge region subset and the second interpolation channel to obtain a detailed perturbation field distribution map;

[0155] S304: Perform spatially adaptive weighted fusion of the basic coupled field distribution map and the detailed perturbation field distribution map. For each target pixel in the uniform grid, determine the fusion weight based on the spatial distance between the target pixel and the data points in the edge region subset to obtain the two-dimensional coupled feature map.

[0156] To more effectively achieve this differentiated processing, a multi-channel interpolation kernel function set, including a first interpolation channel and a second interpolation channel, can be introduced. The purpose of this multi-channel interpolation kernel function set is to use kernel functions with different interpolation characteristics for the spatial variation characteristics of the mutual coupling features of array elements, so as to more accurately express the spatial continuity and smoothness of the two-dimensional coupled feature map and the high-frequency detail features of local regions.

[0157] Specifically, the first interpolation channel uses a first interpolation kernel function, which is characterized by smoothness and low-pass properties, making it suitable for data processing of a subset of the array's central region. The central region is typically the core radiation area of ​​the target antenna array, with densely packed and regularly distributed array elements, exhibiting a relatively stable and continuous spatial variation trend in mutual coupling effects. Utilizing the first interpolation kernel function effectively maintains the smoothness and continuity of the amplitude and phase signals in the central region, avoiding the introduction of unnecessary high-frequency disturbances during the interpolation process, and facilitating the accurate extraction of subsequent low-frequency features.

[0158] The second interpolation kernel function used in the second interpolation channel has high-pass or band-pass characteristics, making it more suitable for data processing of edge region subsets. Because the array elements in the edge regions are relatively sparse and irregularly distributed, the mutual coupling effect often manifests as significant local spatial perturbations, exhibiting rich high-frequency detail features. The second interpolation kernel function can effectively capture and retain this local high-frequency information, preventing important detail perturbation information from being smoothed or lost during the interpolation process, thereby improving the integrity of the detail information in the edge regions.

[0159] For example, a Gaussian or low-order polynomial function can be selected as the first interpolation kernel function to ensure high spatial continuity and smoothness of the interpolated basic coupled field distribution map; while the second interpolation kernel function can be selected as a radial basis function or a spline interpolation kernel function, which has good high-frequency detail preservation ability and is suitable for generating the distribution map of the perturbation field of the edge region.

[0160] In the specific interpolation operation, interpolation processing is performed only using the central region subset and the first interpolation channel to obtain the basic coupled field distribution map. This basic coupled field distribution map mainly reflects the spatial smooth and gradual change trend of the array mutual coupling effect. At the same time, interpolation processing is performed only using the edge region subset and the second interpolation channel to obtain the detailed perturbation field distribution map. This detailed perturbation field distribution map completely preserves the rich high-frequency spatial perturbation features of the edge region.

[0161] After obtaining the field distribution maps of the two different spatial features mentioned above, a spatially adaptive weighted fusion is further performed to construct the final complete two-dimensional coupled feature map. The specific fusion method is as follows:

[0162] First, for each target pixel in the uniform grid within the analysis domain coordinate system, the corresponding pixel value is obtained from both the basic coupled field distribution map and the detail perturbation field distribution map. Then, based on the spatial distance between the target pixel and the actual data points in the edge region subset, the fusion weight at that pixel is dynamically determined.

[0163] For example, for target pixels that are close to data points in the edge region, the detail perturbation field distribution map is given a higher weight to ensure that local high-frequency information can be effectively represented; while for pixels that are close to the center region and far from data points in the edge region, the fusion weight of the basic coupling field distribution map is increased to highlight the continuity and smoothness of the overall spatial trend.

[0164] In this way, by dynamically adjusting the fusion ratio of the two interpolation results, efficient and accurate fusion between the basic coupling field and the smoothing trend and the detailed perturbation features is achieved, ensuring that the final generated two-dimensional coupling feature map can accurately reflect the mutual coupling trend of the overall array and effectively retain the high-frequency detailed features of the local area, thereby more comprehensively and finely expressing the mutual coupling effect features of the actual target antenna array.

[0165] As an optional implementation, in the two-dimensional spatial multi-scale decomposition process in step S103, in order to more effectively separate the feature information of different spatial scales in the two-dimensional coupling feature map, it is necessary to dynamically determine the main decomposition axis and the secondary decomposition axis of the two-dimensional spatial decomposition according to the specific beam scanning angle, so as to more accurately capture the directional features of the array mutual coupling effect in space.

