Method and system for automated amplitude and phase consistency testing of multi-channel electrically adjustable antenna

Abstract

The application provides a method and system for automatically testing amplitude and phase consistency of a multi-channel electrically adjustable antenna, and relates to the technical field of antenna testing. The method comprises the following steps: connecting each channel port of a to-be-tested multi-channel electrically adjustable antenna to a testing device; sequentially performing phase testing and S parameter testing on each channel of the antenna by using the testing device, and acquiring actual phase values and actual amplitude values of each channel; comparing the acquired actual amplitude values and actual phase values with preset standard values to determine amplitude and phase errors of each channel; determining whether the amplitude and phase consistency of the antenna is qualified according to whether the amplitude and phase errors are within a preset tolerance range, and performing repair processing on the antenna when it is determined that the amplitude and phase consistency is unqualified. In this way, the production efficiency, quality control and reliability of the multi-channel electrically adjustable antenna are improved, and the stringent requirements of a high-frequency communication system are ensured.

Description

Technical Field

[0001] This application relates to the field of antenna testing technology, and more specifically, to an automated testing method and system for amplitude and phase consistency of multi-channel electrically tunable antennas. Background Technology

[0002] As a core component of base stations, the amplitude and phase consistency of multi-channel electrically tunable antennas directly affects beamforming accuracy, signal coverage, and interference suppression capabilities. In actual production and deployment, multi-channel electrically tunable antennas must ensure high amplitude and phase consistency across all channels to achieve efficient MIMO transmission and adaptive beam control. However, existing amplitude and phase consistency testing methods mainly rely on manual measurement or simple vector network analyzer testing. These methods have limitations: first, low testing efficiency; for multi-channel antennas, such as 8x8 or higher array antennas, manual port switching for each channel is required, which is time-consuming and prone to human error; second, insufficient accuracy, failing to fully consider the electromagnetic coupling effects between channels, leading to biased error assessment; and third, lack of automated judgment mechanisms, relying solely on threshold comparisons, which cannot quantify consistency risks under complex disturbances. Furthermore, traditional methods often involve testing at the finished product stage; once a defect is found, the antenna cover must be removed for repair, increasing costs and time.

[0003] Therefore, there is an urgent need for an automated, high-precision amplitude and phase consistency testing method that takes into account electromagnetic physics characteristics, in order to improve the production efficiency, quality control and reliability of multi-channel electrically tunable antennas, and ensure that they meet the stringent requirements of high-frequency communication systems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application provides an automated testing method and system for amplitude and phase consistency of multi-channel electrically tunable antennas.

[0005] In a first aspect, this application provides an automated testing method for the amplitude and phase consistency of a multi-channel electrically tunable antenna, including:

[0006] Connect each channel port of the multi-channel electrically adjustable antenna under test to the test equipment;

[0007] Using the testing equipment, phase and S-parameter tests are performed sequentially on each channel of the antenna to obtain the actual phase value and actual amplitude value of each channel.

[0008] The actual amplitude and phase values ​​obtained are compared with preset standard values ​​to determine the amplitude and phase errors of each channel;

[0009] Based on whether the amplitude and phase error is within the preset tolerance range, it is determined whether the amplitude and phase consistency of the antenna is qualified, and if it is determined to be unqualified, the antenna is reworked.

[0010] Optionally, the multi-channel electrically tunable antenna under test is a semi-finished multi-channel electrically tunable antenna, the semi-finished product comprising:

[0011] Metal reflector;

[0012] Multiple radiating elements, a power supply network, and a phase shifter assembly are mounted on the metal reflector.

[0013] The phase test and S-parameter test are performed before the radome is installed on the semi-finished product.

[0014] Optionally, the method also includes:

[0015] Obtain the physical location information of each channel of the multi-channel electrically tunable antenna, and construct an antenna array diagram structure based on the physical location information. Each node in the antenna array diagram structure corresponds to one channel of the multi-channel electrically tunable antenna.

[0016] The actual amplitude and phase values ​​of each channel of the multi-channel electrically adjustable antenna are collected. The amplitude error value and phase error value of each channel are calculated based on the preset amplitude standard value and phase standard value. The amplitude error value and phase error value are mapped to the antenna array diagram structure to form a real amplitude and phase error diagram signal.

[0017] Based on the true amplitude and phase error map signal, a latent space posterior probability distribution characterizing the true amplitude and phase error map signal is generated using a preset graph encoder; at least one latent variable is randomly sampled from the latent space posterior probability distribution, and the latent variable is decoded using a preset graph generator to generate amplitude and phase perturbation samples; the true amplitude and phase error map signal and the amplitude and phase perturbation samples are input to a preset graph discriminator to calculate the authenticity discrimination value of each input signal respectively;

[0018] Based on the true / false discrimination value and the amplitude / phase perturbation sample, the evaluation index of amplitude / phase consistency of the multi-channel electrically adjustable antenna is calculated, and the multi-channel amplitude / phase consistency test result of the multi-channel electrically adjustable antenna is determined according to the evaluation index.

[0019] Optionally, constructing the antenna array diagram structure based on the physical location information includes:

[0020] Obtain the physical coordinate information of each channel of the multi-channel electrically adjustable antenna;

[0021] The Euclidean distance between any two channels is calculated based on the physical coordinate information, and the Euclidean distance is converted into the corresponding electrical length according to the preset test signal center frequency.

[0022] A first distance threshold and a second distance threshold are preset to distinguish between strongly coupled near-field effects and weakly coupled far-field effects, wherein the first distance threshold is smaller than the second distance threshold;

[0023] Based on the Euclidean distance, the connection relationship between any two channels is divided into a first connection relationship or a second connection relationship that are mutually exclusive. The first connection relationship corresponds to the case where the Euclidean distance is less than or equal to the first distance threshold, and the second connection relationship corresponds to the case where the Euclidean distance is greater than the first distance threshold and less than or equal to the second distance threshold.

[0024] Optionally, the step of constructing the antenna array diagram structure based on the physical location information further includes:

[0025] For each pair of channels with the first connection relationship or the second connection relationship, the normalized connection weight value of the corresponding connection relationship is calculated based on the electrical length and the preset path loss model.

[0026] Based on the channels, the first connection relationship, the second connection relationship, and the connection weight values, a weighted heterogeneous graph structure is constructed to characterize the multi-scale electromagnetic coupling characteristics between channels of a multi-channel electrically tunable antenna.

[0027] Optionally, the process of forming the true amplitude and phase error map signal includes:

[0028] Based on the preset maximum amplitude error range, the amplitude error value of each channel is normalized to obtain the normalized amplitude error value corresponding to that channel.

[0029] The phase error value of each channel is transformed by a trigonometric function and mapped to a two-dimensional Cartesian coordinate space to obtain the two-dimensional phase error vector corresponding to that channel.

[0030] The normalized amplitude error value and the two-dimensional phase error vector corresponding to each channel are concatenated to generate the node initial feature vector corresponding to each channel.

[0031] The initial feature vector of each channel is used as a node attribute and assigned to the corresponding node in the weighted heterogeneous graph structure to generate the true amplitude and phase error map signal.

[0032] Optionally, generating the latent space posterior probability distribution for characterizing the true amplitude and phase error map signal includes:

[0033] Using the multi-layer graph convolutional network built into the graph encoder, the node initial feature vectors of the real amplitude and phase error graph signal are aggregated layer by layer to obtain the node output feature vector of each node.

[0034] Perform a residual join operation between the node output feature vector of each node and the node initial feature vector corresponding to that node to obtain a residual feature vector including the initial error features;

[0035] Perform a global pooling operation on the residual feature vectors corresponding to all nodes to obtain a graph-level context vector for the entire antenna array graph structure;

[0036] The residual feature vector corresponding to each node is concatenated with the graph-level context vector to generate a fused feature vector that integrates local node information and global graph information.

[0037] The graph encoder uses a fully connected layer network to map the fused feature vector of each node to the latent space, generating the mean and variance parameters corresponding to the posterior probability distribution of the latent space.

[0038] Optionally, obtaining the node output feature vector for each node includes:

[0039] For each layer of the multilayer graph convolutional network, the following feature update operation is performed on each node of the antenna array graph structure:

[0040] For neighboring nodes with the first connection relationship and neighboring nodes with the second connection relationship, independent weighted information aggregation operations are performed based on the normalized connection weight values ​​corresponding to each neighboring node to obtain intermediate feature vectors that respectively characterize the near-field effect and the far-field effect.

[0041] Linear transformations are performed on the intermediate feature vectors representing the near-field and far-field effects, respectively, and the linearly transformed feature vectors are fused to generate an aggregated feature vector.

[0042] Based on the aggregated feature vector and the previous layer feature vector of the node itself, the feature vector of the node is updated to obtain the updated node feature vector.

[0043] A non-linear activation process is performed on the updated node feature vector to obtain the node output feature vector of the node in the current layer.

[0044] Optionally, generating amplitude and phase perturbation samples includes:

[0045] The latent variables obtained from the latent space posterior probability distribution are decomposed into global latent variables and local latent variables.

[0046] The global latent variables are processed using a first neural network to generate a global bias vector;

[0047] The local latent variables are assigned to each node in the weighted heterogeneous graph structure as initial random features of the nodes.

