Automatic test system and method for high-frequency switch matrix

Abstract

The invention discloses an automatic test system and method for a high-frequency switch matrix, and relates to the technical field of signal processing, and the method comprises the steps: calculating a confidence coefficient spectrum based on a frequency response compensation curve, adjusting a compression sampling rate, and constructing a random sampling mode; according to a random sampling mode, full-band response characteristic verification is carried out on the space-time-environment baseline data, and abnormal frequency points are identified; carrying out point-by-point comparative analysis on abnormal frequency points and historical reference data, calculating a mean square error and a maximum deviation, generating an error frequency response curve and a test report through a piecewise polynomial fitting method, and constructing a random sampling mode through adjusting a compression sampling rate to realize an adjustment mechanism aiming at different frequency band characteristic changes; and equally dividing and calculating a confidence coefficient weight factor based on the confidence coefficient spectrum so as to generate a sub-band sampling scheme with a weight mark, so that high-efficiency and high-quality automatic batch testing is realized.

Description

Technical Field

[0001] The present invention relates to an electrical performance testing apparatus, and more specifically to the field of signal processing technology, particularly to an automated testing system and method for a high-frequency switching matrix. Background Technology

[0002] Automated testing technology for high-frequency switching matrices is a key technology in wireless communication and radar systems, and also falls under the category of electrical performance testing devices. Its core lies in achieving high-precision measurement and analysis of multi-channel RF signals. Currently, the field primarily employs traditional frequency sweep testing methods based on vector network analyzers, acquiring the S-parameter matrix through point-by-point scanning and using environmental temperature compensation algorithms for error correction. The test architecture generally adopts a modular design concept, integrating the switching matrix, signal source, and analysis instrument via a bus, and utilizing standard calibration components for error correction. The test data processing stage typically uses the least squares method to fit the frequency response curve and identifies abnormal frequency points based on a fixed threshold strategy.

[0003] Existing testing methods have two limitations in addressing the dynamic testing requirements under complex electromagnetic environments: First, the spatiotemporal correlation analysis between environmental parameters and test data is insufficient, and traditional temperature compensation models cannot accurately reflect the coupling effects of multiple physical fields, leading to a decrease in testing accuracy at high frequencies; Second, the balance between testing efficiency and accuracy lacks a dynamic adjustment mechanism, and fixed sampling rate strategies cannot adapt to changes in characteristics across different frequency bands, resulting in significant efficiency bottlenecks in batch testing scenarios. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an automated testing method for high-frequency switching matrices to solve the problems of insufficient testing accuracy and low efficiency when performing dynamic testing in complex electromagnetic environments.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an automated testing method for a high-frequency switching matrix, comprising,

[0008] Collect compressed sensing parameter sets and environmental monitoring baseline data and preprocess them to generate spatiotemporal-environmental baseline data;

[0009] Based on the spatiotemporal-environmental baseline data, the gain and phase compensation of synaptic weights under the current environment are calculated, and a frequency response compensation curve is generated through a dynamic pulse weighted fusion algorithm.

[0010] Based on the frequency response compensation curve, the confidence spectrum is calculated, and the compressed sampling rate is adjusted to construct a random sampling mode;

[0011] Based on the random sampling pattern, the full-band response characteristics of the spatiotemporal-environmental baseline data are verified, and abnormal frequency points are identified.

[0012] The abnormal frequency points are compared and analyzed point by point with historical benchmark data. The mean square error and maximum deviation are calculated. The error frequency response curve and test report are generated by piecewise polynomial fitting method.

[0013] As a preferred embodiment of the automated testing method for high-frequency switching matrices described in this invention, the compressed sensing parameter set includes a random measurement matrix, a sparse base dictionary, reconstruction algorithm parameters, and a dynamic sampling rate configuration.

[0014] The environmental monitoring baseline data includes temperature field distribution, three-dimensional vibration spectrum, and near-electromagnetic environment;

[0015] The preprocessing includes noise filtering, time alignment, normalization, and feature extraction.

[0016] As a preferred embodiment of the automated testing method for high-frequency switching matrices described in this invention, the specific steps for generating spatiotemporal-environmental baseline data are as follows:

[0017] The preprocessed compressed sensing parameter set and environmental monitoring baseline data are time-aligned and sorted according to high-precision timestamps to establish a unified time axis;

[0018] Based on a unified time axis, the compressed sensing parameter set and environmental monitoring baseline data are gridded to generate a data cube;

[0019] Spatially align the data cubes to generate spatiotemporal-environmental baseline data.