[0166] During actual operation, the mutual coupling effect of array elements at different scanning angles exhibits significant spatial directivity. Typically, the spatial mutual coupling effect along the beam scanning direction shows a relatively smooth, gradually varying trend, with prominent low-frequency spatial characteristics. However, in directions perpendicular to the beam scanning direction, the mutual coupling effect manifests as obvious local spatial disturbances or high-frequency characteristics. Therefore, determining the axis aligned with the beam scanning direction as the primary decomposition axis can more effectively capture the global trend of the mutual coupling effect, while axes orthogonal to this direction serve as secondary decomposition axes, helping to fully extract the local disturbance characteristics of the mutual coupling effect.

[0167] In practical implementation, based on the current beam scanning angle, the projection direction of the antenna array's beam scanning direction in the analysis domain coordinate system onto the array plane is first determined. This projection direction clearly reflects the geometric relationship between the antenna array's main radiation direction and the array's spatial layout. Therefore, the principal decomposition axis of the two-dimensional spatial multi-scale decomposition is defined as parallel to the aforementioned beam scanning projection direction, while the secondary decomposition axis is defined as a direction orthogonal to the principal decomposition axis.

[0168] For example, for a rectangular antenna array, if the beam scanning direction is horizontal to the array plane, the primary decomposition axis is the horizontal axis, and the secondary decomposition axis is the vertical axis perpendicular to it. If the scanning direction changes with the beam scanning angle, the primary and secondary decomposition axes also rotate accordingly to keep the primary decomposition axis parallel to the beam scanning projection direction. This method of dynamically determining the axes ensures that the decomposition process matches the radiation characteristics of the actual antenna array, effectively improving the relevance and accuracy of spatial frequency analysis.

[0169] By employing the above method, spatial multi-scale decomposition along the primary decomposition axis can efficiently capture and separate the low-frequency spatial components of the array's mutual coupling effect, i.e., the global gradual variation trend of the array; while multi-scale decomposition along the secondary decomposition axis can accurately capture and highlight the array's local spatial perturbation features or high-frequency components. For example, when the scanning angle gradually changes from -45° to +45°, the above axis determination and multi-scale decomposition processing can be performed on the two-dimensional coupling feature map at each angle to obtain more targeted low-frequency and high-frequency feature subsets closely related to each angle.

[0170] This effectively avoids the loss of directional information caused by a single fixed axis, and enhances the adaptability and robustness of two-dimensional spatial multi-scale analysis.

[0171] As an optional implementation, in the two-dimensional spatial multi-scale decomposition process described in step S103, in order to more efficiently and accurately separate the global gradual trend and local spatial disturbance characteristics of the mutual coupling effect of the target antenna array, a two-dimensional multi-scale decomposition technique in stages along the main decomposition axis and the secondary decomposition axis can be adopted.

[0172] In practice, after determining the primary and secondary decomposition axes, the data in the two-dimensional coupled feature map is treated as a two-dimensional matrix, with its rows and columns corresponding to the spatial distribution of the primary and secondary decomposition axes, respectively. Since the direction of the primary decomposition axis usually corresponds to the spatial trend of beam scanning, the first stage uses one-dimensional multi-scale decomposition to process each row of data in the two-dimensional coupled feature map independently along the direction of the primary decomposition axis.

[0173] In practice, spatial frequency decomposition techniques such as one-dimensional discrete wavelet transform or one-dimensional filter bank can be selected to perform low-pass and high-pass filtering operations on each row of data to obtain the low-frequency component map and the high-frequency component map of the main axis.

[0174] The low-frequency component map of the principal axis contains the spatially gradual trend of the two-dimensional coupling feature map along the principal decomposition axis, which characterizes the overall spatial smoothness of the array mutual coupling effect; while the high-frequency component map of the principal axis reflects the spatial local fluctuation or disturbance characteristics along the principal axis.

[0175] Subsequently, in the second stage, the principal axis low-frequency component map is further decomposed along the secondary decomposition axis using a one-dimensional multi-scale decomposition process. This stage specifically includes:

[0176] One-dimensional low-pass filtering is performed on each column of data in the principal axis low-frequency component plot along the secondary decomposition axis to extract a low-frequency feature subset of the overall spatial characteristics of the array. This low-frequency feature subset reflects the smoothest and most gradually changing trend information of the mutual coupling effect, reflecting the global spatial continuity and trend characteristics of the array, and can be used for subsequent global spatial feature analysis and processing for mutual coupling effect compensation.