[0048] Secondly, this application provides an automated testing system for the amplitude and phase consistency of a multi-channel electrically tunable antenna, comprising:

[0049] A construction module is used to obtain the physical location information of each channel of the multi-channel electrically tunable antenna, and construct an antenna array diagram structure based on the physical location information. Each node in the antenna array diagram structure corresponds to one channel of the multi-channel electrically tunable antenna.

[0050] The mapping module is used to collect the actual amplitude and phase values ​​of each channel of the multi-channel electrically tunable antenna, calculate the amplitude error value and phase error value of each channel based on the preset amplitude standard value and phase standard value, and map the amplitude error value and phase error value to the antenna array diagram structure to form a real amplitude and phase error diagram signal.

[0051] The processing module, based on the true amplitude-phase error map signal, uses a preset graph encoder to generate a latent space posterior probability distribution characterizing the true amplitude-phase error map signal; randomly samples at least one latent variable from the latent space posterior probability distribution, decodes the latent variable using a preset graph generator to generate amplitude-phase perturbation samples; and inputs the true amplitude-phase error map signal and the amplitude-phase perturbation samples into a preset graph discriminator to calculate the authenticity discrimination value for each input signal.

[0052] The testing module calculates the evaluation index of amplitude and phase consistency of the multi-channel electrically adjustable antenna based on the true / false discrimination value and the amplitude and phase disturbance sample, and determines the multi-channel amplitude and phase consistency test result of the multi-channel electrically adjustable antenna according to the evaluation index.

[0053] Compared to existing technologies, the testing method in this application differs from traditional methods that directly analyze single, limited measured data. This invention utilizes a graphical model based on the antenna's physical topology to learn the inherent error statistical characteristics exhibited by a single actual measurement. Based on these learned statistical characteristics, the method can generate multiple sets of simulated amplitude and phase perturbation samples, and the final evaluation index is derived from the overall statistical analysis of these simulated samples. Therefore, this method reduces the interference of random errors in a single measurement on the final conclusion, improving the stability of the test results. Simultaneously, since the simulated samples can cover potential error situations that are unlikely to occur in actual measurements, the evaluation results of this method are more comprehensive. Furthermore, by introducing the graphical structure of the antenna array, the model can consider the spatial correlation between channels in the analysis, making its error modeling closer to physical reality, thereby improving the accuracy of the test. Attached Figure Description

[0054] Figure 1 A flowchart of the automated testing method provided in this application;

[0055] Figure 2 A flowchart of a method for constructing an antenna array diagram structure based on the physical location information provided in this application;

[0056] Figure 3 A flowchart of a method for generating a true amplitude and phase error map signal is provided in this application;

[0057] Figure 4 A schematic diagram of the automated test system for amplitude and phase consistency of the multi-channel electrically adjustable antenna provided in this application;

[0058] Figure 5 A flowchart of the automated test method for amplitude and phase consistency of the multi-channel electrically tunable antenna provided in this application. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0060] This invention provides an automated testing method for amplitude and phase consistency of multi-channel electrically tunable antennas, aiming to achieve efficient evaluation of the amplitude and phase consistency of multi-channel electrically tunable antennas through automated testing equipment. This method is applicable to various multi-channel electrically tunable antennas, such as array antennas used in 5G / 6G base stations, and is particularly suitable for electrically tunable antennas with a large number of channels, such as 8 or more channels. The core of this method lies in achieving consistency judgment through port connection, parameter testing, and error comparison, and supports testing at the semi-finished product stage, thereby identifying problems early and reducing rework costs.

[0061] See Figure 5 The flowchart of the automated test method for amplitude and phase consistency of the multi-channel electrically tunable antenna provided in this application includes steps S401 to S404, wherein:

[0062] S401: Connect each channel port of the multi-channel electrically adjustable antenna under test to the test equipment;

[0063] In this step, each channel port of the multi-channel electrically tunable antenna under test is connected to the test equipment via RF cables or connectors to ensure stable signal transmission and low loss. Specifically, the test equipment may include a vector network analyzer (VNA), a phase meter, or an integrated test platform, which supports multi-port switching and automated control. For example, a multi-channel RF switch matrix can be used to automatically switch channel ports, avoiding manual intervention. During the connection process, port matching, such as 50 ohms impedance, must be ensured, and preliminary calibration must be performed to eliminate the effects of cable delay and loss. This step establishes a reliable test link, facilitating subsequent parameter acquisition.

[0064] Optionally, the multi-channel electrically tunable antenna under test is a semi-finished multi-channel electrically tunable antenna, comprising: a metal reflector; and multiple radiating elements, a feed network, and a phase shifter assembly mounted on the metal reflector; wherein the test is performed before the radome is installed on the semi-finished product. Specifically, performing the test at the semi-finished product stage avoids disassembly and rework after the radome is installed, improving efficiency. The metal reflector serves as a substrate, supporting the array arrangement of the radiating elements; the radiating elements are responsible for signal radiation; and the feed network and phase shifter assembly achieve amplitude and phase adjustment. This optional approach is suitable for production lines, integrating the test module before assembling the radome to achieve automated assembly line operation.

[0065] S402: Using the test equipment, perform phase testing and S-parameter testing on each channel of the antenna in sequence to obtain the actual phase value and actual amplitude value of each channel respectively;

[0066] In this step, each channel is tested sequentially using testing equipment. First, phase testing is performed, for example, using a phase comparator or the phase measurement function of a VNA, scanning within the operating frequency band, such as 3.5GHz-6GHz, to obtain the actual phase value of each channel, taking into account the phase shift after phase shifter adjustment. Second, S-parameter testing is performed, such as measuring the reflection coefficient S11 and transmission coefficient S21, extracting the actual amplitude values, for example, through amplitude response curves. The testing process can be automated and script-controlled, executed sequentially according to channel number, ensuring the integrity and real-time nature of data acquisition. The acquired data is stored digitally, including the amplitude and phase values ​​corresponding to the frequency points, for easy subsequent analysis.

[0067] S403: Compare the acquired actual amplitude value and actual phase value with the preset standard value to determine the amplitude and phase error of each channel;

[0068] In this step, the actual measured values ​​are compared channel by channel with preset standard values. The standard values ​​are set based on antenna design specifications, such as ideally uniform amplitude distribution and linearly increasing phase. Amplitude and phase error calculation includes amplitude error (actual amplitude minus standard amplitude) and phase error (actual phase minus standard phase), which can be implemented using software algorithms, such as MATLAB or dedicated testing software, to calculate the difference. The comparison process considers frequency dependence, calculating the average or peak error for multi-frequency data to ensure comprehensive error assessment.

[0069] S404: Determine whether the amplitude and phase consistency of the antenna is qualified based on whether the amplitude and phase error is within the preset tolerance range, and if it is determined to be unqualified, rework the antenna.

[0070] In this step, based on a preset tolerance range, such as the amplitude error being within a reasonable threshold and the phase error being within an acceptable angle, the consistency is judged. The tolerance range is set according to industry standards and application requirements, for example, referring to the 3GPP specification. If the errors of all channels are within the range, it is judged as qualified; otherwise, it is unqualified and a rework process is triggered, such as adjusting the phase shifter, rewelding the feeding network, or replacing the radiation oscillator. The judgment result can output a report through the software interface, including an error distribution diagram and suggested measures, which is convenient for production optimization.

[0071] Through the above steps, this embodiment realizes the automated testing of the amplitude-phase consistency of multi-channel electronically tunable antennas. In practical applications, the test parameters can be adjusted according to the antenna type, such as adding temperature compensation to simulate environmental impacts. This method has strong scalability and can be integrated into intelligent manufacturing systems, and those skilled in the art can implement the invention based on this without excessive experimentation.

[0072] Furthermore, the amplitude-phase consistency testing of multi-channel electronically tunable antennas in the prior art usually relies on direct statistical analysis of single and limited measured data. The test results obtained by this method are easily interfered by accidental errors, and it is difficult to comprehensively and profoundly reflect the internal statistical distribution law and potential extreme situations of amplitude-phase errors caused by factors such as manufacturing tolerances and environmental changes, resulting in insufficient robustness and predictability of the test conclusions.

[0073] Therefore, this application aims to overcome the limitations of the traditional testing method based on single and limited actual measurements to achieve a more comprehensive, robust, and in-depth automated evaluation of the amplitude-phase consistency of multi-channel electronically tunable antennas.

[0074] See Figure 1 As shown, it is a flowchart of the automated testing method provided by this application, including steps S101 to S104, where:

[0075] S101: Obtain the physical position information of each channel of the multi-channel electronically tunable antenna, and construct an antenna array diagram structure according to the physical position information. Each node in the antenna array diagram structure corresponds to a channel of the multi-channel electronically tunable antenna;

[0076] S102: Collect the actual amplitude values and actual phase values of each channel of the multi-channel electronically tunable antenna, calculate the amplitude error values and phase error values of each channel based on the preset amplitude standard values and phase standard values, and map the amplitude error values and phase error values to the antenna array diagram structure to form a true amplitude-phase error map signal;

[0077] S103: Based on the true amplitude-phase error map signal, a latent space posterior probability distribution characterizing the true amplitude-phase error map signal is generated using a preset graph encoder; at least one latent variable is randomly sampled from the latent space posterior probability distribution, and the latent variable is decoded using a preset graph generator to generate amplitude-phase perturbation samples; the true amplitude-phase error map signal and the amplitude-phase perturbation samples are input to a preset graph discriminator to calculate the authenticity discrimination value of each input signal respectively;

[0078] S104: Based on the true / false discrimination value and the amplitude / phase disturbance sample, calculate the evaluation index of amplitude / phase consistency of the multi-channel electrically adjustable antenna, and determine the multi-channel amplitude / phase consistency test result of the multi-channel electrically adjustable antenna according to the evaluation index.

