A method, system and storage medium for coping with antenna performance testing

Abstract

This invention relates to the field of electrical variable measurement technology, solving the technical problems of existing technologies such as lack of dynamic path planning, mechanical wear caused by frequent acceleration and deceleration of the probe, and easy omission of key features in complex multipath environments. In particular, it relates to a method, system, and storage medium for antenna performance testing. The method involves the following steps: acquiring target test frequency data and standard signal modulation data of the target antenna; generating synchronous excitation signal data through multi-dimensional signal processing and synchronous adjustment; significantly improving sampling efficiency and data quality through multi-stage collaborative optimization; initially selecting the sampling domain by combining theoretical radiation patterns and interference distribution to avoid measurement of invalid areas; using reinforcement learning to dynamically plan the path to reduce mechanical loss; reducing redundancy by refining the points; and finally ensuring data independence and coverage through weighted evaluation. It is more adaptable to dynamic channel environments and is especially suitable for high-dimensional antenna testing scenarios such as millimeter waves.

Description

Technical Field

[0001] This invention relates to the field of electrical variable measurement technology, and in particular to a method, system and storage medium for antenna performance testing. Background Technology

[0002] Measuring electrical variables refers to the process of quantifying and analyzing physical quantities that characterize energy or state in a circuit, such as voltage, current, power, and impedance. This involves measuring the input impedance of an antenna using the voltage-to-current ratio, calculating the gain using the power comparison method, detecting the standing wave ratio using an impedance plotter or standing wave meter, and determining the radiated power and calculating the efficiency using the radiant pattern integration method. Modern technology also incorporates network analyzers, field strength meters, and other equipment to accurately acquire parameters such as the antenna's radiation pattern and polarization characteristics in a far-field test field or microwave anechoic chamber, thereby comprehensively evaluating its performance.

[0003] Existing technologies are prone to wasting resources in low signal-to-noise ratio areas, lack dynamic path planning, and cause mechanical wear due to frequent acceleration and deceleration of the probe. They are also difficult to balance global coverage with local details and are prone to missing key features in complex multipath environments. For example, when testing antennas installed on fast-moving vehicles in tunnels, there are multiple sets of delayed and diffused signals, which leads to distortion of the acquired signal. Or, in millimeter-wave testing of indoor antennas, the reflection from metal filing cabinets creates a strong interference angular domain, causing inaccurate measurements. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for antenna performance testing, which solves the technical problems of lack of dynamic path planning, mechanical wear caused by frequent acceleration and deceleration of the probe, and easy omission of key features in complex multipath environments.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for testing antenna performance, the method comprising the following steps: acquiring target test frequency data and standard signal modulation data of the target antenna, and generating synchronous excitation signal data through multi-dimensional signal processing and synchronous adjustment;

[0006] Raw environmental response signals are acquired from reference probe arrays at different spatial locations of the target antenna. Multidimensional dynamic sensing and interactive processing are performed based on the raw environmental response signals and synchronous excitation signal data to generate interference source feature vectors.

[0007] The theoretical radiation pattern of the target antenna is acquired, and the optimal sampling path is planned based on the theoretical radiation pattern and the feature vector of the interference source to obtain a refined sampling point sequence and the expected sampling data quality score.

[0008] The calibration signal matrix is ​​obtained by synchronously acquiring and filtering the refined sampling point sequence;

[0009] The expected sampled data quality score and calibration signal matrix are used to perform performance inversion and uncertainty assessment, and an antenna performance parameter evaluation report is generated.

[0010] Preferably, through multi-dimensional signal processing and synchronous adjustment, the method includes: processing the target test frequency data and standard signal modulation data through an orthogonal frequency division multiplexing modulation matrix to obtain a baseband signal vector;

[0011] Mixing and automatic gain control are performed on the baseband signal vector to obtain a normalized radio frequency signal;

[0012] High-precision clock synchronization protocol processing is performed on the normalized radio frequency signal to obtain synchronization excitation signal data.

[0013] Preferably, multi-dimensional dynamic perception and interactive processing based on the original environmental response signal and synchronous excitation signal data includes: obtaining an environmental data tensor by aligning the original environmental response signal through tensor processing;

[0014] Fuzzy mean clustering is performed on environmental data tensors to obtain component cluster centers and membership matrices.

[0015] The environmental data tensor and membership matrix are subjected to adversarial interference and feature fusion processing to obtain the feature vector of the interference source.

[0016] Preferably, adversarial interference and feature fusion processing are performed on the environmental data tensor and membership matrix, including: generating adversarial processing based on the environmental data tensor and membership matrix through conditional generation to obtain an adversarial generator;

[0017] The component cluster centers, membership matrix and synchronization excitation signal data are deconvolved and path parameter tracking processed to generate dynamic environment channel response values;

[0018] Feature extraction and fusion processing based on adversarial generator and environmental data tensor are used to obtain the feature vector of the interference source.

[0019] Preferably, the optimal sampling path planning process is performed on the theoretical radiation pattern and the feature vector of the interference source, including: calculating the candidate sampling region based on the spatial spectrum avoidance region according to the theoretical radiation pattern and the feature vector of the interference source;

[0020] Spatial information entropy is calculated based on candidate sampling regions and dynamic environment channel response values ​​to obtain candidate point information entropy;

[0021] The information entropy of candidate points is refined and evaluated through sampling points to obtain the expected sampling data quality score.

[0022] Preferably, the candidate point information entropy is refined and evaluated by sampling points, including: updating the candidate point information entropy through Q-value to obtain a preliminary sampling path;

[0023] Based on the theoretical direction map and the preliminary sampling path, a refined sampling point sequence is obtained through sampling point refinement.

[0024] The refined sampling point sequence and the dynamic environment channel response value are weighted and evaluated to obtain the expected sampling data quality score.

[0025] Preferably, the process of synchronous acquisition and filtering based on the refined sampling point sequence includes: acquiring the original response signal of the sampling point through synchronous trigger acquisition processing based on the refined sampling point sequence;

[0026] Adaptive filtering notch processing is performed based on the original response signal of the sampling point, the feature vector of the interference source, and the channel response value of the dynamic environment to obtain the filtered signal;

[0027] The filtered signal is coherently averaged to obtain the calibration signal matrix.