[0177] To extract the local high-frequency disturbance information in the two-dimensional coupled feature map more completely, a one-dimensional high-pass filter needs to be performed on the principal axis low-frequency component map along the secondary decomposition axis to generate a first high-frequency component, which records the local disturbance features in the direction of the secondary decomposition axis.

[0178] Simultaneously, to further and comprehensively extract local perturbation features at different spatial scales, it is necessary to perform one-dimensional low-pass filtering and one-dimensional high-pass filtering along the secondary decomposition axis on the principal axis high-frequency component map obtained in the first stage, thereby obtaining the second and third high-frequency components that respectively reflect the features at different spatial frequency scales. These two high-frequency components represent local spatial perturbation features at different scales in the mutual coupling effect. For example, the second high-frequency component can reflect the mesoscale perturbation features in the secondary axis direction, while the third high-frequency component represents more refined microscale perturbation information.

[0179] Through the above two-stage decomposition, a low-frequency feature subset and a high-frequency feature subset composed of the first, second, and third high-frequency components are finally obtained. Among them, the low-frequency feature subset reflects the overall spatial trend of the array mutual coupling effect, while the high-frequency feature subset fully records the local perturbation characteristics caused by the mutual coupling between array elements at different spatial scales.

[0180] For example, taking a rectangular antenna array as an example, if its beam scanning direction is horizontal in the array plane, then the main decomposition axis is horizontal and the secondary decomposition axis is vertical. In the first stage, each row of the two-dimensional coupling feature map, i.e., the horizontal row data, is processed using a one-dimensional discrete wavelet transform to generate the main axis low-frequency component map and the main axis high-frequency component map. In the second stage, each column of the main axis low-frequency component map is processed by low-pass filtering and high-pass filtering along the vertical direction to generate a low-frequency feature subset and a first high-frequency component. Then, each column of the main axis high-frequency component map is processed by low-pass filtering and high-pass filtering to generate a second high-frequency component and a third high-frequency component.

[0181] In this way, the two-stage directional decomposition can improve the accuracy and completeness of array mutual coupling features in the spatial scale separation process, ensuring that mutual coupling features at different scales are fully and finely expressed, avoiding the loss of spatial features caused by single-scale analysis. By performing multi-scale decomposition on the principal axis and secondary axis respectively, the spatial gradual trend and local high-frequency perturbation features in the array mutual coupling effect can be highlighted in a targeted manner, and the spatial structure of the actual mutual coupling effect can be more accurately characterized. The obtained low-frequency feature subset and high-frequency feature subset directly correspond to the different spatial scale features of the array mutual coupling effect, which can provide more accurate and differentiated input data for the LNN used in the subsequent step S104, thereby further improving the reliability and accuracy of subsequent dynamic compensation and test evaluation of mutual coupling effect.

[0182] As an optional implementation, the low-frequency feature subset and the high-frequency feature subset are processed by the first processing unit and the second processing unit respectively, so as to further improve the accuracy and effectiveness of the dynamic compensation processing of the mutual coupling effect of the target antenna array.

[0183] In specific implementation, unlike the first processing unit mentioned above, the second processing unit specially designed in this embodiment adopts multiple parallel dedicated processing channels to perform independent processing on different scale spatial features contained in the high-frequency feature subset, namely the first high-frequency component, the second high-frequency component and the third high-frequency component, so as to significantly improve the extraction efficiency and accuracy of local spatial disturbance features.

[0184] Specifically, each dedicated processing channel in the second processing unit can be designed as an independent nonlinear feature extraction network during implementation. For example, each channel can adopt a convolutional neural network (CNN) structure. Each dedicated CNN processing channel includes several convolutional layers and nonlinear activation layers. The size and number of convolutional kernels are set according to specific scale requirements. For example, for the first high-frequency component, the convolutional kernel can be larger to capture larger-scale perturbation information; for the third high-frequency component, a smaller-sized convolutional kernel can be selected to more effectively capture small-scale spatial perturbation details.

[0185] Through the implementation design of the aforementioned parallel dedicated processing channels, each high-frequency component feature is independently and specifically subjected to nonlinear feature extraction processing, generating multiple high-frequency dedicated processing results. This specialized design can significantly improve the efficiency and accuracy of capturing local perturbation information of mutual coupling effects, effectively avoiding the limitation of a single nonlinear network in processing high-frequency features at different scales.