[0079] Regarding the above S101:

[0080] In practical implementation, the first step is to obtain the physical location information of each channel in the multi-channel electrically tunable antenna under test. Specifically, in actual engineering implementation, the physical coordinate information of each channel on the multi-channel electrically tunable antenna can be extracted from the CAD design drawings, structural design documents, or antenna design database used during antenna design. For example, the precise spatial position of the center point of each channel in a three-dimensional Cartesian coordinate system. Typically, this physical location information can be precisely represented by spatial coordinate values ​​of (x, y, z), where the x, y, and z coordinates represent the horizontal, vertical, and lateral distances of the channel center point in space, respectively.

[0081] First, basic information such as the overall dimensions, unit spacing, unit arrangement, and channel numbering of the multi-channel electrically tunable antenna can be recorded. Based on this information, the spatial coordinates of each channel can be determined using CAD software such as AutoCAD, SolidWorks, or HFSS, or in a database. Alternatively, if there are no direct design documents for the specific channel positions of the multi-channel electrically tunable antenna, actual measurements can be taken using measuring equipment such as a laser rangefinder or a high-precision coordinate measuring machine to obtain the spatial position information of each channel.

[0082] After obtaining this precise spatial location information, computer programs or scripts, such as Python or MATLAB scripts, can be used to read and process the coordinate data to construct the required antenna array structure.

[0083] For example, a graph structure can be defined first, in which each node uniquely corresponds to a channel, and each node is assigned a channel number or identifier to ensure that the one-to-one correspondence between nodes and actual channels is clear and accurate.

[0084] During the construction of the antenna array diagram structure, the acquired channel coordinate data can be imported into the program and stored in the form of a data table, matrix, or data structure. Subsequently, based on this coordinate data, the relative distance relationships between each channel are calculated to determine whether connections are established between nodes. These connection relationships between nodes then form the complete antenna array diagram structure. The specific methods for determining the connection relationships between nodes will be described in detail in subsequent steps and will not be elaborated upon here.

[0085] In this way, the physical location information of each channel of the multi-channel electrically tunable antenna can be accurately obtained, and the corresponding antenna array diagram structure can be constructed, providing a data foundation for further calculation and analysis and amplitude-phase consistency testing in subsequent steps.

[0086] Regarding S102 above:

[0087] In practical implementation, it is necessary to collect the actual amplitude and phase values ​​of each channel of the multi-channel electrically tunable antenna. In actual engineering implementation, a predetermined test excitation signal can first be applied to each channel of the multi-channel electrically tunable antenna, such as a sinusoidal signal of a specific frequency or other specified waveform signal generated by a network analyzer or vector signal generator. Then, the amplitude and phase of the output signal of each channel can be actually measured using, for example, a vector network analyzer or a phase amplitude measuring instrument.

[0088] It is important to note that during the measurement process, standardized coaxial cables or waveguides must be used as feeders, and appropriate calibration must be performed to ensure the accuracy of the measurement data.

[0089] In addition, during the acquisition process, the actual amplitude and phase values ​​obtained from each channel should be recorded. These measurement data are recorded in numerical form of amplitude and phase and stored in a computer or testing system for subsequent data processing and analysis.

[0090] Next, after obtaining the actual amplitude and phase values, the amplitude error and phase error values ​​for each channel can be calculated based on the preset amplitude and phase standard values.

[0091] For example, the amplitude error value can be calculated by subtracting the actual measured amplitude value from the preset amplitude standard value to obtain the difference, which is then used as the amplitude error value.

[0092] Similarly, the phase error value is calculated by subtracting the actual measured phase value from the preset phase standard value to obtain the corresponding phase error value.

[0093] Among them, the preset amplitude standard value is usually the target amplitude value of the design, and the preset phase standard value is usually the target phase value of the design.

[0094] After obtaining the above amplitude error values ​​and phase error values, in order to facilitate subsequent processing based on the graph structure, these error values ​​need to be mapped onto the antenna array graph structure constructed in step S101 to form a real amplitude and phase error graph signal.

[0095] In practice, this mapping process can be implemented by writing computer scripts or programs to store the calculated amplitude and phase error values ​​of each channel in the corresponding nodes of the previously constructed antenna array structure in the form of node attributes or node feature vectors.

[0096] Specifically, each node corresponds to a channel, and its attributes can be set to the amplitude error and phase error information of the channel corresponding to that node.

[0097] In this way, a complete and true amplitude and phase error map signal can be obtained, including the amplitude and phase error values ​​of all channels, and mapped onto the constructed antenna array diagram structure, providing a reliable data foundation for subsequent calculations, analysis, and amplitude and phase consistency tests based on this diagram structure.

[0098] Regarding the above S103:

[0099] In specific implementation, this embodiment will use the real amplitude and phase error map signal obtained in the aforementioned step S102 to generate amplitude and phase disturbance samples and perform authenticity discrimination processing using a preset map encoder, map generator and map discriminator.

[0100] First, for the real amplitude and phase error map signal constructed in the aforementioned step S102, this embodiment uses a pre-trained or constructed graph encoder for processing to obtain a latent space posterior probability distribution that can effectively characterize the features of the input signal.

[0101] Specifically, in implementation, a deep learning framework written in Python, such as PyTorch or TensorFlow, can be used to build a graph encoder network. This graph encoder network typically contains multiple graph convolutional layers, which can be implemented using standardized graph convolutional modules (such as Graph Convolution Network, GCN) or graph attention modules (GAT). The features input to each node of the graph encoder are a feature vector composed of the magnitude error value and the phase error value of the corresponding node.

[0102] In actual operation, the true amplitude and phase error map signal propagates layer by layer through the multi-layer graph convolutional network of the graph encoder. The information of each node in each layer is weighted, aggregated, and its features are extracted from the information of its neighboring nodes. This is then processed by a non-linear activation function such as ReLU, and finally passed through a fully connected network to map the features of each node into a latent space, thereby generating the posterior probability distribution of that latent space. In practice, this posterior probability distribution is typically represented by the mean and variance parameters of the latent variables corresponding to each node.

[0103] Then, in this embodiment, at least one latent variable is randomly sampled from the generated latent space posterior probability distribution. Specifically, this can be achieved using a standardized random sampling method, such as a reparameterization method. For example, computer scripting code can be used to implement the random sampling operation of the latent variable distribution parameters for each node, thereby obtaining the specific latent variable.

[0104] Next, this embodiment utilizes a preset graph generator to decode the sampled latent variables, thereby generating amplitude and phase perturbation samples. The graph generator can also be implemented using a decoding network based on a graph convolutional network, which typically includes several graph convolutional layers and fully connected layers. Specifically, the latent variables are programmed to pass and process information layer by layer through the graph generator, decoding to generate amplitude and phase perturbation samples with the same structure as the true amplitude and phase error graph signal, i.e., containing the perturbation amplitude error value and perturbation phase error value corresponding to each node.

[0105] Finally, in this embodiment, the obtained true amplitude and phase error map signal and the generated amplitude and phase perturbation samples are respectively input into a preset graph discriminator for processing. The graph discriminator is usually implemented based on a graph convolutional network, containing multiple graph convolutional layers and fully connected layers, which are used to output the true or false discrimination value of each input signal.

[0106] In implementation, the graph discriminator first performs layer-by-layer graph convolution feature extraction and spatial information aggregation on the input signal to obtain the feature representation of each node, and then obtains the overall graph-level feature representation of the input signal through global pooling. Subsequently, the fully connected network of the graph discriminator further processes the above-mentioned node features and graph-level features to obtain global true / false discrimination values ​​representing the overall authenticity of the input signal and local true / false discrimination values ​​representing the local authenticity of each node. Finally, the above discrimination values ​​are combined to obtain a composite true / false discrimination result.

[0107] This completes the process of generating amplitude and phase error diagram samples from the actual amplitude and phase error diagram signal, and determining the authenticity of both. The above implementation steps can be run on a computing platform or server with graph processing capabilities using specially written computer programs or code, thereby achieving automated and efficient processing and providing necessary data support for subsequent amplitude and phase consistency evaluation.

[0108] Regarding S104 above:

[0109] In practice, the amplitude error value and phase error value corresponding to each channel in the amplitude and phase disturbance sample generated in step S103 can be statistically analyzed by a computer program to automatically calculate the evaluation index reflecting the amplitude and phase consistency of the multi-channel electrically tunable antenna.

[0110] For example, in the specific implementation process, the computer program can automatically read the amplitude error value in the generated perturbation sample and automatically calculate the statistical indicators such as the standard deviation, range or coefficient of variation of the amplitude error of all channels to quantitatively characterize the amplitude consistency between channels; similarly, it can automatically read the phase error value and calculate the statistical indicators such as the standard deviation and range of the phase error of all channels to quantitatively characterize the phase consistency between channels.