[0020] As a preferred embodiment of the automated testing method for high-frequency switching matrices described in this invention, the specific steps for calculating the gain and phase compensation of synaptic weights under the current environment based on spatiotemporal-environmental baseline data are as follows:

[0021] Spatial correlation clustering is performed on the spatiotemporal-environmental baseline data to obtain the hierarchical relationship between the spatiotemporal-environmental baseline data nodes, and the spatiotemporal-environmental topological relationship is obtained through a dynamic spatiotemporal causal inference algorithm.

[0022] Based on the spatiotemporal-environment topology relationship, a spiking neural network with a spatiotemporal-environment topology structure is constructed using the topology mapping method;

[0023] Based on a spatiotemporal-environment topology spiking neural network, the gain and phase compensation of synaptic weights are calculated in the current environment.

[0024] As a preferred embodiment of the automated testing method for high-frequency switching matrices described in this invention, the specific steps for generating the frequency response compensation curve using a dynamic pulse weighted fusion algorithm are as follows:

[0025] Based on the gain and phase compensation of the synaptic weights, the synaptic connection strength between neurons in a spatiotemporal-environment topology spiking neural network is obtained through a dynamic pulse weighted fusion algorithm, and then weighted fusion is performed to obtain the compensated frequency response.

[0026] The frequency response is smoothed to generate a frequency response compensation curve.

[0027] As a preferred embodiment of the automated testing method for high-frequency switching matrices described in this invention, the specific steps for calculating the confidence spectrum based on the frequency response compensation curve are as follows:

[0028] The frequency response compensation curve is divided into frequency bands according to a fixed frequency to generate the frequency response interval of each frequency point, and the amplitude and phase values ​​of each frequency point are extracted by the orthogonal matching pursuit method.

[0029] The amplitude and phase values ​​at each frequency point are defined as state variables to construct the process noise matrix;

[0030] Based on the process noise matrix, the state estimate and covariance values ​​of the frequency points are calculated, and principal component fusion analysis is performed to generate the state-covariance matrix.

[0031] Calculate the trace of the state-covariance matrix and perform spectral analysis to generate the frequency confidence spectrum.

[0032] As a preferred embodiment of the automated testing method for high-frequency switching matrices described in this invention, the specific steps for adjusting the compressed sampling rate and constructing the random sampling mode are as follows:

[0033] The frequency confidence spectrum is evenly divided to obtain equally divided frequency confidence sub-spectrums, and the confidence weight factor is calculated by average weighting.

[0034] By fusing and comparing the confidence weight factor with historical confidence-sampling rate mapping data, a sub-band sampling scheme with weighted labels is generated, and the compressed sampling rate is adjusted to construct a random sampling mode.

[0035] As a preferred embodiment of the automated testing method for high-frequency switching matrices described in this invention, the specific steps for verifying the full-band response characteristics of spatiotemporal-environmental baseline data and identifying abnormal frequency points based on a random sampling mode are as follows:

[0036] Based on the random sampling pattern, wavelet packet decomposition is performed on the spatiotemporal-environment baseline data to generate an energy distribution matrix, and discrete wavelet transform is performed to obtain the approximate coefficients and detail coefficients of the energy distribution.

[0037] Based on the approximation coefficient and detail coefficient of the energy distribution, the wavelet packet energy value of each node is calculated, and the time-frequency energy distribution matrix is ​​generated by the time-frequency gridding algorithm;

[0038] Principal component analysis is performed on the time-frequency energy distribution matrix to extract the dominant modes, and spatiotemporal calibration is performed to generate time-frequency energy feature vectors;

[0039] Based on the time-frequency energy feature vector, the median absolute deviation is calculated and compared with the time-frequency energy deviation safety threshold to identify abnormal frequency points.

[0040] As a preferred embodiment of the automated testing method for high-frequency switching matrices described in this invention, the steps of comparing and analyzing abnormal frequency points with historical benchmark data point by point, calculating the mean square error and maximum deviation, and generating an error frequency response curve and test report using a piecewise polynomial fitting method are as follows.

[0041] The abnormal frequency points are compared and analyzed point by point with historical benchmark data to calculate the abnormal mean square error and maximum deviation, and the amplitude difference is obtained through the difference calculation function.

[0042] Based on the amplitude difference, an error frequency response curve is generated using a piecewise polynomial fitting method, and then integrated with the frequency response compensation curve and abnormal frequency points to generate a test report.

[0043] Secondly, this invention provides an automated testing system for a high-frequency switching matrix, comprising:

[0044] The data acquisition module is used to collect compressed sensing parameter sets and environmental monitoring baseline data and perform preprocessing to generate spatiotemporal-environmental baseline data;

[0045] The compensation curve generation module is used to calculate the gain and phase compensation of synaptic weights under the current environment based on spatiotemporal-environmental baseline data, and generate frequency response compensation curves through a dynamic pulse weighted fusion algorithm.