[0186] To further improve the fusion effect between high-frequency dedicated processing results, this implementation adopts a context-guided attention fusion method to dynamically fuse multiple high-frequency dedicated processing results into the final high-frequency processing result.

[0187] In practice, the low-frequency processing result obtained by the first processing unit is used as the global context guidance information. The main reason for using the low-frequency processing result as the context guidance information is that the low-frequency processing result reflects the overall spatial trend of the array mutual coupling effect, which can provide a global reference for the fusion process of high-frequency features, thereby effectively improving the spatial consistency and accuracy of the fusion result.

[0188] Next, dynamic fusion weights are calculated using a pre-configured attention network. This attention network can include several convolutional and fully connected layers. For example, one to three convolutional or fully connected layers can be used to extract features from the low-frequency processing results. The extracted features are then input into the attention mechanism to generate a set of fusion weights for the high-frequency processing results. Essentially, the attention mechanism dynamically adjusts the importance of each high-frequency processing result in the final fusion based on the global contextual guidance information provided by the low-frequency processing results, achieving efficient coordination between local perturbation information and global spatial trends.

[0189] In the specific fusion operation, the fusion weights calculated above are used to perform a pixel-by-pixel weighted summation on multiple high-frequency dedicated processing results to obtain the final fused high-frequency processing result. Through this context-guided attention fusion method, local perturbation information at various spatial scales in the high-frequency processing results can be more effectively integrated and highlighted, especially achieving a precise and efficient balance between local detail features and global spatial trends, significantly improving the spatial accuracy and robustness of dynamic compensation for mutual coupling effects.

[0190] In this way, the second processing unit not only effectively achieves independent and refined extraction of multi-scale high-frequency perturbation features, but also achieves precise coordination between global trends and local perturbation information through a context-guided attention fusion method. This effectively improves the accuracy and robustness of dynamic compensation processing for array mutual coupling effects, and ultimately significantly improves the radiation characteristics and compensation processing effect of the array.

[0191] As an optional implementation, using low-frequency processing results as global context guidance information includes:

[0192] Based on the low-frequency processing results, a global intensity scalar is calculated to characterize the amplitude characteristics of the global slowly varying trend.

[0193] The low-frequency processing result is input into a pre-configured context feature extraction network to generate an attention-guided vector that characterizes the spatial structure features of the global slowly changing trend.

[0194] The attention guidance vector is used as the query input to the attention network to calculate the initial fusion weights for each high-frequency dedicated processing result;

[0195] Based on the initial fusion weights, the multiple high-frequency dedicated processing results are weighted and summed to obtain preliminary high-frequency fusion results;

[0196] An adaptive gain scaling operation is performed on the preliminary high-frequency fusion result using the global intensity scalar, wherein the gain scaling operation is to multiply the preliminary high-frequency fusion result by a scaling factor determined by the global intensity scalar to adjust the overall amplitude of the high-frequency processing result.

[0197] In practical implementation, the technical solution of using low-frequency processing results as global context guidance information aims to further improve the effectiveness and stability of high-frequency processing result fusion, so as to more accurately reflect the actual characteristics of the mutual coupling effect of multi-channel electrically tunable base station antennas.

[0198] First, based on the low-frequency processing results obtained from the first processing unit, a global intensity scalar is calculated to characterize the global gradually varying amplitude characteristics. Specifically, this global intensity scalar is a scalar value obtained by global pooling or spatial averaging based on the low-frequency processing results. This scalar value essentially reflects the average amplitude intensity characteristics of the mutual coupling effect of the entire array at the current beam scanning angle, and can provide an effective global amplitude reference for subsequent fusion operations.

[0199] Simultaneously, to further reflect the spatial structural characteristics of the low-frequency processing results, these results need to be input into a pre-configured context feature extraction network to generate an attention-guided vector. In practice, the context feature extraction network can be designed as a deep feature extraction network consisting of several convolutional or fully connected layers. For example, it can use two to three convolutional layers to progressively extract spatial structural features, and then convert these features into a compact feature representation, i.e., the attention-guided vector, through global pooling or dimensionality reduction. This attention-guided vector actually represents the spatial structural information of a globally gradually changing trend.

[0200] Subsequently, in a specific embodiment of this application, the attention-guided vector obtained above is input as a "query" into a pre-configured attention network to calculate the initial fusion weights for each high-frequency specialized processing result. Specifically, this attention network can be implemented using a classic attention mechanism, such as dynamically generating the initial fusion weights for each high-frequency specialized processing result by using a similarity measure between the feature vectors corresponding to each high-frequency specialized processing result, such as cosine similarity or dot product similarity.