[0111] Furthermore, the computer program can automatically read the authenticity discrimination values ​​obtained in step S103, including, for example, the global authenticity discrimination value and the local authenticity discrimination values ​​corresponding to each channel, and automatically filter and label the perturbation samples according to the pre-stored authenticity discrimination thresholds. Specifically, if the authenticity discrimination value corresponding to a certain perturbation sample exceeds the threshold, the computer program automatically labels the sample as a reliable perturbation sample and includes it in the calculation range of the amplitude-phase consistency evaluation index; otherwise, it automatically labels it as an abnormal or unreliable sample, thereby automatically completing the screening of reliable perturbation samples.

[0112] Finally, the computer program or testing system automatically compares the obtained amplitude and phase consistency indices with engineering evaluation standards pre-stored in the testing system or database. In practice, allowable ranges or standard thresholds for indices such as amplitude error standard deviation and phase error standard deviation can be pre-set within the computer program. The program automatically compares the calculated indices with the preset thresholds and automatically determines whether the currently tested multi-channel electrically tunable antenna meets the technical requirements for multi-channel amplitude and phase consistency.

[0113] For example, when all the calculated evaluation indicators fall within the pre-stored allowable range, the computer program automatically marks the multi-channel electrically adjustable antenna test results as qualified; if any indicator is automatically detected to be outside the allowable range, the test results are automatically marked as unqualified, and a test report is automatically generated, indicating the unqualified channel and corresponding indicator, for subsequent analysis and adjustment.

[0114] Thus, the testing method of this application differs from traditional methods that directly analyze single, limited measured data. This invention utilizes a graphical model based on the antenna's physical topology to learn the inherent error statistical characteristics exhibited by a single actual measurement. Based on these learned statistical characteristics, the method can generate multiple sets of simulated amplitude and phase perturbation samples, and the final evaluation index is derived from the overall statistical analysis of these simulated samples. Therefore, this method reduces the interference of random errors in a single measurement on the final conclusion and improves the stability of the test results. Simultaneously, since the simulated samples can cover potential error situations that are unlikely to occur in actual measurements, the evaluation results of this method are more comprehensive. Furthermore, by introducing the graphical structure of the antenna array, the model can consider the spatial correlation between channels in the analysis, making its error modeling closer to physical reality, thereby improving the accuracy of the test.

[0115] As an optional implementation, please refer to Figure 2 The flowchart provided in this application illustrates a method for constructing an antenna array diagram structure based on the physical location information, including steps S201 to S204, wherein:

[0116] S201: Obtain the physical coordinate information of each channel of the multi-channel electrically adjustable antenna;

[0117] S202: Calculate the Euclidean distance between any two channels based on the physical coordinate information, and convert the Euclidean distance into the corresponding electrical length according to the preset test signal center frequency;

[0118] S203: A first distance threshold and a second distance threshold are preset to distinguish between strongly coupled near-field effects and weakly coupled far-field effects, wherein the first distance threshold is smaller than the second distance threshold;

[0119] S204: Based on the Euclidean distance, the connection relationship between any two channels is divided into a first connection relationship or a second connection relationship that are mutually exclusive, wherein the first connection relationship corresponds to the case where the Euclidean distance is less than or equal to the first distance threshold, and the second connection relationship corresponds to the case where the Euclidean distance is greater than the first distance threshold and less than or equal to the second distance threshold.

[0120] In practice, the physical coordinate information of each channel is first automatically read using computer scripts or programs. This information is typically represented in three-dimensional Cartesian coordinates (x, y, z), and the coordinate information for each channel is stored in an array or data table, serving as the data source for program input.

[0121] The program then automatically iterates through the coordinate information of all channels and calculates the spatial Euclidean distance between any two channels. Specifically, the program can perform distance calculations as follows: for the coordinate data of each pair of channels, it calculates the distance difference along the corresponding coordinate axis, and automatically calculates the actual spatial distance between any two channels according to the definition of standard Euclidean distance.

[0122] Then, based on the preset test signal center frequency in this embodiment, which is typically preset in a program or system database, the computer program automatically converts the obtained spatial Euclidean distance between each pair of channels into the corresponding electrical length. Specifically, the electrical length is calculated as follows: the program automatically obtains the corresponding signal wavelength based on the test signal center frequency, for example, by calculating it using the relationship between electromagnetic wave speed and frequency. Then, it automatically divides the actual spatial distance between each pair of channels by the wavelength to obtain the electrical length corresponding to that distance. This conversion makes the determination of connection relationships in the graph structure closer to the actual laws governing electromagnetic field action.

[0123] Furthermore, in practical implementation, the computer program pre-stores two distance thresholds: a first distance threshold and a second distance threshold, to delineate different electromagnetic coupling regions. The first distance threshold corresponds to the near-field effect boundary of strong coupling, while the second distance threshold corresponds to the far-field effect boundary of weak coupling, and the first distance threshold is significantly smaller than the second distance threshold. These two distance thresholds are generally determined in advance based on empirical values ​​or engineering simulation results and stored in the computer program or system database for automatic program recall.

[0124] Next, the computer program automatically compares the electrical length between each pair of channels with the stored first and second distance thresholds, and determines the connection relationship between the nodes accordingly. The specific implementation is as follows:

[0125] If the electrical length between any two channels is less than or equal to the first distance threshold, the program automatically determines that there is a first connection between the two nodes, representing a strong coupling near-field effect;

[0126] If the electrical length between any two channels is greater than the first distance threshold and less than or equal to the second distance threshold, the program automatically determines that there is a second connection relationship between the two nodes, representing a weak coupling far-field effect.

[0127] If the electrical length between any two channels exceeds the second distance threshold, the program will automatically determine that there is no connection between the two nodes, that is, there is no obvious electromagnetic coupling effect between the nodes.

[0128] Through the above automated implementation method, the connection relationship between every two channels can be automatically determined, ensuring the mutual exclusivity and clarity of the connection relationship.

[0129] Finally, the computer program automatically records and stores the determined connection relationships, including the first and second connection relationships, and the corresponding node information, ultimately forming an antenna array diagram structure with practical physical meaning. The resulting antenna array diagram structure can fully represent the electromagnetic coupling relationships between each channel, accurately reflect the electromagnetic physical characteristics of the actual antenna structure, and provide a precise analytical framework and computational basis for subsequent amplitude and phase error analysis and consistency testing.

[0130] As an optional implementation, the step of constructing the antenna array diagram structure based on the physical location information further includes:

[0131] For each pair of channels with the first connection relationship or the second connection relationship, the normalized connection weight value of the corresponding connection relationship is calculated based on the electrical length and the preset path loss model.

[0132] Based on the channels, the first connection relationship, the second connection relationship, and the connection weight values, a weighted heterogeneous graph structure is constructed to characterize the multi-scale electromagnetic coupling characteristics between channels of a multi-channel electrically tunable antenna.

[0133] In specific implementation, this embodiment, based on determining the first and second connection relationships between the multi-channel electrically tunable antenna channels, further constructs a weighted heterogeneous graph structure that can reflect the multi-scale electromagnetic coupling characteristics between channels. The specific implementation process is as follows:

[0134] First, for each pair of channels with a first connection relationship or a second connection relationship, the normalized connection weight value of the corresponding connection relationship is calculated based on the electrical length information determined in the aforementioned steps and the preset path loss model.

[0135] In practice, the path loss model can be the free space path loss model widely used in engineering or a more refined empirical path loss model.

[0136] For example, during implementation, the electromagnetic wave attenuation of the signal along the propagation path can be determined based on the electrical length between channels, thereby quantitatively characterizing the coupling strength between the two channels. Specifically, attenuation data corresponding to different electrical lengths can be collected in advance, and a data table or interpolation function can be established as a means of implementing the path loss model, thereby achieving accurate prediction of the coupling strength under arbitrary electrical length conditions.

[0137] Furthermore, to ensure that the weight values ​​of different connection relationships have a uniform scale in subsequent calculations, implementers can normalize the calculated connection strength values.

[0138] For example, in practical implementation, a linear normalization method can be adopted, which unifies all connection strength values ​​to the range of 0 to 1, where the strongest connection corresponds to a normalized weight value of 1 and the weakest connection corresponds to a normalized weight value of 0, so as to facilitate direct comparison of different connection relationships and subsequent unified processing.

[0139] After calculating the normalized connection weights, a complete weighted heterogeneous graph structure can be constructed based on the determined channel information, the first connection relationship, the second connection relationship, and the corresponding normalized weights.

[0140] In practical implementation, existing graph processing software or tools, such as open-source graph computing libraries like NetworkX and PyTorchGeometric, can be used to realize the weighted heterogeneous graph structure of the antenna array. For example, each antenna channel is first treated as a node in the graph, and each node is assigned a unique identifier. Then, for the channel node pairs with the aforementioned first or second connection relationship, corresponding edge connections are created, and the calculated normalized connection weight values ​​are assigned to the corresponding edges and stored as their attributes.

[0141] The resulting weighted heterogeneous graph structure reflects the actual electromagnetic coupling relationships at different scales between the channels of the multi-channel electrically tunable antenna, and accurately characterizes the differences in coupling strength between channels through normalized weight values, thus providing an accurate and sufficient data foundation for subsequent more refined amplitude and phase consistency analysis.