[0046] The compressed sampling module is used to calculate the confidence spectrum based on the frequency response compensation curve, adjust the compressed sampling rate, and construct a random sampling mode.

[0047] The anomaly identification module is used to verify the full-band response characteristics of spatiotemporal-environmental baseline data based on random sampling patterns and to identify anomalous frequency points.

[0048] The report generation module is used to perform point-by-point comparison and analysis of abnormal frequency points with historical benchmark data, calculate the mean square error and maximum deviation, and generate error frequency response curves and test reports through piecewise polynomial fitting.

[0049] The beneficial effects of this invention are as follows: by calculating the gain and phase compensation of synaptic weights under the current environment based on spatiotemporal-environmental baseline data, a precise response to dynamic testing requirements under complex electromagnetic environments is achieved; by adjusting the compressed sampling rate and constructing a random sampling mode, an adjustment mechanism for changes in characteristics of different frequency bands is realized; by dividing the confidence spectrum equally and calculating the confidence weight factor, a sub-band sampling scheme with weighted labels is generated, thereby achieving high-efficiency and high-quality automated batch testing. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart of an automated testing method for high-frequency switching matrices.

[0052] Figure 2 This is a schematic diagram of an automated testing system for a high-frequency switching matrix.

[0053] Figure 3 A flowchart for generating frequency response compensation curves.

[0054] Figure 4 A flowchart for constructing a random sampling mode based on the frequency response compensation curve. Detailed Implementation

[0055] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0056] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0057] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0058] Reference Figures 1-4 This embodiment of the invention provides an automated testing method for a high-frequency switching matrix, comprising the following steps: (The apparatus for testing the electrical performance of the present invention is described in the original text.)

[0059] S1. Collect compressed sensing parameter sets and environmental monitoring baseline data and preprocess them to generate spatiotemporal-environmental baseline data;

[0060] The compressed sensing parameter set includes a random measurement matrix, a sparse base dictionary, reconstruction algorithm parameters, and dynamic sampling rate configuration;

[0061] Furthermore, pseudo-random numbers are used to generate random measurement matrices that follow a specific distribution (such as Gaussian or Bernoulli distribution). The dimension of the random measurement matrix is ​​dynamically determined based on the Nyquist sampling theorem and the signal bandwidth. A sparse basis dictionary is selected according to the signal characteristics: Fourier basis is used for signals with obvious periodic characteristics, wavelet basis is used for transient signals, and K-SVD algorithm is used to train the sparse basis dictionary for complex signals. The reconstruction algorithm parameters are determined through iterative optimization in the simulation environment, including key parameters such as the number of iterations and regularization coefficients. The dynamic sampling rate configuration is dynamically adjusted based on real-time monitoring of channel conditions. For example, a 30% sampling rate is used when the signal-to-noise ratio is higher than 40dB, a 50% sampling rate is used when the signal-to-noise ratio is between 30-40dB, and the sampling rate is increased to 70% when the signal-to-noise ratio is lower than 30dB. All compressed sensing parameter sets are verified in the hardware-in-the-loop to ensure their stability and reliability in an environment ranging from -40℃ to 85℃.

[0062] Environmental monitoring baseline data includes temperature field distribution, three-dimensional vibration spectrum, and near-electromagnetic environment;

[0063] Furthermore, the acquisition of environmental monitoring baseline data is achieved through multi-sensor fusion. The temperature field distribution is obtained using a high-precision infrared thermal imager and a distributed temperature sensor network; the three-dimensional vibration spectrum is acquired using a triaxial MEMS accelerometer array; and the near-electromagnetic environment is acquired using a broadband radio frequency probe and a spectrum analyzer. All environmental monitoring baseline data are synchronized at the μs level through the PTP precision time protocol to generate environmental monitoring baseline data containing spatiotemporal tags.

[0064] Preprocessing includes noise filtering, time alignment, normalization, and feature extraction;

[0065] Furthermore, noise removal employs a combination of wavelet thresholding and Kalman filtering. A five-level decomposition is performed using the Daubechies wavelet basis, and the reduced sensing parameter set and environmental monitoring baseline data are reduced using the SURE thresholding strategy. Random noise is suppressed using a self-Kalman filter. Time alignment is based on a precise time protocol, employing least squares to fit the clock offsets of each sensor, achieving μs-level synchronization of multi-source data. Normalization utilizes a dual mechanism of Z-score standardization and range scaling, mapping the reduced sensing parameter set and environmental monitoring baseline data to a unified dimension of [0,1]. Feature extraction is performed in the time domain, frequency domain, and time-frequency domain. All preprocessing steps are accelerated by FPGA hardware to ensure real-time performance.