[0028] Preferably, performance inversion and uncertainty assessment are performed on the expected sampled data quality score and calibration signal matrix, including:

[0029] A joint observation model is generated by embedding a physical model based on the calibration signal matrix and the dynamic environment channel response value.

[0030] Basis tracking denoising was performed on the joint observation model and calibration signal matrix to obtain the initial current distribution coefficients, and a high-precision calibration antenna parameter material library was constructed.

[0031] Variational autoencoders are trained based on a high-precision calibration antenna parameter library to obtain optimized encoders and optimized decoders.

[0032] The optimal antenna intrinsic parameter estimate is obtained by maximizing the posterior probability processing of the joint observation model, initial current distribution coefficient, calibration signal matrix, optimized encoder and optimized decoder;

[0033] An antenna performance parameter evaluation report is obtained by combining the optimal antenna intrinsic parameter estimate and the expected sampled data quality score through uncertainty propagation.

[0034] This technical solution also provides a system for applying the aforementioned method to test antenna performance, the system comprising:

[0035] The synchronization excitation module is used to acquire target test frequency data and standard signal modulation data from the target antenna, and generate synchronization excitation signal data through multi-dimensional signal processing and synchronization adjustment.

[0036] The multi-dimensional interaction module is used to acquire raw environmental response signals from the reference probe array at different spatial locations of the target antenna, and to perform multi-dimensional dynamic sensing and interactive processing based on the raw environmental response signals and synchronous excitation signal data to generate interference source feature vectors.

[0037] The refinement planning module is used to acquire the theoretical radiation pattern of the target antenna, perform optimal sampling path planning on the theoretical radiation pattern and the feature vector of the interference source, and obtain the refined sampling point sequence and the expected sampling data quality score.

[0038] The synchronous filtering module is used to obtain the calibration signal matrix based on the refined sampling point sequence through synchronous acquisition and filtering.

[0039] The inversion evaluation module is used to perform performance inversion and uncertainty assessment on the expected sampled data quality score and calibration signal matrix, and generate an antenna performance parameter evaluation report.

[0040] According to one aspect of some embodiments of this disclosure, a computer-readable storage medium is provided that stores computer program instructions thereon, which, when executed, implement the steps of any of the antenna performance testing methods mentioned above.

[0041] By employing the above technical solutions, the present invention provides a method, system, and storage medium for antenna performance testing, which has at least the following beneficial effects:

[0042] 1. This invention employs an orthogonal frequency division multiplexing modulation matrix generation method to generate high-quality baseband signals, accurately realize signal frequency conversion and power adjustment, ensure signal stability and reliability, provide precise time information to the signal, greatly improve signal synchronization, and fully meet the needs of complex test scenarios, providing more accurate and efficient support for antenna testing.

[0043] 2. This invention integrates multi-dimensional information by constructing a three-dimensional spatiotemporal-frequency tensor, retains the fuzzy boundaries of multipath components using fuzzy clustering, generates adversarial networks to adaptively model non-stationary noise, and dynamically extracts channel response by combining deconvolution with path tracking. Finally, it synthesizes interference feature vectors, which can unsupervisedly separate dense multipath components and overcome the limitation of traditional algorithms that require a preset number of information sources. It generates highly realistic environmental data for enhanced modeling, which together achieves high-fidelity and adaptive modeling of dynamic environmental characteristics and unknown interference, significantly improving the completeness of environmental perception and robustness to unconventional interference.

[0044] 3. This invention significantly improves sampling efficiency and data quality through multi-stage collaborative optimization. It combines theoretical radiation patterns and interference distribution to initially select the sampling domain, avoiding invalid area measurements. Information entropy quantifies the value of multipath signals, reinforcement learning dynamically plans paths to reduce mechanical losses, refines points to reduce redundancy, and finally ensures data independence and coverage through weighted evaluation. It is more adaptable to dynamic channel environments and is especially suitable for high-dimensional antenna testing scenarios such as millimeter waves.

[0045] 4. This invention explicitly integrates environmental effects through a joint observation model, avoiding parameter distortion caused by environmental interference in traditional methods. It solves the problem of inaccurate prior assumptions in traditional Bayesian methods through prior constraints. The uncertainty quantification combined with sampling quality scoring overcomes the shortcomings of existing technologies that ignore the impact of data quality on reliability. Finally, even in complex electromagnetic environments, such as multipath interference or dynamic channels, it can still output reliable parameters and error ranges, providing interpretable uncertainty quantification. Attached Figure Description

[0046] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0047] Figure 1 This is a flowchart of a method for testing antenna performance according to the present invention;

[0048] Figure 2 This is a structural block diagram of an antenna performance testing system according to the present invention. Detailed Implementation

[0049] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This will allow for a full understanding of how the present application uses technical means to solve technical problems and achieve technical effects, and to facilitate its implementation.

[0050] Example 1:

[0051] Due to the lack of dynamism in existing technology path planning, frequent acceleration and deceleration of the probe lead to mechanical wear, and key features are easily missed in complex multipath environments. Please refer to [the relevant documentation / reference]. Figure 1 This embodiment provides a method for antenna performance testing that can significantly improve sampling efficiency and data quality, avoid invalid area measurements, use reinforcement learning to dynamically plan paths to reduce mechanical losses, refine point locations to reduce redundancy, and finally ensure data independence and coverage through weighted evaluation. It is more adaptable to dynamic channel environments. The method includes the following steps:

[0052] S1. Collect target test frequency data and standard signal modulation data from the target antenna, and generate synchronous excitation signal data through multi-dimensional signal processing and synchronous adjustment. Existing multi-standard signal synthesis technology may lack flexibility, resulting in poor signal quality, insufficient control over the accuracy of signal frequency conversion and power adjustment, and easy to cause signal instability. In addition, the clock synchronization accuracy is not high enough.