[0142] As an optional implementation, see [link to implementation details]. Figure 3 The flowchart provided in this application illustrates a method for generating a true amplitude-phase error map signal, including steps S301 to S304, wherein:

[0143] S301: Based on the preset maximum amplitude error range, normalize the amplitude error value of each channel to obtain the normalized amplitude error value corresponding to that channel;

[0144] S302: Perform a trigonometric function transformation on the phase error value of each channel to map the phase error value to a two-dimensional Cartesian coordinate space to obtain the two-dimensional phase error vector corresponding to that channel;

[0145] S303: Concatenate the normalized amplitude error value and the two-dimensional phase error vector corresponding to each channel to generate the node initial feature vector corresponding to each channel;

[0146] S304: Assign the initial feature vector of each channel as a node attribute to the corresponding node in the weighted heterogeneous graph structure to generate the true amplitude and phase error map signal.

[0147] In specific implementation, in order to effectively represent the amplitude error value and phase error value obtained in step S102 on the antenna array diagram structure, so as to form a more suitable real amplitude and phase error diagram signal for subsequent processing, the following steps can be performed:

[0148] First, based on engineering requirements and the preset allowable amplitude error range, a maximum amplitude error value is selected as the normalization benchmark for amplitude error. For example, this value can be determined empirically based on the design specifications of multi-channel electrically adjustable antennas or historical measurement data, serving as a unified amplitude error scale.

[0149] Subsequently, for the amplitude error value calculated for each channel, linear normalization can be applied. The actual amplitude error value is divided by a preset maximum amplitude error benchmark value, so that the normalized amplitude error value is uniformly limited to the range of 0 to 1. The normalized amplitude error value is the normalized amplitude error value for each channel, thereby achieving a unified scale of amplitude error between different channels, which facilitates further analysis and comparison.

[0150] To eliminate the influence of phase periodicity and ensure the continuity and directionality of the phase error representation during implementation, the original phase error value can be mapped to a two-dimensional Cartesian coordinate space using trigonometric function transformations. For example, in practice, the phase error value of each channel can be subjected to cosine and sine transformations, with the cosine value used as the horizontal coordinate and the sine value as the vertical coordinate, thereby forming a two-dimensional phase error vector for each channel.

[0151] Furthermore, to comprehensively characterize the error information of each channel, the normalized amplitude error value and the two-dimensional phase error vector corresponding to each channel obtained in the previous step can be concatenated or cascaded to form the initial feature vector of the node for that channel. For example, in a specific implementation, the normalized amplitude error value can be used as the first dimension of the feature vector, and then the two-dimensional phase error vector can be added sequentially to form an initial feature vector of node dimension 3, reflecting the amplitude and phase error information of each node (channel).

[0152] Finally, the initial feature vector of each channel is assigned as a node attribute to the corresponding node in the weighted heterogeneous graph structure established in step S103 to generate a complete true amplitude and phase error map signal. In specific implementation, the node attribute storage mechanism provided by the graph computing library can be used to store the initial feature vector of the node and the corresponding node in a one-to-one correspondence, thereby obtaining a true amplitude and phase error map signal with a clear structure.

[0153] In this way, the real amplitude and phase error map signal generated by this application can more clearly, accurately and completely reflect the actual error characteristics of each channel of the multi-channel electrically tunable antenna. This not only facilitates further analysis and processing of the graph neural network model, but also provides a precise and rich data foundation for the automated testing of the amplitude and phase consistency of the antenna multi-channel.

[0154] As an optional implementation, generating the latent space posterior probability distribution for characterizing the true amplitude-phase error map signal includes:

[0155] Using the multi-layer graph convolutional network built into the graph encoder, the node initial feature vectors of the real amplitude and phase error graph signal are aggregated layer by layer to obtain the node output feature vector of each node.

[0156] Perform a residual join operation between the node output feature vector of each node and the node initial feature vector corresponding to that node to obtain a residual feature vector including the initial error features;

[0157] Perform a global pooling operation on the residual feature vectors corresponding to all nodes to obtain a graph-level context vector for the entire antenna array graph structure;

[0158] The residual feature vector corresponding to each node is concatenated with the graph-level context vector to generate a fused feature vector that integrates local node information and global graph information.

[0159] The graph encoder uses a fully connected layer network to map the fused feature vector of each node to the latent space, generating the mean and variance parameters corresponding to the posterior probability distribution of the latent space.

[0160] In specific implementation, in order to effectively extract the intrinsic statistical features of amplitude and phase errors from the real amplitude and phase error map signal constructed in the aforementioned steps, and accurately generate the latent space posterior probability distribution characterizing the real amplitude and phase error map signal, the specific implementation process is as follows:

[0161] First, the graph encoder used in this embodiment can be implemented based on a commonly used deep learning framework and has several layers of graph convolutional networks, such as GCN or GAT, built in to achieve effective information aggregation and feature extraction of input node features.

[0162] Specifically, each node in the real amplitude and phase error map signal has an initial feature vector of the node determined in the aforementioned steps, which may include, for example, a combination of normalized amplitude error value and two-dimensional phase error vector.

[0163] During implementation, the initial feature vectors of these nodes are used as input features and fed into the first graph convolutional layer of the graph encoder. Subsequently, through the layer-by-layer processing of the graph convolutional network, the features of each node are gradually fused with information from itself and its neighboring nodes to form the node's output feature vector.

[0164] For example, the output feature vector of each node can be achieved by weighted summation of the feature vectors of neighboring nodes or by attention aggregation. Taking a typical graph convolution process as an example, the features of a node itself will interact with the features of its neighboring nodes through graph convolution operators to extract features. After, for example, 2 to 4 layers of graph convolution, a node output feature vector with richer node hierarchy and higher order can be obtained.

[0165] Subsequently, to further enhance the network's ability to capture amplitude and phase error details, this embodiment can perform a residual concatenation operation on the output feature vector of each node and its corresponding initial feature vector. Specifically, this involves adding or concatenating the node's output feature vector with the initial feature vector dimension by dimension, thereby obtaining a node residual feature vector that contains both initial error information and depth features extracted by higher-order graph convolution.

[0166] After obtaining the residual feature vectors corresponding to all nodes, in order to enable the graph encoder to have a holistic grasp of the entire graph structure, this embodiment further implements a global pooling operation.

[0167] In practical implementation, methods such as global mean pooling or global maximum pooling can be used to pool the residual feature vectors of all nodes along their feature dimensions, thereby forming a unified graph-level context vector. This graph-level context vector can effectively capture the amplitude and phase error distribution characteristics of the entire antenna array, providing global information support for the subsequent generation of the latent space probability distribution.

[0168] Furthermore, to more comprehensively reflect the relationship between local node features and global graph information, this embodiment concatenates the residual feature vector corresponding to each node with the graph-level context vector. For example, the two types of features can be concatenated along the feature dimension. Exemplarily, through feature vector concatenation, the final fused feature vector of each node contains both the local features of the node and the global information of the overall graph structure.

[0169] Finally, in this embodiment, the fused feature vector is mapped to the latent space through the fully connected layer network built into the graph encoder. In specific implementation, the fully connected layer network can be composed of several linear transformation layers and nonlinear activation functions, such as ReLU, LeakyReLU, etc., to gradually project the fused feature vector into the latent space. Through the final mapping of the network, the fused feature vector corresponding to each node will generate mean and variance parameters representing the posterior probability distribution of the latent space.

[0170] For example, when building a graph encoder, the last layer can be set to two parallel fully connected layers, which output the mean and variance parameters of the latent variables for each node, respectively, thus representing the latent space probability distribution of each node. These two parameters can be stored as node attributes or in separate data structures for use in subsequent steps when randomly sampling latent variables.

[0171] In this way, this application can effectively learn and extract the latent space posterior probability distribution that accurately represents the intrinsic statistical characteristics of the amplitude and phase errors of each channel from the real amplitude and phase error map signal. This lays a solid data and technical foundation for subsequent random generation of amplitude and phase perturbation samples and further automated amplitude and phase consistency testing.

[0172] As an optional implementation, obtaining the node output feature vector of each node includes:

[0173] For each layer of the multilayer graph convolutional network, the following feature update operation is performed on each node of the antenna array graph structure:

[0174] For neighboring nodes with the first connection relationship and neighboring nodes with the second connection relationship, independent weighted information aggregation operations are performed based on the normalized connection weight values ​​corresponding to each neighboring node to obtain intermediate feature vectors that respectively characterize the near-field effect and the far-field effect.

[0175] Linear transformations are performed on the intermediate feature vectors representing the near-field and far-field effects, respectively, and the linearly transformed feature vectors are fused to generate an aggregated feature vector.

[0176] Based on the aggregated feature vector and the previous layer feature vector of the node itself, the feature vector of the node is updated to obtain the updated node feature vector.

[0177] A non-linear activation process is performed on the updated node feature vector to obtain the node output feature vector of the node in the current layer.

[0178] In practical implementation, to more accurately and effectively express the amplitude and phase error characteristics between channels of a multi-channel electrically tunable antenna and fully reflect the multi-scale electromagnetic coupling relationship between channels, this embodiment improves the feature update mechanism in the graph convolutional network of the graph encoder. The specific implementation process is as follows:

[0179] When implementing a multi-layer graph convolutional network, each layer needs to perform a feature update operation for each node in the antenna array graph structure.