[0201] Furthermore, a pixel-by-pixel weighted summation operation is performed on multiple high-frequency dedicated processing results based on the initial fusion weights to obtain a preliminary high-frequency fusion result. Although this preliminary high-frequency fusion result already reflects the effective coordination relationship between high-frequency local perturbation information and low-frequency global context, due to the significant variations in amplitude characteristics under different array conditions and scanning angles, it is still necessary to perform an adaptive gain scaling operation on the preliminary high-frequency fusion result using the aforementioned global intensity scalar.

[0202] In practice, the gain scaling operation achieves adaptive adjustment of the overall amplitude of the initial high-frequency fusion result by multiplying the initial high-frequency fusion result pixel by pixel with a scaling factor dynamically calculated based on the global intensity scalar.

[0203] Specifically, the scaling factor can be obtained by mapping the global intensity scalar to a suitable gain range through nonlinear mapping. In this way, when the global intensity scalar is high, it indicates that the global mutual coupling effect has a significant gradual trend, and the gain factor is high, so that the amplitude of the initial high-frequency fusion result is appropriately enhanced; when the global intensity scalar is low, the gain factor is reduced to avoid excessive amplification of the amplitude of the initial high-frequency fusion result, thereby achieving precise control and adaptive adjustment of the overall amplitude.

[0204] As an optional implementation, the method provided in this application for performing feature mapping and weighted superposition of the processing results of the first processing unit and the second processing unit to obtain the amplitude and phase calibration coefficients is based on the introduction of the concept of unit factor effectiveness mapping map to achieve refined processing of dynamic compensation for the mutual coupling effect of the target antenna array.

[0205] In practice, firstly, based on the beam scanning angle corresponding to the current measurement and combined with the preset physical radiation model of the target antenna array, a unit factor effectiveness mapping map corresponding one-to-one with the position of each array element is calculated and generated. This unit factor effectiveness mapping map is a spatial feature map used to quantify and characterize the actual radiation performance of each element at the current beam scanning angle. The specific meaning of its pixel value is the radiation efficiency or radiation gain effectiveness of each element in a specific scanning direction; the higher the value, the greater the contribution of that element to beamforming at the current scanning angle. By constructing such a unit factor effectiveness mapping map, the spatial modulation effect of the actual physical radiation performance of the array elements on the mutual coupling effect can be effectively reflected, providing a clear physical reference for the subsequent compensation process.

[0206] Then, using the unit factor validity map obtained above, parameterized nonlinear feature mapping processing is performed on the high-frequency processing result output by the second processing unit to obtain the perturbation compensation field. Specifically, this nonlinear feature mapping processing can be implemented using a parameterized nonlinear activation network, such as a parameterized nonlinear function network, an adaptive activation network, or a parameterized rectified linear unit. This processing method effectively utilizes the spatial feature distribution of the unit factor validity map to adaptively modulate the amplitude and spatial structure of the high-frequency perturbation information, making the high-frequency perturbation compensation field more accurately reflect the nonlinear spatial perturbation mode of the actual array element radiation characteristics in the local region.

[0207] Next, the low-frequency processing result output by the first processing unit is used as the basic compensation field, and the disturbance compensation field obtained above is subjected to spatial adaptive weighted superposition processing with the basic compensation field.

[0208] A spatially adaptive fusion strategy can be adopted to ensure the efficient fusion of the base compensation field and the perturbation compensation field at different array locations, thereby obtaining the final amplitude and phase calibration coefficients. In practice, a spatial weight matrix needs to be calculated and generated. The values ​​of the elements in this spatial weight matrix are determined by two factors: first, the local amplitude distribution of the low-frequency processing results; and second, the element radiation effectiveness distribution in the element factor effectiveness mapping diagram.

[0209] For example, regarding the local amplitude distribution of low-frequency processing results, the local spatial gradient or local amplitude mean can be calculated by analyzing the amplitude variation characteristics of the local region, reflecting the importance or stability of the global gradual variation trend within the local region. On the other hand, the element radiation effectiveness distribution in the unit factor effectiveness mapping map is used to reflect the physical radiation contribution of each element in the current scanning direction. Specifically, feature normalization processing can be performed on these two aspects of data, followed by pixel-by-pixel or local-region nonlinear fusion operations to generate the final spatial weight matrix.