[0053] To solve the above problems, the specific implementation steps are as follows:

[0054] S11. Based on the target test frequency data and standard signal modulation data, the baseband signal vector is obtained through orthogonal frequency division multiplexing modulation matrix processing. In this step, the frequency corresponding to each subcarrier is first determined according to the input target test frequency data, and relevant modulation information is obtained from the standard signal modulation data. Then, for each subcarrier, its amplitude, the frequency determined according to the target test frequency data, and the set initial phase are combined and calculated. Specifically, the product of twice pi, frequency, and time is calculated first, and then the initial phase is added. Then, exponential operation is performed with the base of the natural constant as the base. Finally, the result is multiplied by the amplitude. After that, all subcarriers are processed... The results obtained from the above calculations are summed to obtain the baseband signal vector. The target test frequency data contains all the specific frequency parameters required for antenna testing, such as the center frequency value, the start and end frequencies of the sweep, the frequency step interval, and the test bandwidth corresponding to each frequency point. The standard signal modulation data is a predefined set of parameters that specifies in detail the key generation parameters of various modulation systems, such as QPSK, 16QAM, and OFDM. These parameters include the name of the modulation type, the constellation diagram mapping relationship, the roll-off factor of the pulse shaping filter, the orthogonal subcarrier spacing, the cyclic prefix length, and the frame structure and pilot pattern, which will not be elaborated here.

[0055] S12. Mix the baseband signal vector and perform automatic gain control to obtain a normalized radio frequency signal. In this step, a mixing operation is first performed, multiplying the baseband signal by a complex exponential signal with a base of the natural constant and an exponent of twice pi multiplied by the product of the carrier center frequency and time. This step realizes the conversion of the signal from the baseband frequency to the radio frequency. Then, automatic gain control is performed. First, the average power of the baseband signal is calculated, and then the target transmit power is divided by the average power to obtain the normalized gain coefficient. Then, the signal obtained from the previous mixing is multiplied by this normalized gain coefficient. Through this series of operations, the normalized radio frequency signal is finally obtained. This signal not only completes the up-conversion of the frequency, but also realizes the power normalization processing, so that it meets the predetermined transmit power requirements.

[0056] S13. High-precision clock synchronization protocol processing based on normalized radio frequency signals to obtain synchronization excitation signal data. In this step, the synchronization start time is first determined by using the high-precision clock synchronization protocol based on BeiDou, and the period of each signal frame is defined. For each signal frame, the synchronization start time is used as the reference, plus the product of the frame number and the frame period, plus the residual time difference obtained after compensation by the clock synchronization protocol, where the absolute value of the residual time difference is less than the maximum synchronization error allowed by the protocol, thereby calculating the absolute timestamp corresponding to each signal frame. Finally, the normalized radio frequency signal is bound one by one with the calculated high-precision absolute timestamp sequence to obtain the synchronization excitation signal data. This data realizes the fusion of signal and precise time information, providing a foundation for subsequent precise processing and analysis.

[0057] This invention employs an orthogonal frequency division multiplexing modulation matrix generation method to generate high-quality baseband signals, accurately realize signal frequency conversion and power adjustment, ensure signal stability and reliability, provide precise time information to the signal, greatly improve signal synchronization, and fully meet the needs of complex test scenarios, providing more accurate and efficient support for antenna testing.

[0058] S2. Collect raw environmental response signals from reference probe arrays at different spatial locations of the target antenna. Perform multi-dimensional dynamic sensing and interactive processing based on raw environmental response signals and synchronous excitation signal data to generate interference source feature vectors. Existing technologies are difficult to fully capture features such as multipath effects and dynamic interference. They rely on fixed thresholds or hard clustering, are prone to losing fuzzy boundary components, are difficult to simulate non-stationary interference, have low accuracy in channel response extraction due to noise, and lack adaptive capabilities. For example, in UAV antenna communication scenarios, the rapid movement of interference sources leads to non-stationary noise statistical characteristics. Existing technologies cannot match the actual interference distribution and are prone to dynamic interference failure.

[0059] To solve the above problems, the specific implementation steps are as follows:

[0060] S21. Based on the original environmental response signal, an environmental data tensor is obtained through alignment tensor processing. In this step, multi-probe data alignment is first performed. Based on the previously acquired synchronization time information, the signals collected by different probes are accurately aligned in the time dimension to ensure that the data of each probe at the same time can be correctly matched. Then, tensor processing is performed. According to the three dimensions of probe spatial position, time sampling point, and pre-set test frequency, the corresponding data is extracted from the original environmental response signal. Specifically, the probe spatial position is used as the first dimension index, the synchronization time sampling point is used as the second dimension index, and the test frequency is used as the third dimension index. The corresponding data is selected from the original signal and combined. After this series of processing, the environmental data tensor is finally obtained. This tensor completely retains the information of the three dimensions of space, time, and frequency, which can provide comprehensive and reliable data support for the subsequent accurate evaluation of the antenna performance in complex environments.

[0061] S22. Based on the environmental data tensor, perform fuzzy mean clustering to obtain component cluster centers and membership matrices. In this step, the fuzzy C-means clustering algorithm is used for multipath component separation. Specifically, first, the estimated number of multipath components is set as the number of cluster centers, and the fuzzy weight index is determined. Then, for each data point in the environmental data tensor, its membership degree to each cluster center is calculated. The calculation process is as follows: calculate the square of the difference between each data point and the feature vector of each cluster center, and then multiply the fuzzy weight index of the membership degree by the square of this difference. The fuzzy weight index is generally set to 2, but it can be slightly adjusted according to the actual situation, such as 2.5. Then, sum the calculation results of all data points for each cluster center to obtain the objective function. By continuously iterating and adjusting the membership degree and cluster centers, the value of the objective function is minimized. The number of iterations can be set to 100. Finally, the multipath component cluster centers representing the characteristics of each multipath component and the membership matrix that reflects the degree of belonging of each data point to different multipath components are obtained, providing key data for accurately analyzing the impact of multipath effects on antenna performance.

[0062] S23. Perform anti-interference and feature fusion processing on the environmental data tensor and membership matrix to obtain the feature vector of the interference source.