[0180] Specifically, when updating node features, the neighborhood of the node to be updated is first divided. That is, the neighboring nodes with the first connection relationship with the node are divided into one group, and the neighboring nodes with the second connection relationship are divided into another group, thereby reflecting the spatial electromagnetic coupling characteristics of strong coupling near-field effect and weak coupling far-field effect.

[0181] For the neighbor node group with the first connection relationship, during implementation, weighted information aggregation is performed using the node feature vector corresponding to each neighbor node and the pre-calculated normalized connection weight value. For example, each neighbor node feature vector can be multiplied by its corresponding normalized connection weight value, and then these weighted neighbor node feature vectors can be summed or averaged to generate an intermediate feature vector representing the strong coupling near-field effect.

[0182] For the neighbor node group of the second connection relationship, the implementation method is the same as described above, that is, the feature vector of each neighbor node and the corresponding normalized connection weight value are independently weighted and aggregated to generate an intermediate feature vector representing the weak coupling far-field effect.

[0183] The two intermediate feature vectors obtained in the above manner accurately reflect the near-field and far-field feature information of the current node, avoiding the defect of ignoring the electromagnetic coupling relationship at different scales in traditional graph convolution operations.

[0184] Furthermore, after obtaining the two intermediate feature vectors for the near-field and far-field effects, this embodiment performs independent linear transformation operations on these two feature vectors to enhance the feature representation at different electromagnetic scales. For example, in a specific implementation, two different sets of learnable weight matrices can be used to perform linear transformation operations on the intermediate feature vectors of the near-field and far-field effects respectively, generating enhanced intermediate feature vectors. This implementation ensures that the electromagnetic effect features at different scales can be independently and specifically enhanced and extracted.

[0185] Subsequently, the two intermediate feature vectors after linear transformation are fused to obtain a unified aggregated feature vector. In practice, the two feature vectors can be concatenated according to their feature dimensions or directly added before generating the fused aggregated feature vector through linear mapping, thereby achieving a unified representation of near-field and far-field effects.

[0186] After obtaining the aggregated feature vector, this embodiment further updates the node features by combining the node's own feature vector from the previous layer. For example, the aggregated feature vector can be added to or concatenated with the node's feature vector from the previous layer, and then another independent linear transformation layer can be used to fuse the two into the updated node feature vector. This implementation process ensures that the updated node features fully reflect the electromagnetic coupling information of neighboring nodes while retaining the node's own historical feature information, thus enhancing the expressive power of the node features.

[0187] Finally, to further improve the network's expressive power and nonlinear modeling capabilities, this embodiment processes the updated node feature vectors using a nonlinear activation function to obtain the final output feature vectors of the nodes in the current graph convolutional layer. For example, commonly used ReLU or LeakyReLU activation functions can be selected to enhance the expressive power of the feature vectors through element-wise nonlinear transformation, making the output node feature vectors more suitable for subsequent feature extraction and latent space mapping.

[0188] This approach realizes a graph convolution feature update method sensitive to multi-scale electromagnetic coupling, which not only significantly improves the expression accuracy of node features, but also effectively avoids the limitation of traditional methods that ignore coupling relationships at different scales. This provides a more accurate and effective feature extraction method and data foundation for automated testing of amplitude and phase consistency of multi-channel electrically tunable antennas.

[0189] As an optional implementation, the generation of amplitude and phase perturbation samples includes:

[0190] The latent variables obtained from the latent space posterior probability distribution are decomposed into global latent variables and local latent variables.

[0191] The global latent variables are processed using a first neural network to generate a global bias vector;

[0192] The local latent variables are assigned to each node in the weighted heterogeneous graph structure as initial random features of the nodes.

[0193] In practical implementation, to efficiently generate samples that can accurately reflect the amplitude and phase perturbation characteristics of multi-channel electrically tunable antennas and facilitate subsequent amplitude and phase consistency analysis, this embodiment proposes an amplitude and phase perturbation sample generation method, the specific implementation process of which is as follows:

[0194] First, based on the latent space posterior probability distribution obtained in the preceding steps, latent variables are sampled. These latent variables are typically random vectors defined in the latent space by the graph encoder, representing high-level feature representations of the true amplitude-phase error map signal. In practical implementations, for example, a reparameterization method can be used to obtain the latent variable vector corresponding to each node through random sampling from a standard normal distribution.

[0195] Subsequently, in order to model the overall consistency deviation of the multi-channel electrically tunable antenna and the differences between local random disturbances in each channel in more detail, this embodiment decomposes the obtained latent variable vector into two parts. One part is defined as global latent variables, which are used to describe the overall disturbance trend of the entire multi-channel electrically tunable antenna; the other part is defined as local latent variables, which are used to describe the differences in local random disturbances in each antenna channel.

[0196] In practice, the latent variable vector can be divided according to the predetermined dimensions. For example, the first part of the latent variable vector can be used as global latent variables, and the second part can be used as local latent variables.

[0197] Furthermore, to accurately generate a feature representation describing the overall consistency deviation of the multi-channel electrically tunable antenna, this embodiment processes the aforementioned global latent variables through a pre-defined first neural network to generate a global bias vector. In implementation, the first neural network typically consists of several fully connected layers. For example, the input layer receives the global latent variable vector, which undergoes linear transformation and nonlinear activation operations through several hidden layers, and finally generates the global bias vector via the output layer. This global bias vector is specifically represented as a bias value with the same dimension as the node feature vector, used to describe the overall amplitude and phase error offset trend of the antenna.

[0198] After processing the global latent variables and obtaining the global bias vector, this embodiment further assigns local latent variables to each node of the antenna array graph structure, making them the initial random features of each node. In actual implementation, each local random variable vector in the local latent variables can be matched one by one with the corresponding node in the weighted heterogeneous graph structure, and these local latent variables can be directly assigned as the initial feature information of the node. This assignment method enables the subsequent graph neural network to consider the individual perturbation differences of each channel during the node feature initialization stage, thereby achieving more detailed and accurate local perturbation feature modeling.

[0199] To ensure that the generated amplitude and phase perturbation samples accurately reflect the actual amplitude and phase error characteristics, after obtaining the initial random features of the nodes, further implementation typically requires integrating or fusing the features of each node with the aforementioned global bias vector. For example, during implementation, the initial random features of the nodes can be added element-wise to the global bias vector to generate the final initial perturbation feature vector of the nodes. This integration method allows the perturbation features of each node to simultaneously reflect both the overall consistency deviation trend and individual random differences, thereby generating more realistic and effective amplitude and phase perturbation samples.

[0200] In this way, by explicitly dividing global and local perturbation information, the amplitude and phase error characteristics of multi-channel electrically tunable antennas can be described more accurately and comprehensively. By specifically implementing the fusion of global bias characteristics and local random characteristics, this embodiment can generate amplitude and phase perturbation samples with physical meaning and accurate feature representation, effectively supporting the high-quality implementation of subsequent automated amplitude and phase consistency testing of multi-channel electrically tunable antennas.

[0201] As an optional implementation, the generation of amplitude and phase perturbation samples further includes:

[0202] Using the graph convolutional network built into the graph generator, multi-layer information propagation operations are performed on the initial random features of each node to obtain the local perturbation vector corresponding to each node. Each layer of information propagation includes:

[0203] For each node, the first weighted information aggregation operation is performed on the neighboring nodes connected to that node through the first connection relationship;

[0204] For each node, a second weighted information aggregation operation is performed on the neighboring nodes connected to that node through the second connection relationship;

[0205] The results of the first and second weighted information aggregation operations are fused to generate updated node features;

[0206] For each node, the updated node features are combined with the global bias vector to perform vector synthesis, and the synthesized vector is subjected to inverse feature transformation to obtain the amplitude error value and phase error value corresponding to each channel in the amplitude and phase perturbation sample.

[0207] To effectively generate amplitude and phase perturbation samples for multi-channel electrically tunable antennas and accurately simulate the amplitude and phase perturbations between channels, this embodiment uses a graph convolutional network inside the graph generator to perform multi-layer information propagation operations on the initial random features of the nodes to obtain amplitude and phase perturbation samples. The specific implementation steps are as follows:

[0208] First, based on the weighted heterogeneous graph structure determined in the preceding steps, the initial random features of each node are used as input data. The multi-layer graph convolutional network built within the graph generator performs information propagation and node feature updates layer by layer. Each layer of the multi-layer graph convolutional network includes two key processes: weighted aggregation of neighboring node information and updating of its own node features.

[0209] In the implementation process, taking a certain layer of graph convolution as an example, for each node in the antenna array graph structure, two independent weighted information aggregation operations are performed on the information of neighboring nodes. Specifically:

[0210] The first weighted information aggregation operation targets neighboring nodes connected to the current node through the first connection relationship. These nodes represent near-field adjacent channels with strong coupling effects in the antenna array. In practice, information aggregation can be performed by weighted summation, that is, multiplying the current feature vector of each neighboring node by a pre-calculated normalized connection weight value and then summing the results to obtain the information aggregation result for the near-field neighboring nodes, reflecting the disturbance characteristics generated by near-field electromagnetic coupling.

[0211] The second weighted information aggregation operation targets the neighboring nodes connected to the current node through the second connection relationship, which represents the far-field neighbor channel with weak coupling effect.