[0210] After obtaining the spatial weight matrix, a spatially adaptive weighted fusion process is further performed. In practice, the spatial weight matrix, the base compensation field, and the perturbation compensation field can be used to perform pixel-by-pixel or local region weighting operations, thereby accurately superimposing global trend and local perturbation information in an appropriate proportion to obtain the final amplitude and phase calibration coefficients.

[0211] For example, taking a rectangular target antenna array, when the beam scanning angle is 15°, firstly, based on a preset array physical radiation model, a unit factor effectiveness mapping map is calculated, in which the radiation effectiveness of the array elements in the central region is higher, and the edge region is relatively lower. Based on this unit factor effectiveness mapping map, parameterized nonlinear feature mapping is performed on the high-frequency processing results output by the second processing unit, successfully generating a disturbance compensation field that effectively reflects the spatial disturbance characteristics. Subsequently, based on the low-frequency processing results output by the first processing unit as the compensation field, a spatial weight matrix is ​​further calculated. This matrix comprehensively considers the local amplitude information of the low-frequency processing results and the radiation effectiveness information of the array elements, clearly showing that the central region has a higher weight and the edge region has a lower weight. Finally, through weighted fusion using the spatial weight matrix, amplitude and phase calibration coefficients that accurately reflect the amplitude and phase characteristics of the mutual coupling effect at this scanning angle are successfully obtained.

[0212] In this way, by introducing the unit factor effectiveness mapping diagram, the influence of the actual radiation characteristics of the array on the local space can be more accurately reflected, and the high-frequency disturbance information can be more accurately adjusted and incorporated into the final compensation process. By combining parameterized nonlinear feature mapping processing with spatial adaptive weighted fusion method, the optimization fusion between global trend and local disturbance information can be dynamically and finely realized, improving the reliability and accuracy of amplitude and phase calibration coefficients, and further improving the overall performance and effect of actual array mutual coupling effect compensation.

[0213] Based on the same inventive concept, this application also provides an array antenna mutual coupling calibration system based on spatial multi-scale feature decoupling, which corresponds to the array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling. Since the principle of the system in this application is similar to the array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling described above, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0214] Reference Figure 4 The diagram shown is a schematic of the array antenna mutual coupling calibration system based on spatial multi-scale feature decoupling provided in this application. The system includes:

[0215] The acquisition module 10 is used to measure the scattering parameters of each element port of the target antenna array at multiple scanning angles, and obtain amplitude and phase test data characterizing the mutual coupling effect of the array.

[0216] The construction module 20 is used to construct a two-dimensional coupling feature map based on the amplitude and phase test data, wherein the pixel position of the two-dimensional coupling feature map corresponds one-to-one with the spatial position of each element of the target antenna array, and the pixel value corresponds to the scattering parameter measured at the corresponding element position.

[0217] The decomposition module 30 is used to perform two-dimensional spatial multi-scale decomposition processing on the two-dimensional coupling feature map to separate the low-frequency feature subset that characterizes the global slow-changing trend of the mutual coupling effect and the high-frequency feature subset that characterizes the local spatial perturbation of the mutual coupling effect.

[0218] The processing module 40 is used to process the low-frequency feature subset and the high-frequency feature subset respectively based on a parallel processing model including a first processing unit and a second processing unit. The first processing unit is used to calculate the low-frequency processing result characterizing the global gradual change trend, and the second processing unit is used to calculate the high-frequency processing result characterizing the local spatial perturbation. The processing results of the first processing unit and the second processing unit are subjected to feature mapping and weighted superposition to calculate the amplitude and phase calibration coefficient characterizing the amplitude and phase deviation of the current array mutual coupling effect, and the amplitude and phase calibration coefficient is output.

[0219] Evaluation module 50 is used to calculate, based on the amplitude and phase calibration coefficients, using a preset neural network quantization model, and generate and output test evaluation results characterizing the effectiveness of the amplitude and phase calibration coefficients in dynamically compensating for array mutual coupling effects.