[0063] S231. Based on the environmental data tensor and membership matrix, an adversarial generator is obtained through conditional generative adversarial processing. In this step, a conditional generative adversarial network is used to model environmental noise and interference. Specifically, a generator and a discriminator are constructed. The generator takes random noise and cluster centers extracted from the membership matrix as conditional inputs and attempts to generate simulated environmental noise and interference data through complex internal neural network operations. The discriminator also uses cluster centers as conditions, receives real environmental data and data generated by the generator, and determines whether the received data is real or generated data. During training, the generator and discriminator compete against each other, continuously adjusting parameters to optimize their performance. The generator strives to make the generated data more closely resemble real data in order to fool the discriminator, while the discriminator strives to improve its ability to distinguish between real and generated data. This process is based on a specific adversarial loss calculation method, which first calculates the logarithmic sum of the probability that the discriminator judges the real data as true, and then adds the logarithmic sum of the probability that the discriminator judges the generated data as false. Optimization is achieved by minimizing the generator's loss and maximizing the discriminator's loss. After multiple rounds of iterative training, for example, 80 iterations, a well-trained generator is obtained. This generator can simulate background noise and interference data that conform to the characteristics of the actual environment based on different cluster center conditions, providing interference simulations that are closer to real scenarios for subsequent antenna performance testing.

[0064] S232. Perform deconvolution and path parameter tracking on the component cluster centers, membership matrix, and synchronization excitation signal data to generate dynamic environment channel response values. In this step, the deconvolution and path parameter tracking method is used to extract the dynamic channel impulse response. The specific steps are as follows: First, based on the component cluster centers and membership matrix, perform dynamic tracking on each cluster to obtain the changes in path complex gain, delay, and phase over time.

[0065] Specifically, path complex gain reflects the intensity variation of the signal along different paths, time delay reflects the time difference required for the signal to travel from transmission to reception along different paths, and phase represents the phase shift of the signal during propagation. Then, using the synchronous excitation signal data, the excitation signal is combined with the tracked path parameters through deconvolution. In this process, for each cluster, the path complex gain, the unit impulse function offset by time delay, and the phase rotation factor are multiplied together, and the results of all clusters are summed to finally obtain the dynamic environment channel response value. It fully describes the response characteristics of the channel in the dynamic environment where the antenna is located at different times and with different time delays, providing a key basis for accurately evaluating the performance of the antenna in complex dynamic environments.

[0066] S233. Feature extraction and fusion processing based on adversarial generator and environmental data tensor to obtain interference source feature vector. In this step, a series of typical interference samples are first generated using adversarial generator. These samples simulate various interference situations that may occur in the actual environment. At the same time, the environmental data tensor is analyzed in detail to find regions with abnormal energy. These regions often hide key interference information. Then, key features are extracted from the generated typical interference samples and abnormal energy regions in the environmental data tensor. Specifically, the center frequency is determined by analyzing the spectral distribution of the signal, the bandwidth is obtained by calculating the width occupied by the signal in the frequency domain, the average power of the signal is obtained by averaging the power of the signal over a period of time, and the modulation type is determined based on the waveform characteristics and variation law of the signal. Finally, the four key features extracted—center frequency, bandwidth, average power, and modulation type identifier—are combined in a certain order to form interference source feature vector. This feature vector can comprehensively and accurately describe the characteristics of the dominant interference, providing an important basis for the identification and suppression of interference in subsequent antenna performance testing.

[0067] This invention integrates multi-dimensional information by constructing a three-dimensional spatiotemporal-frequency tensor, preserves the fuzzy boundaries of multipath components using fuzzy clustering, generates adversarial networks to adaptively model non-stationary noise, and dynamically extracts channel response by combining deconvolution with path tracking. Finally, it synthesizes interference feature vectors, which can unsupervisedly separate dense multipath components and overcome the limitation of traditional algorithms that require a preset number of information sources. It generates highly realistic environmental data for enhanced modeling, which together achieves high-fidelity and adaptive modeling of dynamic environmental characteristics and unknown interference, significantly improving the completeness of environmental perception and robustness to unconventional interference.

[0068] S3. Acquire the theoretical radiation pattern of the target antenna, perform optimal sampling path planning on the theoretical radiation pattern and interference source feature vector to obtain a refined sampling point sequence and expected sampling data quality score; Existing technologies are prone to wasting resources in low signal-to-noise ratio areas, lack dynamic path planning, and the frequent acceleration and deceleration of the probe leads to mechanical wear. It is difficult to balance global coverage and local details, and key features are easily missed in complex multipath environments. For example, when testing an antenna installed on a fast-moving vehicle in a tunnel, there are multiple sets of delayed and diffused signals, which leads to distortion of the acquired signal. Or, in the millimeter-wave test of an indoor antenna, the reflection from a metal filing cabinet forms a strong interference angular domain, causing inaccurate measurements.

[0069] To solve the above problems, the specific steps are as follows:

[0070] S31. Based on the theoretical radiation pattern and the interference source feature vector, the candidate sampling region is obtained through spatial spectrum avoidance region calculation. In this step, the theoretical radiation pattern of the antenna under test is first obtained through theoretical calculation or simulation software. This radiation pattern reflects the radiation characteristics of the antenna at different spatial angles, such as the radiation intensity in a certain direction. Then, combined with the input interference source feature vector, which contains key information such as the center frequency, bandwidth, average power and modulation type of the interference, the intensity distribution of the interference in different spatial directions is deduced.

[0071] Then, spatial spectrum analysis and avoidance region calculation are performed. Specifically, the theoretical radiation intensity of the target antenna in a certain direction is divided by the interference intensity in that direction. If the calculation result is less than the preset signal-to-interference-plus-noise ratio (SIR) threshold (in practice, the SIR threshold is set to 3dB, with a threshold range of -3dB to 20dB), then that direction is determined to be a corner region that needs to be avoided. Finally, these corner regions that need to be avoided are removed from the total sampling sphere, and the remaining area is the candidate sampling region. This candidate sampling region can avoid areas with strong interference, providing a more accurate and less interference-prone sampling space for subsequent antenna performance testing. In practice, the simulation software used is HFSS, a commonly used frequency domain analysis software based on the finite element method, which will not be elaborated here.