[0212] During implementation, the normalized connection weight values ​​are also used to perform a weighted summation operation on the feature vectors of each far-field neighbor node to obtain the information aggregation result for the far-field neighbor node, reflecting the propagation of far-field perturbation features.

[0213] Subsequently, this embodiment further fuses the aforementioned near-field neighbor aggregation features and far-field neighbor aggregation features. Specifically, this fusion process can be achieved using a linear transformation, for example, by performing a linear mapping transformation on each of the two aggregation results, followed by element-wise addition or concatenation, to generate the updated feature vector for the current node. This fusion process reflects the combined effect of electromagnetic coupling at different scales, thus more realistically and accurately describing the perturbation characteristics between channels.

[0214] After completing the node feature update, the updated node feature vector is further combined with the previously generated global bias vector.

[0215] For example, the updated feature vector of the node is added element by element to the global bias vector to explicitly consider the influence of the overall bias trend on the local perturbation features, and finally a synthetic feature vector that reflects both global and local perturbation information is obtained.

[0216] To generate the final amplitude and phase perturbation samples, this embodiment also requires performing an inverse feature transformation operation on the synthesized vector. In practice, a pre-trained linear or nonlinear transformation network can be used, for example, through a multi-layer fully connected network, to map the synthesized feature vector back to the original physical error space, outputting channel amplitude error values ​​and phase error values ​​with physical meaning.

[0217] In practice, for example, the synthesized feature vector is input into an inverse feature transformation neural network containing multiple hidden layers. Each hidden layer performs linear transformation and nonlinear activation, and finally the amplitude error value and phase error value corresponding to each node are obtained in the output layer.

[0218] In this way, the graph generator's built-in graph convolutional network enables accurate aggregation and effective propagation of multi-scale neighbor node information of the initial random features of nodes. Furthermore, global bias information is fused, and through inverse feature transformation, samples that accurately reflect the actual amplitude and phase perturbation characteristics of multi-channel electrically tunable antennas are explicitly generated. This provides a sufficient, realistic, and effective sample foundation for subsequent automated amplitude and phase consistency testing.

[0219] As an optional implementation, the step of inputting the true amplitude-phase error map signal and the amplitude-phase perturbation sample into a preset map discriminator, and calculating the authenticity discrimination value of each input signal, includes:

[0220] Using the multi-layer graph convolutional network built into the graph discriminator, the input real amplitude and phase error graph signal or the amplitude and phase perturbation sample is processed by layer-by-layer spatial information aggregation to obtain the final feature vector of each node.

[0221] The final feature vector of each node is subjected to global pooling to obtain the graph-level feature vector corresponding to the input signal.

[0222] The graph-level feature vectors are processed using a first fully connected network to obtain a global authenticity value that characterizes the overall authenticity of the input signal.

[0223] The final feature vector of each node is processed by the second fully connected network to obtain the local authenticity discrimination value that represents the local authenticity of each node.

[0224] The global true / false discrimination value and the local true / false discrimination values ​​corresponding to all nodes are combined to form a composite true / false discrimination result.

[0225] To effectively determine the authenticity of the input true amplitude and phase error map signal and the generated amplitude and phase perturbation sample, this embodiment utilizes a pre-built and trained graph discriminator to calculate the authenticity discrimination value for each input signal. The specific implementation steps are as follows:

[0226] First, the initial feature vectors of the nodes of the real amplitude and phase error map signal to be judged or the generated amplitude and phase perturbation sample are input into the pre-trained graph discriminator.

[0227] The graph discriminator internally comprises multiple layers of graph convolutional networks to achieve layer-by-layer spatial information aggregation processing of the input signal. Specifically, for each node, the graph discriminator extracts and aggregates neighboring node information layer by layer through the graph convolutional networks. The implementation of each layer of the graph convolutional network is as follows: after weighted summation of the feature vectors of all neighboring nodes for each node, a linear transformation and a non-linear activation function are applied to obtain the node feature representation of each node in the current layer. This process is continued layer by layer until the final feature vector of each node is obtained.

[0228] Subsequently, to evaluate the overall authenticity of the input signal, the graph discriminator performs global pooling on the final feature vectors of all nodes. In practice, commonly used global mean pooling or max pooling methods can be employed. For example, the feature vectors of all nodes can be averaged element-wise to obtain a graph-level feature vector of fixed dimensions, thereby effectively representing the overall structural features of the input signal.

[0229] Furthermore, in this embodiment, a pre-trained first fully connected network is used to process the graph-level feature vectors obtained above, so as to calculate and output the global true / false discrimination value corresponding to the input signal.

[0230] For example, a graph-level feature vector is input into a fully connected network containing several hidden layers. After each hidden layer undergoes a linear transformation, it is processed by a nonlinear activation function. Finally, a scalar global authenticity value is obtained in the output layer, representing the overall authenticity or forgery of the input signal.

[0231] Meanwhile, in order to further accurately assess the authenticity of the local input signal, this embodiment also uses a pre-trained second fully connected network to process the final feature vector of each node, thereby calculating the local authenticity discrimination value of each node.

[0232] For example, for each node's feature vector, the input is fed into a specially designed fully connected network containing one or more hidden layers. After layer-by-layer linear transformations and nonlinear activation functions, the final local true / false value is obtained at the output layer. The local true / false value of each node can explicitly characterize whether the local amplitude and phase error information at the corresponding node has true characteristics.

[0233] Finally, the global true / false discrimination value obtained above is effectively combined with the local true / false discrimination value corresponding to each node to form a composite true / false discrimination result.

[0234] For example, the global true / false score can be weighted and averaged or concatenated with the local true / false scores of all nodes, and then processed through a fusion layer to obtain a single composite true / false score. Alternatively, the global and local scores can be output separately for further, more refined analysis.

[0235] This approach simultaneously considers both the overall and local information of the input signal, providing a more comprehensive and reliable determination of whether the signal is genuine or spoofed. This embodiment not only significantly improves the ability to distinguish the authenticity of amplitude and phase disturbance samples during automated antenna amplitude and phase consistency testing, but also further enhances the robustness and reliability of this method in practical engineering applications.

[0236] It should be noted that the graph encoder, graph generator, and graph discriminator mentioned above are all deep neural network models, and can be implemented using supervised or semi-supervised end-to-end training methods.

[0237] First, in actual implementation, it is necessary to obtain a large amount of real amplitude and phase consistency test data of multi-channel electrically tunable antennas in advance. This data can be obtained by conducting actual tests on a large number of antenna samples to obtain the real amplitude and phase error values ​​of each channel as the source of training data, and to calculate the real amplitude and phase error map signal as the sample for model input.

[0238] The graph encoder and graph generator can together form a variational autoencoder structure. During the training phase, the actual amplitude-phase error graph signal obtained earlier is input into the graph encoder to generate the posterior probability distribution of the latent space. Then, latent variables are obtained through random sampling and input into the graph generator to generate amplitude-phase perturbation samples. The difference between the generated amplitude-phase perturbation samples and the actual input signal is then used as the reconstruction error. The model parameters are optimized by defining appropriate loss functions, such as reconstruction error loss and latent space regularization loss. The training process can use stochastic gradient descent (SGD) or the Adam optimizer, with an appropriate learning rate, and multiple training rounds are performed until the model converges.

[0239] For the graph discriminator, its training can be conducted using adversarial training. In practice, the input to the graph discriminator includes the actual measured amplitude-phase error graph signal and the amplitude-phase perturbation samples generated by the graph generator. The training objective is to distinguish between genuine and fake input signals; that is, real data is judged as genuine, and generated data as fake. Specifically, during training, a suitable binary cross-entropy loss function is defined, and the Adam optimizer is used to repeatedly update the discriminator's parameters, enabling it to effectively distinguish between real and generated signals. Simultaneously, the graph generator updates its own parameters during training with the goal of deceiving the discriminator, thus forming an adversarial training mechanism. Through multiple alternating iterative training iterations, the graph generator can generate more realistic perturbation samples, while the graph discriminator can more accurately judge the authenticity of the input signal.

[0240] During implementation, the acquisition of training data, model training, and the setting of optimized hyperparameters can all be achieved using mature deep learning frameworks such as PyTorch and TensorFlow, thereby ensuring that the model can be trained effectively and quickly and has good generalization performance.

[0241] Based on the same inventive concept, this application also provides an automated test system for the amplitude and phase consistency of a multi-channel electrically adjustable antenna, corresponding to the automated test method for amplitude and phase consistency of a multi-channel electrically adjustable antenna. Since the principle of the system in this application is similar to the automated test method for amplitude and phase consistency of a multi-channel electrically adjustable antenna described above, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0242] Reference Figure 4 The diagram shown is a schematic of an automated amplitude and phase consistency testing system for a multi-channel electrically tunable antenna provided in this application. The system includes:

[0243] Construction module 10 is used to obtain the physical location information of each channel of the multi-channel electrically tunable antenna, and construct an antenna array diagram structure based on the physical location information. Each node in the antenna array diagram structure corresponds to one channel of the multi-channel electrically tunable antenna.

[0244] The mapping module 20 is used to collect the actual amplitude and phase values ​​of each channel of the multi-channel electrically tunable antenna, calculate the amplitude error value and phase error value of each channel based on the preset amplitude standard value and phase standard value, and map the amplitude error value and phase error value to the antenna array diagram structure to form a real amplitude and phase error diagram signal.