[0066] The preprocessed compressed sensing parameter set and environmental monitoring baseline data are time-aligned and sorted according to high-precision timestamps to establish a unified time axis;

[0067] Furthermore, the preprocessed compressed sensing parameter set and environmental monitoring baseline data are synchronized at the nanosecond level using high-precision timestamps. The timestamps of the compressed sensing parameter set and the environmental monitoring baseline data are aligned using an interpolation algorithm. The aligned compressed sensing parameter set and environmental monitoring baseline data are arranged in ascending order of time. Each parameter entry in the compressed sensing parameter set is strictly matched with each monitoring entry in the environmental monitoring baseline data according to the timestamp, establishing a unified time axis based on absolute time. The starting point of the unified time axis is referenced to the GPS clock, and the time interval is fixed. All data points in the compressed sensing parameter set and environmental monitoring baseline data are reindexed according to the unified time axis to ensure complete consistency in the time dimension.

[0068] Based on a unified time axis, a data cube is generated by gridding the compressed sensing parameter set and environmental monitoring baseline data through a spatial interpolation algorithm.

[0069] Furthermore, based on a unified time axis, the Kriging spatial interpolation algorithm is used to grid the compressed sensing parameter set and environmental monitoring baseline data. The random measurement matrix, sparse basis dictionary, reconstruction algorithm parameters, and dynamic sampling rate configuration parameters of the compressed sensing parameter set are mapped to the temperature field distribution data, three-dimensional vibration spectrum data, and near-electromagnetic environment data of the environmental monitoring baseline data according to a unified time axis sequence. The grid dimension is set to time × space × parameter. Missing grid points are filled by radial basis function interpolation, and finally a data cube containing multiple dimensions of time, space, and parameters is generated. The dimensional structure and numerical range of the data cube are completely determined by the original characteristics of the compressed sensing parameter set and environmental monitoring baseline data.

[0070] The data cubes are spatially aligned using a dynamic spatiotemporal registration algorithm to generate spatiotemporal-environmental baseline data.

[0071] Furthermore, multimodal spatial feature points are extracted from the data cube using the SIFT (Scale Invariant Feature Transform) algorithm. Specifically, these include abrupt changes in the amplitude of the sparse base dictionary atoms in the compressed sensing parameter set, and extreme values ​​of the temperature field gradient, peak values ​​of the vibration spectrum energy, and abrupt changes in the electromagnetic field intensity of the environmental monitoring baseline data. The RANSAC algorithm is then used to calculate the spatial transformation parameters of these multimodal spatial feature points. Based on these parameters, the spatial coordinates of the data cube are elastically registered using a non-rigid registration method, with the point spacing set to, for example, 10 grid units, to ensure complete spatial alignment between the compressed sensing parameter set and the environmental monitoring baseline data. Finally, spatiotemporal-environmental baseline data, fusing temporal, spatial, and environmental multidimensional features, is generated. The spatial consistency error of the spatiotemporal-environmental baseline data is controlled to, for example, within 0.1 mm, while the temporal dimension maintains a unified timeline sequence completely consistent with the data cube.

[0072] S2. Based on the spatiotemporal-environmental baseline data, calculate the gain and phase compensation of the synaptic weights under the current environment, and generate the frequency response compensation curve through the dynamic pulse weighted fusion algorithm;

[0073] The DBSCAN algorithm is used to perform spatial correlation clustering on the spatiotemporal-environment baseline data to obtain the hierarchical relationship between the spatiotemporal-environment baseline data nodes, and the spatiotemporal-environment topological relationship is obtained through the dynamic spatiotemporal causal inference algorithm.

[0074] Furthermore, a neighborhood radius parameter ε and a minimum number of points MinPts are defined. The DBSCAN algorithm is used to calculate the number of neighboring points within the ε neighborhood of the spatiotemporal-environment baseline data. If a point's neighborhood contains at least MinPts points, it is marked as a core point, and these neighboring points are included in the same cluster. The clustering range is continuously expanded through density reachability propagation. Points with fewer than MinPts in their neighborhood but located within the ε neighborhood of a core point are marked as boundary points. Other isolated points that do not meet the density connectivity conditions are identified as noise points. A hierarchical relationship tree of spatiotemporal-environment baseline data nodes is constructed based on core points, boundary points, and noise points. The root node of the hierarchical relationship tree represents the center of a high-density region, and the child nodes are arranged in descending order of spatial density. Subsequently, a dynamic spatiotemporal causal inference algorithm is used to calculate the time series causal relationship index between spatiotemporal-environment baseline data nodes and analyze the time series causal relationship between them. A hierarchical causal fusion algorithm is used to integrate and analyze the hierarchical relationship tree and the time series causal relationship to generate a spatiotemporal-environment topology containing parent-child associations and causal orientations.