[0072] S32. Calculate the spatial information entropy based on the candidate sampling region and the dynamic environment channel response value to obtain the information entropy of the candidate points. In this step, for each sampling point within the candidate sampling region, predict the posterior probability of the received signal belonging to each multipath component based on the dynamic environment channel response value. Then, take the base-2 logarithm of the posterior probability corresponding to each multipath component, multiply it by the posterior probability, and take the negative number. Finally, sum the calculation results corresponding to all multipath components. The result is the information entropy of the sampling point. Calculate this for all sampling points within the candidate sampling region in this way to obtain the information entropy corresponding to each sampling point within the candidate region. This set of information entropies reflects the richness of signal information contained in each sampling point. The larger the information entropy, the more complex the signal components received by the sampling point and the more information it contains, providing a basis for selecting more representative sampling points in the future.

[0073] S33. Refine and evaluate the information entropy of candidate points to obtain the expected sampling data quality score;

[0074] S331. Based on the entropy of candidate points, the Q-value is updated to obtain the initial sampling path. In this step, a deep Q-network reinforcement learning method is used to generate a dynamic path sequence. First, the probe movement is used for constraint, mainly constraining speed and acceleration. Usually, it is determined comprehensively based on the physical characteristics of the probe in the actual test scenario, such as mechanical structure, driving capability and test efficiency requirements. For example, if the probe is a mechanical scanning antenna test turntable, its speed constraint needs to consider the upper limit of motor speed, and the acceleration constraint needs to avoid measurement jitter caused by inertia.

[0075] In specific processing, the current probe position and the set of sampled points together constitute the state. The selection of the next sampling point is taken as the action. The reward value is calculated based on the information entropy of the candidate points. Points with high information entropy are selected first to obtain more signal features. At the same time, the movement cost, such as the time cost obtained by dividing the distance between two points by the probe speed, is combined for a trade-off. The decision is iteratively optimized through the Q-value update formula: the current Q-value plus the learning rate multiplied by the difference between the current reward and the discount of the future maximum Q-value minus the current Q-value. The future maximum Q-value is obtained by taking the maximum value of the Q-value of all possible actions. After multiple rounds of training, the algorithm generates an initial sampling path that takes into account both information acquisition efficiency and movement constraints. This path prioritizes the coverage of key areas based on information entropy, while also meeting the physical movement constraints of the probe, providing an efficient sampling sequence for subsequent accurate testing. The Q-value update formula is a method commonly used in deep Q-networks to iteratively optimize the value evaluation of agents for different state-action combinations, which will not be elaborated here.

[0076] S332. Based on the theoretical radiation pattern and the initial sampling path, a refined sampling point sequence is obtained through sampling point refinement. In this step, the sampling point positions in the initial path and their corresponding antenna radiation pattern observations are used as the initial dataset. The spatial distribution law of the theoretical radiation pattern is fitted by a Gaussian process model. This model characterizes the continuity and fluctuation of the function value by calculating the covariance matrix of the spatial distance between each point and the difference in the radiation pattern. On this basis, the expected improvement value of each candidate point is calculated for the unsampled area. The predicted radiation pattern value of the point is subtracted from the currently observed best value. If the result is positive, the difference is retained; otherwise, it is zero.

[0077] Subsequently, the expected value of the difference is calculated for all possible posterior distribution samples. Finally, the point with the largest expected improvement value is selected as the new sampling point. By iteratively executing the loop of model fitting, expectation calculation, and point selection, high-information-value points are gradually added to the sampling sequence, and finally, a refined sampling point sequence is output. This sequence effectively reduces redundant sampling while maintaining coverage of key features of the antenna pattern. For example, dense sampling is performed in the main lobe region to capture details, and sparse sampling is performed in the side lobe region to reduce testing costs, thereby achieving an optimal balance between testing accuracy and efficiency. Among them, the Gaussian process model is a non-parametric regression method based on statistics. It assumes that the value of the antenna theoretical pattern at any point in space follows a joint Gaussian distribution. Using the known sampling point positions and pattern observations, a probabilistic model that can predict the pattern values ​​of unsampled areas is constructed. This will not be elaborated here.

[0078] S333. The refined sampling point sequence and the dynamic environment channel response value are weighted and evaluated to obtain the expected sampling data quality score. In this step, the average intensity of the channel response value corresponding to all sampling points is first calculated by summing the response values ​​of each point and dividing by the total number of sampling points. At the same time, the correlation between the channel responses of each point is analyzed, and the average value of the correlation of all points is taken as the overall redundancy index. Then, the average intensity and the redundancy index are weighted and combined. The higher the average intensity, the more fully the signal energy is captured. The lower the redundancy, the more independent information each sampling point carries. The two are assigned different weights and then added together. The weight coefficient can be obtained by the analytic hierarchy process. Finally, the expected data quality score is obtained. This score comprehensively reflects whether the sampling data can effectively avoid redundant measurements caused by point overlap while covering the key information of the antenna radiation characteristics. It provides a quantitative basis for the final confirmation of the subsequent test path.

[0079] This invention significantly improves sampling efficiency and data quality through multi-stage collaborative optimization. It combines theoretical radiation patterns and interference distribution to initially select the sampling domain, avoiding measurements in invalid areas. Information entropy quantifies the value of multipath signals, and reinforcement learning dynamically plans paths to reduce mechanical losses. It also refines points to reduce redundancy and finally ensures data independence and coverage through weighted evaluation. This invention is more adaptable to dynamic channel environments and is especially suitable for high-dimensional antenna testing scenarios such as millimeter waves.

[0080] S4. Based on the refined sampling point sequence, the calibration signal matrix is ​​obtained through synchronous acquisition and filtering. Existing technologies mostly adopt open-loop positioning and single-trigger acquisition, which are easily affected by mechanical vibration, environmental clutter and random noise, resulting in low signal-to-noise ratio, distorted radiation pattern, lack of dynamic interference suppression and multi-cycle signal enhancement methods, weak signals are easily submerged by noise, and the test efficiency and accuracy are low.