[0245] The processing module 30, based on the real amplitude-phase error map signal, uses a preset graph encoder to generate a latent space posterior probability distribution characterizing the real amplitude-phase error map signal; randomly samples at least one latent variable from the latent space posterior probability distribution, decodes the latent variable using a preset graph generator to generate amplitude-phase perturbation samples; and inputs the real amplitude-phase error map signal and the amplitude-phase perturbation samples into a preset graph discriminator to calculate the authenticity discrimination value of each input signal.

[0246] The test module 40 calculates the evaluation index of the amplitude and phase consistency of the multi-channel electrically adjustable antenna based on the true / false discrimination value and the amplitude and phase disturbance sample, and determines the multi-channel amplitude and phase consistency test result of the multi-channel electrically adjustable antenna according to the evaluation index.

[0247] 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 automated testing of amplitude and phase consistency of a multi-channel electrically adjustable antenna, characterized in that, The method comprises: connecting each channel port of a to-be-tested multi-channel electrically tunable antenna to a test device; sequentially performing phase testing and S parameter testing on each channel of the antenna by the test device to obtain actual phase values and actual amplitude values of each channel respectively; comparing the obtained actual amplitude values and actual phase values with preset standard values to determine amplitude and phase errors of each channel; judging whether the amplitude and phase consistency of the antenna is qualified according to whether the amplitude and phase errors are within a preset tolerance range, and performing repair processing on the antenna when it is judged as unqualified; The method further comprises: obtaining physical position information of each channel of a multi-channel electrically tunable antenna, and constructing an antenna array graph structure according to the physical position information, each node in the antenna array graph structure corresponding to one channel of the multi-channel electrically tunable antenna; collecting actual amplitude values and actual phase values of each channel of the multi-channel electrically tunable antenna, and calculating amplitude error values and phase error values of each channel based on preset amplitude standard values and phase standard values, and mapping the amplitude error values and phase error values to the antenna array graph structure to form a real amplitude and phase error graph signal; based on the real amplitude and phase error graph signal, using a preset graph encoder to generate a latent space posterior probability distribution for representing the real amplitude and phase error graph signal; randomly sampling at least one latent variable from the latent space posterior probability distribution, and decoding the latent variable using a preset graph generator to generate an amplitude and phase perturbation sample; inputting the real amplitude and phase error graph signal and the amplitude and phase perturbation sample into a preset graph discriminator to calculate a true and false discrimination value of each input signal respectively; based on the true and false discrimination value and the amplitude and phase perturbation sample, calculating an evaluation index of the amplitude and phase consistency of the multi-channel electrically tunable antenna, and determining a multi-channel amplitude and phase consistency test result of the multi-channel electrically tunable antenna according to the evaluation index.

2. The automated test method for amplitude and phase consistency of a multi-channel electrically adjustable antenna according to claim 1, characterized in that, The to-be-tested multi-channel electrically tunable antenna is a multi-channel electrically tunable antenna semi-finished product, and the semi-finished product comprises: a metal reflector plate; a plurality of radiation dipoles, a feed network and a phase shifter assembly mounted on the metal reflector plate; wherein the phase testing and the S parameter testing are performed before the semi-finished product is installed with a radome.

3. The automated test method for amplitude and phase consistency of a multi-channel electrically adjustable antenna according to claim 1, characterized in that, The construction of the antenna array graph structure according to the physical position information comprises: obtaining physical coordinate information of each channel of the multi-channel electrically tunable antenna; calculating the Euclidean distance between any two channels based on the physical coordinate information, and converting the Euclidean distance into a corresponding electrical length according to a preset test signal center frequency; presetting a first distance threshold and a second distance threshold for distinguishing between strong coupling near-field effect and weak coupling far-field effect, wherein the first distance threshold is smaller than the second distance threshold; according to the Euclidean distance, dividing the connection relationship between any two channels into mutually exclusive first connection relationship or second connection relationship, wherein the first connection relationship corresponds to the case that the Euclidean distance is less than or equal to the first distance threshold, and the second connection relationship corresponds to the case that the Euclidean distance is greater than the first distance threshold and less than or equal to the second distance threshold.

4. The automated test method for amplitude and phase consistency of a multi-channel electrically adjustable antenna according to claim 3, characterized in that, The constructing an antenna array graph structure according to the physical position information further comprises: For each pair of channels having the first connection relationship or the second connection relationship, a normalized connection weight value of the corresponding connection relationship is calculated based on the electrical length and a preset path loss model; A weighted heterogeneous graph structure for representing multi-channel electrically steerable antenna channel multi-scale electromagnetic coupling characteristics is constructed based on the channels, the first connection relationship, the second connection relationship and the connection weight value.

5. The automated test method for amplitude and phase consistency of a multi-channel electrically adjustable antenna according to claim 4, characterized in that, The forming a real amplitude and phase error graph signal comprises: Based on a preset maximum amplitude error range, performing normalization processing on the amplitude error value of each channel to obtain a normalized amplitude error value corresponding to the channel; Performing a trigonometric function transformation on the phase error value of each channel, and mapping the phase error value to a two-dimensional Cartesian coordinate space to obtain a two-dimensional phase error vector corresponding to the channel; Concatenating the normalized amplitude error value and the two-dimensional phase error vector corresponding to each channel to generate a node initial feature vector corresponding to each channel; The node initial feature vector corresponding to each channel is taken as a node attribute, and is assigned to a corresponding node in the weighted heterogeneous graph structure to generate a real amplitude and phase error graph signal.

6. The automated test method for amplitude and phase consistency of a multi-channel electrically adjustable antenna according to claim 3, characterized in that, The generating an implicit space posterior probability distribution for representing the real amplitude and phase error graph signal comprises: Using a multi-layer graph convolution network built in the graph encoder, performing layer-by-layer information aggregation on the node initial feature vector of the real amplitude and phase error graph signal to obtain a node output feature vector of each node; Performing a residual connection operation on the node output feature vector of each node and the node initial feature vector corresponding to the node to obtain a residual feature vector including an initial error feature; Performing a global pooling operation on the residual feature vector corresponding to all nodes to obtain a graph-level context vector for the entire antenna array graph structure; Concatenating the residual feature vector corresponding to each node and the graph-level context vector to generate a fusion feature vector that fuses local node information and global graph information; Mapping the fusion feature vector corresponding to each node to an implicit space through a fully connected layer network built in the graph encoder to generate a mean parameter and a variance parameter corresponding to the implicit space posterior probability distribution.

7. The automated test method for amplitude and phase consistency of a multi-channel electrically adjustable antenna according to claim 6, characterized in that, The obtaining a node output feature vector of each node comprises: For each layer of the multi-layer graph convolution network, the following feature update operation is performed on each node of the antenna array graph structure: For the neighbor nodes having the first connection relationship and the neighbor nodes having the second connection relationship with the node respectively, a separate weighted information aggregation operation is performed based on the normalized connection weight value corresponding to each neighbor node to obtain an intermediate feature vector representing near-field effect and far-field effect respectively; Performing a linear transformation on the intermediate feature vectors representing near-field effect and far-field effect respectively, and fusing the linearly transformed feature vectors to generate an aggregated feature vector; Performing feature update based on the aggregated feature vector and the previous layer feature vector of the node itself to obtain an updated node feature vector; Performing nonlinear activation processing on the updated node feature vector to obtain a node output feature vector of the node at the current layer.

8. The automated test method for amplitude and phase consistency of a multi-channel electrically adjustable antenna according to claim 1, characterized in that, The generating of the amplitude-phase perturbation sample includes: Decomposing the latent variable sampled from the latent space posterior probability distribution into a global latent variable and a local latent variable; Processing the global latent variable by using a first neural network to generate a global bias vector; Assigning the local latent variable to each node in the weighted heterogeneous graph structure as an initial random feature of the node.

9. An automated test system for amplitude and phase consistency of a multi-channel electrically adjustable antenna, characterized in that, It includes: The construction module is used for acquiring physical position information of each channel of the multi-channel electrically adjustable antenna, and constructing an antenna array graph structure according to the physical position information, each node in the antenna array graph structure corresponding to one channel of the multi-channel electrically adjustable antenna; The mapping module is used for collecting actual amplitude values and actual phase values of each channel of the multi-channel electrically adjustable antenna, calculating amplitude error values and phase error values of each channel based on preset amplitude standard values and phase standard values, and mapping the amplitude error values and the phase error values to the antenna array graph structure to form a real amplitude-phase error graph signal; The processing module generates a latent space posterior probability distribution for representing the real amplitude-phase error graph signal by using a preset graph encoder based on the real amplitude-phase error graph signal; at least one latent variable is randomly sampled from the latent space posterior probability distribution, and the latent variable is decoded by using a preset graph generator to generate an amplitude-phase perturbation sample; The real amplitude-phase error graph signal and the amplitude-phase perturbation sample are input into a preset graph discriminator to calculate a true-false discrimination value of each input signal; The test module calculates an evaluation index of the amplitude-phase consistency of the multi-channel electrically adjustable antenna based on the true-false discrimination value and the amplitude-phase perturbation sample, and determines a multi-channel amplitude-phase consistency test result of the multi-channel electrically adjustable antenna according to the evaluation index.

Images