[0220] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

Claims

1. A method for mutual coupling calibration of array antennas based on spatial multi-scale feature decoupling, characterized in that, include: The target antenna array is placed in a fully anechoic chamber, and the first and second ports of its multiple channel ports are connected to a vector network analyzer. Terminate all remaining unconnected channel ports to a matching load; The vector network analyzer measures and obtains the transmission coefficient between the first port and the second port within a preset operating frequency band to characterize the coupling strength between the two ports. By iterating through multiple port pairs, repeatedly performing the steps of connecting the port pairs to the vector network analyzer, terminating the remaining unconnected channel ports to matched loads, and measuring the transmission coefficient, scattering parameters characterizing the antenna mutual coupling effect are obtained. Also includes: The scattering parameters of each element port of the target antenna array are measured at multiple scanning angles to obtain amplitude and phase test data characterizing the array mutual coupling effect. A two-dimensional coupling feature map is constructed based on the amplitude and phase test data, wherein the pixel position of the two-dimensional coupling feature map corresponds one-to-one with the spatial position of each element of the target antenna array, and the pixel value corresponds to the scattering parameter measured at the corresponding element position. Two-dimensional spatial multi-scale decomposition processing is performed on the two-dimensional coupling feature map to separate the low-frequency feature subset that characterizes the global slow-changing trend of the mutual coupling effect and the high-frequency feature subset that characterizes the local spatial perturbation of the mutual coupling effect. Based on a parallel processing model including a first processing unit and a second processing unit, the low-frequency feature subset and the high-frequency feature subset are processed respectively. The first processing unit is used to calculate the low-frequency processing result characterizing the global gradual change trend, and the second processing unit is used to calculate the high-frequency processing result characterizing the local spatial perturbation. The processing results of the first processing unit and the second processing unit are subjected to feature mapping and weighted superposition to calculate the amplitude and phase calibration coefficient characterizing the amplitude and phase deviation of the current array mutual coupling effect, and the amplitude and phase calibration coefficient is output. Based on the amplitude and phase calibration coefficients, a preset neural network quantization model is used to calculate and output test evaluation results that characterize the effectiveness of the amplitude and phase calibration coefficients in dynamically compensating for array mutual coupling effects.

2. The array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling according to claim 1, characterized in that, include: The measured scattering parameters are compared with preset technical specifications to determine whether the antenna's isolation performance is up to standard.

3. The array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling according to claim 1, characterized in that, The construction of the two-dimensional coupled feature map based on the amplitude-phase test data includes: Based on the Cartesian coordinates of each element of the target antenna array in physical space, the amplitude and phase test data are mapped to the Cartesian coordinate system to obtain the initial physical domain coupling feature map. Based on the beam scanning angle corresponding to the current measurement, an analysis domain coordinate system is constructed, wherein the first coordinate axis of the analysis domain coordinate system is parallel to the projection direction of the beam scanning direction in the target antenna array plane, and the second coordinate axis is orthogonal to the first axis; An affine transformation operation is performed on the initial physical domain coupling feature map to map the amplitude and phase test data in the initial physical domain coupling feature map from the Cartesian coordinate system to the analysis domain coordinate system, thereby obtaining coupling feature data that is non-uniformly distributed in the analysis domain coordinate system. The non-uniformly distributed coupling feature data is interpolated and resampled to obtain a two-dimensional coupling feature map with a uniform grid distribution in the coordinate system of the analysis domain.

4. The array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling according to claim 3, characterized in that, The process of obtaining a two-dimensional coupled feature map that exhibits a uniform grid distribution in the analysis domain coordinate system includes: Based on the spatial location of the array elements in the target antenna array, the non-uniformly distributed coupling feature data is divided into regions to generate at least one central region subset corresponding to the array central region element and at least one edge region subset corresponding to the array edge region element.

5. The array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling according to claim 4, characterized in that, The process of obtaining a two-dimensional coupled feature map that is uniformly grid-distributed in the coordinate system of the analysis domain further includes: The configuration includes a multi-channel interpolation kernel function set including a first interpolation channel and a second interpolation channel, wherein the first interpolation channel adopts a first interpolation kernel function for maintaining the smooth continuity of the amplitude and phase signals, and the second interpolation channel adopts a second interpolation kernel function for maintaining the high-frequency detail features of the amplitude and phase signals; Interpolation processing is performed using only the central region subset and the first interpolation channel to obtain the basic coupling field distribution map; Interpolation processing is performed using only the subset of the edge regions and the second interpolation channel to obtain a detailed perturbation field distribution map; The basic coupled field distribution map and the detailed perturbation field distribution map are spatially adaptively weighted and fused. For each target pixel in the uniform grid, the fusion weight is determined according to the spatial distance between the target pixel and the data points in the edge region subset, and the two-dimensional coupled feature map is obtained.