[0081] To solve the above problems, the specific implementation steps are as follows:

[0082] S41. Based on the refined sampling point sequence, the original response signals of the sampling points are acquired through synchronous triggering and processing. In this step, the three-dimensional spatial coordinates of each sampling point are first converted into motion commands for the mechanical probe. The probe is then driven to the target position by a servo motor. During the movement, the position deviation is fed back in real time by an encoder, and the motor speed is dynamically adjusted by a proportional-integral-derivative controller to ensure that the error between the actual position of the probe and the target position is less than a preset threshold. In practice, the preset threshold is set to 0.2 degrees. When the probe reaches the designated point, the system immediately sends a hardware trigger pulse to the signal acquisition device to synchronously start the reception and recording of the antenna transmission signal. The acquired raw response signal contains amplitude and phase information, and its duration is determined according to the preset channel delay spread range. The final data is a set of raw channel responses arranged in the order of the refined sampling point sequence. Each data point corresponds to the complete channel impulse response at a specific spatial location, providing basic measurement data for subsequent radiation pattern reconstruction and multipath analysis. The encoder used is the Bocheng BCE108K36 absolute encoder, and the proportional-integral-derivative controller is a commonly used PID controller, which is a commonly used tool for comparing the desired position with the actual position fed back by the encoder, calculating the control signal, and dynamically adjusting the motor speed. It will not be elaborated here.

[0083] S42. Adaptive filtering and notch filtering are performed based on the original response signal of the sampling point, the feature vector of the interference source, and the dynamic environment channel response value to obtain the filtered signal. In this step, strong multipath clutter components unrelated to antenna testing are first extracted from the dynamic environment channel response. A clutter reference signal is generated by time delay alignment and amplitude weighting. Then, the superposition result of all clutter components is subtracted from the original response signal to achieve preliminary suppression of multipath interference. At the same time, the periodic interference in a specific frequency band is located according to the feature vector of the interference source. A digital notch filter is designed to attenuate the signal in this frequency band to avoid its interference with the antenna pattern reconstruction. The final filtered signal is a pure channel response after removing environmental coupling clutter and fixed frequency band interference. Its time domain waveform is smoother and the energy of the interference frequency band in the spectrum is significantly reduced, providing a reliable data foundation for subsequent high-precision pattern extraction and channel parameter estimation.

[0084] S43. Perform coherent averaging on the filtered signal to obtain the calibration signal matrix. In this step, the start time of each excitation is marked based on the synchronous trigger clock. The response signals of the same spatial location in multiple independent tests are precisely aligned along the time axis. Then, the arithmetic mean of all aligned signal segments is performed point by point. That is, for each time sampling point, the signal amplitude values ​​of all corresponding positions of all periods are added and then divided by the number of tests. In the final calibration signal matrix, each element represents the enhanced response value of a specific spatial location at a specific time. This method enhances the effective components by superimposing coherent signals and suppresses random noise, which can significantly improve the detection capability of weak signals. It is especially suitable for low-power antenna testing or far-field attenuation scenarios. The final output calibration signal matrix provides high signal-to-noise ratio and low distortion basic data for subsequent pattern reconstruction.

[0085] This invention achieves submicron-level positioning accuracy through closed-loop servo control, ensures signal spatiotemporal alignment through synchronous triggering, dynamically suppresses environmental clutter and fixed-frequency interference using adaptive filtering and notch filtering techniques, and improves the signal-to-noise ratio with time-domain coherent averaging, significantly enhancing the detection capability of weak signals. It is especially suitable for far-field attenuation testing or high-interference scenarios, and the test results are more stable and reliable.

[0086] S5. Perform performance inversion and uncertainty assessment on the expected sampled data quality score and calibration signal matrix, and generate an antenna performance parameter assessment report; Existing technologies have difficulty handling dynamic and complex multipath, resulting in large calibration errors. The parameter inversion model is simple and does not fully consider the residual effects of the environment and the complex relationships between parameters. The results are not accurate and lack reliable credibility evaluation. For example, in satellite communication, ionospheric scintillation causes random fluctuations in the amplitude of the received signal, and in UAV antenna testing, flight jitter causes uneven distribution of sampling points, making it difficult to dynamically correct the uncertainty.

[0087] To solve the above problems, the specific implementation steps are as follows:

[0088] S51. Based on the calibration signal matrix and the dynamic environment channel response value, a physical model is embedded to generate a joint observation model. In this step, the calibrated signal matrix and the dynamic environment channel response are first deeply fused. Specifically, a physical model of the antenna radiation characteristics is established, and the mathematical relationship between the antenna current distribution and far-field radiation is transformed into a forward calculation operator. This operator is then spatially convolved with the dynamic environment channel response to simulate the superposition effect of the antenna signal after multipath propagation in the environment during actual testing. Finally, a residual noise term is superimposed on the convolution result to fully characterize the entire physical process from the antenna's intrinsic radiation to environmental coupling and then to the received signal. The final result is... The joint observation model uses mathematical mapping to uniformly describe the complex nonlinear relationship between antenna parameters, environmental characteristics, and measurement signals, providing interpretable physical constraints for subsequent parameter inversion. The physical model is based on Maxwell's equations, and its core structure includes a current distribution description layer, a radiation integral operator layer, and a propagation medium parameter layer. It can correlate the antenna current distribution with the spatial radiation field to form a positive mapping relationship. The residual noise term is random interference that is not completely suppressed by environmental channel modeling or signal processing in actual measurements. Its sources include sensor thermal noise, unmodeled weak multipath scattering, and background noise of system electronic devices, which will not be elaborated here.

[0089] S52. Basis tracking denoising is performed on the joint observation model and calibration signal matrix to obtain the initial current distribution coefficients. In this step, the joint observation model is first discretized into a numerically computable observation matrix. This matrix describes the mapping relationship between the antenna current distribution and the measurement signal. Based on the physical characteristic that the antenna current has a sparse distribution under a specific basis, such as the characteristic mode basis, the basis tracking denoising method is adopted. By minimizing the sum of the absolute values ​​of the current sparse coefficients, i.e., the first norm, and simultaneously constraining the sum of the squared errors between the inverted signal and the measured calibration signal to be less than the noise tolerance threshold, the residual noise interference is suppressed while ensuring the sparsity of the solution. The final initial current distribution coefficients are a sparse representation of the equivalent current on the antenna surface or in the bulk domain. Its non-zero elements correspond to the radiation-dominant region, providing sparse prior information for subsequent high-precision current reconstruction. It is especially suitable for the rapid inversion of the current distribution of complex structure antennas in multipath environments.