[0075] It should be noted that the neighborhood radius parameter ε and the minimum number of points MinPts are defined based on the spatial distribution density characteristics of the spatiotemporal-environmental baseline data and the clustering granularity requirements.

[0076] It should be noted that the hierarchical relationship between nodes refers to the tree-like structural dependency relationship formed by nodes at different spatial locations in the spatiotemporal-environmental baseline data based on the density clustering results. Specifically, it is a topological hierarchy in which nodes in high-density areas act as parent nodes and dominate child nodes in low-density areas.

[0077] Based on the spatiotemporal-environment topology relationship, a spiking neural network with a spatiotemporal-environment topology structure is constructed using the topology mapping method;

[0078] Furthermore, through a topology mapping mechanism, the node hierarchy and connection relationships in the spatiotemporal-environment topology are transformed into the neuron structure and synaptic connection pattern of the spiking neural network. Each node in the spatiotemporal-environment topology is mapped to a neuron in the spiking neural network, the parent-child relationship in the spatiotemporal-environment topology is mapped to the inter-layer feedforward connection in the spiking neural network, and the causal orientation in the spatiotemporal-environment topology is mapped to the cross-layer recursive connection in the spiking neural network. The neuron parameters of the spiking neural network are initialized according to the node attributes of the spatiotemporal-environment topology, thus constructing a spiking neural network with a complete spatiotemporal-environment topology. The topology of the spiking neural network with the spatiotemporal-environment topology maintains a hierarchy and causal relationship that is completely consistent with the spatiotemporal-environment topology.

[0079] It should be noted that the training process of the spatiotemporal-environment topology spiking neural network first constructs the initial connection topology of neurons based on the input spatiotemporal-environment information, and sets the initial values ​​of the gain and phase compensation of the synaptic weights between each neuron; then, it inputs a training pulse signal sequence into the spatiotemporal-environment topology spiking neural network, calculates the synaptic connection strength between neurons through a dynamic pulse weighted fusion algorithm, and performs weighted fusion to obtain the current frequency response; it compares the obtained frequency response with the expected frequency response (generated by joint prediction of the standard frequency response curve in historical benchmark data and environmental compensation), and uses the weighted least squares method to perform joint quantization to calculate the error signal, and adjusts the gain and phase compensation of the synaptic weights based on the error signal using a gradient descent method; it repeats the above forward propagation and parameter adjustment process until the error of the frequency response converges to below the preset safe error threshold (usually ranging from 80% to 85%), thus completing the training of the spatiotemporal-environment topology spiking neural network.

[0080] It should be noted that the preset safety error threshold is determined based on the signal integrity requirements of the cumulative contribution rate of principal component analysis, combined with the convergence characteristics of spiking neural network training and hardware quantization error, and is dynamically calibrated through reverse verification of historical benchmark data and real-time confidence spectrum analysis.

[0081] It should be noted that the initial values ​​of the gain and phase compensation of the synaptic weights between neurons are determined based on the spatial density distribution of the spatiotemporal-environmental baseline data nodes (DBSCAN clustering results) and the causal inference strength (output of the dynamic spatiotemporal causal inference algorithm).

[0082] Based on a spatiotemporal-environment topology spiking neural network, the gain and phase compensation of synaptic weights are calculated in the current environment.

[0083] Furthermore, based on the spatiotemporal-environment topology of the spiking neural network, real-time environmental monitoring data is input into the corresponding neurons of the spiking neural network. By using the pulse timing-dependent plasticity rule, the pulse firing time of presynaptic and postsynaptic neurons is continuously observed to obtain the time interval, and weighted accumulation is performed to calculate the synaptic weight adjustment between neurons. The impulse response integral method is used to continuously integrate the synaptic weight adjustment over time to obtain the gain compensation of the synaptic weight. The gain compensation of the synaptic weight is calculated through the linear relationship between the pulse interval time and the phase shift. The final output includes the specific gain compensation and phase compensation of each synaptic connection.

[0084] It should be noted that the pulse timing dependence plasticity rule is obtained by jointly calibrating the dynamic spatiotemporal causal inference results of spatiotemporal-environmental baseline data with the statistical characteristics of historical pulse intervals, combined with the hardware response characteristics of the high-frequency switching matrix.