6. The array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling according to claim 1, characterized in that, The two-dimensional spatial multi-scale decomposition process performed on the two-dimensional coupling feature map separates a low-frequency feature subset characterizing the global slowly varying trend of the mutual coupling effect and a high-frequency feature subset characterizing the local spatial perturbation of the mutual coupling effect, including: Based on the beam scanning angle corresponding to the two-dimensional coupling feature map, the primary and secondary decomposition axes of the two-dimensional spatial multi-scale decomposition are determined. The primary decomposition axis is parallel to the projection direction of the beam scanning direction in the target antenna array plane, and the secondary decomposition axis is orthogonal to the primary decomposition axis.

7. The array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling according to claim 6, characterized in that, The two-dimensional spatial multi-scale decomposition process includes: The first-stage one-dimensional multi-scale decomposition process is performed independently on each row of data in the two-dimensional coupled feature map along the main decomposition axis to obtain the main axis low-frequency component map and the main axis high-frequency component map. A second-stage one-dimensional multi-scale decomposition process is performed on the principal axis low-frequency component map and the principal axis high-frequency component map along the secondary decomposition axis, wherein: Each column of data in the principal axis low-frequency component plot is subjected to one-dimensional low-pass filtering along the secondary decomposition axis to generate the low-frequency feature subset. The low-frequency component map of the main axis is subjected to a one-dimensional high-pass filter along the secondary decomposition axis to obtain the first high-frequency component. The high-frequency component map of the main axis is subjected to one-dimensional low-pass filtering and one-dimensional high-pass filtering along the secondary decomposition axis to obtain the second high-frequency component and the third high-frequency component, respectively. The first high-frequency component, the second high-frequency component, and the third high-frequency component are collectively used as the high-frequency feature subset, wherein the high-frequency feature subset characterizes the different spatial scale features of the local spatial perturbation of the mutual coupling effect.

8. The array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling according to claim 7, characterized in that, The parallel processing model, which includes a first processing unit and a second processing unit, processes the low-frequency feature subset and the high-frequency feature subset respectively, including: The first processing unit processes the low-frequency feature subset to calculate the low-frequency processing result; The high-frequency feature subset is processed by the second processing unit, wherein the second processing unit includes multiple parallel dedicated processing channels, each of which corresponds to a first high-frequency component, a second high-frequency component, and a third high-frequency component in the high-frequency feature subset, and performs independent processing on the first high-frequency component, the second high-frequency component, and the third high-frequency component to obtain multiple high-frequency dedicated processing results. A context-guided attention fusion method is used to fuse the multiple high-frequency specialized processing results, wherein: The low-frequency processing results are used as global context guidance information; The fusion weights of the multiple high-frequency dedicated processing results are dynamically calculated based on the global context guidance information through a pre-configured attention network. The multiple high-frequency dedicated processing results are weighted and summed based on the fusion weights to obtain the final high-frequency processing result.

9. An array antenna mutual coupling calibration system based on spatial multi-scale feature decoupling, used to implement the array antenna mutual coupling calibration method based on spatial multi-scale feature decoupling as described in any one of claims 1-8, characterized in that, include: The acquisition module is used to measure the scattering parameters of each element port of the target antenna array at multiple scanning angles, and obtain amplitude and phase test data characterizing the mutual coupling effect of the array. The construction module is used to construct a two-dimensional coupling feature map based on the amplitude and phase test data, wherein the pixel position of the two-dimensional coupling feature map corresponds one-to-one with the spatial position of each element of the target antenna array, and the pixel value corresponds to the scattering parameter measured at the corresponding element position. The decomposition module is used to perform two-dimensional spatial multi-scale decomposition processing on the two-dimensional coupling feature map to separate the low-frequency feature subset that characterizes the global slow-changing trend of the mutual coupling effect and the high-frequency feature subset that characterizes the local spatial perturbation of the mutual coupling effect. The processing module is used to process the low-frequency feature subset and the high-frequency feature subset respectively based on a parallel processing model including a first processing unit and a second processing unit. The first processing unit is used to calculate the low-frequency processing result characterizing the global gradual change trend, and the second processing unit is used to calculate the high-frequency processing result characterizing the local spatial perturbation. The processing results of the first processing unit and the second processing unit are subjected to feature mapping and weighted superposition to calculate the amplitude and phase calibration coefficient characterizing the amplitude and phase deviation of the current array mutual coupling effect, and the amplitude and phase calibration coefficient is output. The evaluation module is used to calculate, based on the amplitude and phase calibration coefficients, using a preset neural network quantization model, and generate and output test evaluation results characterizing the effectiveness of the amplitude and phase calibration coefficients in dynamically compensating for array mutual coupling effects.

Images