[0090] S53. Based on a high-precision calibrated antenna parameter material library, a variational autoencoder is trained to obtain an optimized encoder and an optimized decoder. This step first constructs a historical high-precision calibrated antenna parameter material library. The construction process involves performing full-parameter scan tests on multiple typical antennas in a standard testing environment, such as a microwave anechoic chamber, recording key parameters such as radiation efficiency, radiation pattern characteristics, and impedance matching under different frequency bands and excitation conditions. This forms a material library containing thousands of sets of high-confidence data. Subsequently, a variational autoencoder architecture is adopted. The encoder part maps the input antenna parameters to a probability distribution in the latent space, for example, by calculating the relative entropy balance between the input parameters and the standard normal distribution. Due to differences in parameter distribution, the decoder reconstructs the original parameters from the latent variables. For example, it evaluates the reconstruction quality using the log-likelihood function of the reconstructed parameters and the original parameters. During training, the joint loss function is obtained by minimizing the relative entropy and maximizing the reconstruction quality, thus optimizing the network parameters. The resulting optimized encoder can compress any antenna parameters into low-dimensional latent variables. The optimized decoder can generate parameter predictions that conform to the distribution of historical data based on the latent variables, providing prior probability constraints for antenna parameters in subsequent inversion. This is especially suitable for fast parameter estimation of complex antennas under small sample conditions. The joint loss function is a commonly used loss function calculation formula in variational autoencoders (VAEs), which will not be elaborated here.

[0091] S54. Maximize the posterior probability processing of the joint observation model, initial current distribution coefficients, calibration signal matrix, optimization encoder, and optimization decoder to obtain the optimal antenna intrinsic parameter estimate. In this step, the likelihood relationship between the measured signal and the antenna parameters is first constructed based on the joint observation model. The sum of squared errors between the calibration signal matrix and the model prediction signal is used as the likelihood term, reflecting the degree of matching between the parameter estimate and the measured data. At the same time, the trained variational autoencoder provides the parameter prior distribution. The initial current distribution coefficients are mapped to latent variables by the optimization encoder. Then, the optimization decoder generates the parameter prior probability density that conforms to the statistical law of historical data, which is the prior term. Finally, the maximum posterior probability criterion is adopted. Specifically, the product of the likelihood term and the prior term is used as the optimization objective. The antenna parameters are iteratively adjusted to maximize this product. This ensures that the inversion result is highly consistent with the measured data and that the parameters conform to the physical prior constraints. The final optimal antenna intrinsic parameter estimate integrates the measurement data and prior knowledge, which significantly improves the inversion accuracy of key parameters such as antenna radiation efficiency and radiation pattern in complex electromagnetic environments. It is especially suitable for antenna performance evaluation in small sample or high noise scenarios.

[0092] S55. Based on the optimal antenna intrinsic parameter estimation and the expected sampling data quality score, an antenna performance parameter evaluation report is obtained through uncertainty propagation synthesis. In this step, the posterior variance of each performance parameter is first extracted from the covariance matrix obtained by Bayesian inversion. This variance reflects the statistical fluctuation range of the parameter estimation. Then, combined with the expected sampling data quality score, the posterior variance is dynamically corrected. The square root of the posterior variance of each parameter is used as the basic uncertainty, and then multiplied by the reciprocal of the data quality score as the adjustment coefficient. If the data quality score is low, such as the presence of signal interference or insufficient sampling points, the adjustment coefficient is increased, and the final uncertainty value increases accordingly.

[0093] For example, the uncertainty of the gain parameter is calculated by first taking the square root of the variance element corresponding to the gain in the covariance matrix, and then weighting it according to the data quality score. If the score is low, the uncertainty is significantly amplified. The final antenna performance parameter evaluation report includes the estimated values ​​and quantified uncertainties of various parameters, including gain, sidelobe level, beamwidth, and other performance data. The reliability range of the parameters is clearly marked. For example, if the gain is 8.5 dB, the uncertainty is ±0.3 dB. It takes into account the data quality correction and provides a quantifiable basis for judging whether the antenna performance meets the standard. It is especially suitable for the verification of test results and engineering decisions in complex electromagnetic environments. Bayesian inversion is a commonly used parameter estimation method based on probability statistics. It combines the likelihood information of the observed data with the prior distribution of the parameters and uses Bayes' theorem to calculate the posterior probability distribution of the parameters, thereby achieving more robust parameter inference under the influence of uncertainty and noise.

[0094] This invention explicitly integrates environmental effects through a joint observation model, avoiding parameter distortion caused by environmental interference in traditional methods. It addresses the problem of inaccurate prior assumptions in traditional Bayesian methods through prior constraints. By combining uncertainty quantification with sampling quality scoring, it overcomes the shortcomings of existing technologies that ignore the impact of data quality on reliability. Ultimately, even in complex electromagnetic environments, such as multipath interference or dynamic channels, it can still output reliable parameters and error ranges, providing interpretable uncertainty quantification.

[0095] Example 2:

[0096] Due to the lack of dynamism in existing technology path planning, frequent acceleration and deceleration of the probe lead to mechanical wear, and critical features are easily missed in complex multipath environments. Please refer to [link / reference needed]. Figure 2 The diagram shown is a structural block diagram of an antenna performance testing system provided in this embodiment. The system includes a synchronous excitation module, a multi-dimensional interaction module, a refined planning module, a synchronous filtering module, and an inversion evaluation module.

[0097] The synchronization excitation module is used to acquire target test frequency data and standard signal modulation data from the target antenna, and generate synchronization excitation signal data through multi-dimensional signal processing and synchronization adjustment.

[0098] The multi-dimensional interaction module is used to acquire raw environmental response signals from the reference probe array at different spatial locations of the target antenna, and to perform multi-dimensional dynamic sensing and interactive processing based on the raw environmental response signals and synchronous excitation signal data to generate interference source feature vectors.

[0099] The refinement planning module is used to acquire the theoretical radiation pattern of the target antenna, perform optimal sampling path planning on the theoretical radiation pattern and the feature vector of the interference source, and obtain the refined sampling point sequence and the expected sampling data quality score.

[0100] The synchronous filtering module is used to obtain the calibration signal matrix based on the refined sampling point sequence through synchronous acquisition and filtering.