[0085] It should be noted that the gain and phase compensation of the synaptic weights refer to the adjustment parameter values ​​calculated based on the pulse neural network, which are used to correct the amplitude attenuation and phase shift during signal transmission. The gain compensation is in decibels to compensate for signal strength loss, and the phase compensation is in degrees to correct signal timing deviation.

[0086] Based on the gain and phase compensation of the synaptic weights, the synaptic connection strength between neurons in a spatiotemporal-environment topology spiking neural network is obtained through a dynamic pulse weighted fusion algorithm, and then weighted fusion is performed to obtain the compensated frequency response.

[0087] Furthermore, based on the gain and phase compensation of the synaptic weights, the input pulse sequence of each neuron in the spatiotemporal-environment topology of the spiking neural network is determined. For each input pulse, the pulse amplitude is adjusted according to the gain of the corresponding synaptic weight, and the pulse firing time is offset and corrected according to the phase compensation. After completing the gain and phase adjustment of a single pulse, all adjusted pulse signals from different synapses are dynamically weighted and fused according to their connection relationships in the spatiotemporal-environment topology using a dynamic pulse weighting fusion algorithm, and time and space calibration is performed to form a fused pulse response sequence. The fused pulse response sequence is then subjected to a Fourier transform to convert the time domain to the frequency domain, thereby obtaining the response amplitude and phase information corresponding to each frequency component and acquiring the compensated frequency response.

[0088] It should be noted that synaptic connection strength refers to the degree of connection strength between neurons in a spatiotemporal-environment topology spiking neural network, which is jointly determined by the gain of synaptic weights and the amount of phase compensation when signals are transmitted between neurons through synapses.

[0089] The compensated frequency response is smoothed by Savitzky-Golay filtering to generate a frequency response compensation curve.

[0090] Furthermore, when smoothing the compensated frequency response using Savitzky-Golay filtering, a set of continuous frequency points and their corresponding response value sequences are selected from the frequency response. These sequences are moved across the impulse response sequence in a sliding window manner. Within each window, a polynomial least squares method is used for fitting. The smoothed response value at the center point of the window is calculated using the fitted polynomial. This process is repeated for all frequency points to eliminate high-frequency noise fluctuations in the frequency response. After completing the smoothing calculation for all frequency points, all smoothed response values ​​are connected in frequency order to generate a frequency response compensation curve.

[0091] S3. Based on the frequency response compensation curve, calculate the confidence spectrum and adjust the compressed sampling rate to construct a random sampling mode;

[0092] The frequency response compensation curve is divided into frequency bands according to a fixed frequency to generate the frequency response interval of each frequency point, and the amplitude and phase values ​​of each frequency point are extracted by the orthogonal matching pursuit method.

[0093] Furthermore, when dividing the frequency response compensation curve into frequency bands at a fixed frequency, the entire effective frequency range is divided into several continuous frequency bands with equal frequency steps. The center frequency of each frequency band is taken as a frequency point, and the frequency response interval corresponding to the frequency point is defined by the left and right boundary frequencies (the inflection points where the amplitude of the frequency response compensation curve attenuates significantly). For the frequency response interval of each frequency point, the frequency response compensation curve data within the frequency response interval is taken as input, and the orthogonal matching pursuit method is used to iteratively select the basis function with the highest correlation to the current residual from the preset sparse basis function dictionary, gradually reconstruct the signal, and extract the amplitude and phase values ​​corresponding to the frequency point.

[0094] It should be noted that the pre-set process for the sparse basis function dictionary is as follows: Based on the frequency range and typical variation patterns that the frequency response signal may cover, a set of basis functions with different frequencies, phases, scales, and time shift parameters are constructed based on mathematical functions such as sine, cosine, Gaussian wavelet, or exponential decay; the basis functions are discretized and sampled in the time-frequency domain to form a basis function set containing multiple candidate atoms; the basis function set is optimized through orthogonalization or redundancy control methods to ensure that the basis functions have good distinguishability and sparse representation ability; finally, the optimized function set is used as the sparse basis function dictionary in the orthogonal matching pursuit method.

[0095] The amplitude and phase values ​​at each frequency point are defined as state variables, and a process noise matrix is ​​constructed using a fusion calibration method.

[0096] Furthermore, when defining the amplitude and phase values ​​of each frequency point as state variables, the amplitude and phase values ​​extracted from each frequency point are used as independent state variables to form elements of the state vector. For the elements of each state variable, they are aligned by time and weighted and fused to form a comprehensive response curve. The comprehensive response curve is transformed in the frequency domain to obtain amplitude and phase information at different frequencies. By comparing the frequency domain characteristics under various operating conditions, the dominant frequency of the fluctuation and its variation law are identified, and a complete process noise matrix is ​​constructed.