[0101] The inversion evaluation module is used to perform performance inversion and uncertainty assessment on the expected sampled data quality score and calibration signal matrix, and generate an antenna performance parameter evaluation report.

[0102] Example 3:

[0103] Because existing technology path planning lacks dynamism, frequent acceleration and deceleration of the probe leads to mechanical wear, and key features are easily missed in complex multipath environments, this embodiment also provides a computer-readable storage medium storing computer program instructions, which, when executed, implement any of the steps mentioned above for antenna performance testing methods.

[0104] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code, including but not limited to disk storage, CD-ROM, optical storage, etc.

[0105] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method of coping with an antenna performance test, characterized by, The steps of this method are as follows: acquire target test frequency data and standard signal modulation data of the target antenna, and generate synchronous excitation signal data through multi-dimensional signal processing and synchronous adjustment; Raw environmental response signals are acquired from reference probe arrays at different spatial locations of the target antenna. Multidimensional dynamic sensing and interactive processing are performed based on the raw environmental response signals and synchronous excitation signal data to generate interference source feature vectors. Among them, multi-dimensional dynamic perception and interactive processing based on the original environmental response signal and synchronous excitation signal data includes: obtaining the environmental data tensor by aligning the tensor according to the original environmental response signal; Fuzzy mean clustering is performed on environmental data tensors to obtain component cluster centers and membership matrices. The environmental data tensor and membership matrix are subjected to anti-interference and feature fusion processing to obtain the feature vector of the interference source. The theoretical radiation pattern of the target antenna is acquired, and the optimal sampling path is planned based on the theoretical radiation pattern and the feature vector of the interference source to obtain a refined sampling point sequence and the expected sampling data quality score. The calibration signal matrix is ​​obtained by synchronously acquiring and filtering the refined sampling point sequence; The performance inversion and uncertainty assessment are performed on the expected sampled data quality score and calibration signal matrix to generate an antenna performance parameter assessment report; Among them, the performance inversion and uncertainty assessment of the expected sampled data quality score and calibration signal matrix include: embedding a physical model based on the calibration signal matrix and the dynamic environment channel response value to generate a joint observation model; Basis tracking denoising was performed on the joint observation model and calibration signal matrix to obtain the initial current distribution coefficients, and a high-precision calibration antenna parameter material library was constructed. Variational autoencoders are trained based on a high-precision calibration antenna parameter library to obtain optimized encoders and optimized decoders. The optimal antenna intrinsic parameter estimate is obtained by maximizing the posterior probability processing of the joint observation model, initial current distribution coefficient, calibration signal matrix, optimized encoder and optimized decoder; An antenna performance parameter evaluation report is obtained by combining the optimal antenna intrinsic parameter estimate and the expected sampled data quality score through uncertainty propagation.

2. The method of claim 1, wherein, Through multi-dimensional signal processing and synchronous adjustment, including: processing the target test frequency data and standard signal modulation data through an orthogonal frequency division multiplexing modulation matrix to obtain the baseband signal vector; Mixing and automatic gain control are performed on the baseband signal vector to obtain a normalized radio frequency signal; High-precision clock synchronization protocol processing is performed on the normalized radio frequency signal to obtain synchronization excitation signal data.

3. The method of claim 1, wherein, Adversarial interference and feature fusion processing are performed on environmental data tensors and membership matrices, including: generating adversarial generators based on environmental data tensors and membership matrices through conditional adversarial processing. The component cluster centers, membership matrix and synchronization excitation signal data are deconvolved and path parameter tracking processed to generate dynamic environment channel response values; Feature extraction and fusion processing based on adversarial generator and environmental data tensor are used to obtain the feature vector of the interference source.

4. The method of claim 1, wherein, The optimal sampling path planning process is performed on the theoretical radiation pattern and the feature vector of the interference source, including: calculating the candidate sampling region based on the spatial spectrum avoidance region according to the theoretical radiation pattern and the feature vector of the interference source; Spatial information entropy is calculated based on candidate sampling regions and dynamic environment channel response values ​​to obtain candidate point information entropy; The information entropy of candidate points is refined and evaluated through sampling points to obtain the expected sampling data quality score.

5. The method of claim 4, wherein, The candidate point information entropy is refined and evaluated through sampling point processing, including: updating the candidate point information entropy using Q-value to obtain a preliminary sampling path; Based on the theoretical direction map and the preliminary sampling path, a refined sampling point sequence is obtained through sampling point refinement. The refined sampling point sequence and the dynamic environment channel response value are weighted and evaluated to obtain the expected sampling data quality score.

6. The method of claim 1, wherein, Based on the refined sampling point sequence, synchronous acquisition and filtering are performed, including: acquiring the original response signal of the sampling point through synchronous triggering acquisition and processing based on the refined sampling point sequence; Adaptive filtering notch processing is performed based on the original response signal of the sampling point, the feature vector of the interference source, and the channel response value of the dynamic environment to obtain the filtered signal; The filtered signal is coherently averaged to obtain the calibration signal matrix.

7. A system for use in a method of testing an antenna according to any one of claims 1 to 6, wherein The system includes: The synchronization excitation module is used to acquire target test frequency data and standard signal modulation data from the target antenna, and generate synchronization excitation signal data through multi-dimensional signal processing and synchronization adjustment. The multi-dimensional interaction module is used to acquire raw environmental response signals from the reference probe array at different spatial locations of the target antenna, and to perform multi-dimensional dynamic sensing and interactive processing based on the raw environmental response signals and synchronous excitation signal data to generate interference source feature vectors. The refinement planning module is used to acquire the theoretical radiation pattern of the target antenna, perform optimal sampling path planning on the theoretical radiation pattern and the feature vector of the interference source, and obtain the refined sampling point sequence and the expected sampling data quality score. The synchronous filtering module is used to obtain the calibration signal matrix based on the refined sampling point sequence through synchronous acquisition and filtering. The inversion evaluation module is used to perform performance inversion and uncertainty assessment on the expected sampled data quality score and calibration signal matrix, and generate an antenna performance parameter evaluation report.

8. A computer-readable storage medium having stored thereon computer program instructions that, when executed, implement the steps of the method for testing antenna performance as described in any one of claims 1 to 6.

Images