[0097] Based on the process noise matrix, the state estimate and covariance of the frequency points are calculated by the Kalman filter algorithm, and the principal component fusion analysis is performed by the Bayesian fusion algorithm to generate the state-covariance matrix.

[0098] Furthermore, based on the process noise matrix, when calculating the state estimates and covariance values ​​of frequency points using the Kalman filter algorithm, the state at the previous time step is mapped to the current time step based on the state estimates and covariance values ​​of each frequency point at the previous time step. The state prediction value at the current time step is obtained through Kalman filter prediction, and considering the influence of process noise, the predicted covariance matrix at the current time step is calculated using the covariance propagation formula. The process noise matrix is ​​used as the input for updating the predicted covariance matrix and added to the calculation of the predicted covariance matrix. Subsequently, the difference between the observed value and the state prediction value at the current time step is used to calculate the predicted covariance matrix through covariance propagation. The Kalman gain is calculated, and the state prediction value is corrected based on the Kalman gain to obtain the state estimate value of each frequency point at the current time. At the same time, the covariance value at the current time is updated using the Kalman gain and the prediction covariance matrix. After obtaining the state estimate value and covariance value of each frequency point, they are arranged in frequency point order to form a state vector and a covariance matrix. The state vector and covariance matrix are subjected to principal component fusion analysis by Bayesian fusion algorithm. Based on the contribution rate of the principal components, each state component and its uncertainty information are weighted and fused to generate a state-covariance matrix containing joint information of state estimation and covariance.

[0099] It should be noted that the training process of the state transition model is as follows: Time series data of amplitude and phase values ​​at various frequency points under different time steps are collected, and the state variables at adjacent time points are used as input and output sample pairs; a training sample set is constructed using the current state variable as input and the next state variable as the expected output; the parameters of the state transition matrix are estimated using the least squares method or recursive least squares method, and the parameters are optimized by minimizing the sum of squared errors between the predicted and actual states; after each parameter update, forward prediction is performed using the new state transition matrix, and the prediction error is calculated. This iteration is repeated until the error converges to below a preset safety error threshold, thus completing the training of the state transition model.

[0100] It should be noted that the preset safety error threshold is set comprehensively based on historical operating data, equipment performance indicators, and fault tolerance capabilities in actual application scenarios.

[0101] Calculate the trace of the state-covariance matrix and perform spectral analysis to generate frequency confidence spectra;

[0102] Furthermore, the trace of the state-covariance matrix is ​​obtained by summing all elements on the main diagonal of the state-covariance matrix. The sequence of the trace of the state-covariance matrix changing over time or under different operating conditions is used as the analysis object for spectral analysis. The sequence of the trace of the state-covariance matrix changing over time or under different operating conditions is converted to the frequency domain using Fast Fourier Transform to obtain the amplitude distribution of each frequency component. The amplitude distribution in the frequency domain is normalized to reflect the relative importance of each frequency point in the uncertainty evolution process, and the frequency confidence spectrum is generated.

[0103] The frequency confidence spectrum is uniformly divided by a dynamic bandwidth allocation method to obtain equally divided frequency confidence sub-spectrums, and the confidence weight factor is calculated by a weighted average algorithm.

[0104] The specific expression for calculating the confidence weight factor is as follows:

[0105] ;

[0106] in, Indicates the first Confidence weighting factors for each frequency point confidence sub-spectrum; Indicates the first The original confidence values ​​of each frequency point in the frequency confidence spectrum; Indicates the first Confidence value at each frequency point The weighting coefficient (usually ranging from 0 to 1). Indicates the first The set of all frequency points contained in the confidence subspectrum of a frequency point.

[0107] Furthermore, when uniformly dividing the frequency confidence spectrum using the dynamic bandwidth allocation method, the total frequency range of the frequency confidence spectrum is divided into several continuous and non-overlapping frequency bands according to the preset number of sub-spectrums. Each frequency band corresponds to an equally divided frequency confidence sub-spectrum. For the confidence values ​​of all frequency points within each equally divided frequency confidence sub-spectrum, a weighted average algorithm is used to calculate the confidence weight factor of the sub-spectrum. Finally, the confidence weight factor corresponding to each sub-spectrum is obtained.

[0108] It should be noted that the preset number of sub-spectrums is determined based on the total frequency range covered by the frequency confidence spectrum and the desired bandwidth resolution. By analyzing the application scenario's requirements for the fineness of frequency band division, a number of sub-spectrums that can balance computational complexity and feature discrimination is selected. The number of sub-spectrums must ensure that each equally divided frequency confidence sub-spectrum contains a sufficient number of frequency points to guarantee statistical stability, while avoiding excessive fluctuations in the confidence weight factor due to overly fine division. The number of sub-spectrums is usually set in integer form.

[0109] By fusing and comparing the confidence weight factor with historical confidence-sampling rate mapping data, a sub-band sampling scheme with weighted labels is generated, and the compressed sampling rate is adjusted to construct a random sampling mode.

[0110] Furthermore, when performing a fusion and comparative analysis of the confidence weight factor and historical confidence-sampling rate mapping data, the confidence weight factor corresponding to each equally divided frequency confidence sub-spectrum is matched with the sampling rate records of the same confidence interval in the historical confidence-sampling rate mapping data. Based on the correspondence between confidence and sampling rate in the historical data, the initial sampling rate recommendation value corresponding to the current confidence weight factor is determined. The initial sampling rate recommendation value is then proportionally adjusted according to the magnitude of the confidence weight factor to generate a weighted sub-band sampling scheme. Based on the sampling rate requirements of each sub-band in the weighted sub-band sampling scheme, the overall compressed sampling rate is redistributed. Combined with a random sequence generator, the specific sampling time is determined within each sub-band to construct a random sampling mode.

[0111] It should be noted that the confidence-sampling rate mapping data refers to the correspondence between each confidence level and the corresponding sampling rate in the frequency confidence spectrum recorded during the historical sampling process.

[0112] S4. Based on the random sampling mode, verify the full-band response characteristics of the spatiotemporal-environmental baseline data and identify abnormal frequency points;

[0113] Based on the random sampling pattern, wavelet packet decomposition is performed on the spatiotemporal-environmental baseline data using the dynamic decomposition method to generate an energy distribution matrix. Then, discrete wavelet transform is performed to obtain the approximate coefficients and detail coefficients of the energy distribution.

[0114] Furthermore, based on the random sampling pattern, data segments matching the random sampling time are extracted from the spatiotemporal-environmental baseline data. A dynamic decomposition method is used to perform wavelet packet decomposition on the extracted data segments. Wavelet basis functions and decomposition levels are selected to obtain wavelet packet coefficients. The high-frequency and low-frequency components of the data segments matching the random sampling time are recursively decomposed layer by layer. After each decomposition, the squares of all wavelet packet coefficients within the corresponding time period for each frequency band node are integrated to obtain the energy value of each frequency band node. The energy values ​​of each frequency band node are arranged according to the decomposition level and frequency band order to generate an energy distribution matrix. Discrete wavelet transform is performed on each row of the energy distribution matrix to extract low-frequency trend information and high-frequency fluctuation information at different scales, obtaining the approximate coefficients and detail coefficients of the energy distribution at the corresponding level.

[0115] Based on the approximation coefficient and detail coefficient of the energy distribution, the wavelet packet energy value of each node is calculated by the energy integration method, and the time-frequency energy distribution matrix is ​​generated by the time-frequency gridding algorithm.

[0116] Furthermore, based on the approximation coefficients and detail coefficients of the energy distribution, the sum of squares of the approximation coefficient sequence and detail coefficient sequence of each wavelet packet decomposition node is calculated using the energy integration method. The sum of squares is used as the wavelet packet energy value of the node, and the energy value calculation of all nodes is completed. A time-frequency gridding algorithm is adopted to divide the frequency interval corresponding to each wavelet packet decomposition node into a fixed frequency band grid and the time axis into a time grid with a fixed time step. According to the time period and frequency segment to which the energy value of each node belongs, it is assigned to the corresponding time-frequency grid. If a single grid contains multiple node energy values, an arithmetic average is performed. Finally, a time-frequency energy distribution matrix is ​​formed with time as the horizontal axis, frequency as the vertical axis, and energy value as the element.

[0117] Principal component analysis is performed on the time-frequency energy distribution matrix to extract the dominant modes, and spatiotemporal calibration is performed to generate time-frequency energy feature vectors;

[0118] Furthermore, when performing principal component analysis on the time-frequency energy distribution matrix, the covariance matrix of the time-frequency energy distribution matrix is ​​calculated, and the eigenvalues ​​and corresponding eigenvectors of the covariance matrix are solved. The eigenvalues ​​are sorted from largest to smallest, and the eigenvectors corresponding to the top K largest eigenvalues ​​are selected as the dominant modes. The value of K is determined based on the cumulative contribution rate exceeding a safety threshold (typically ranging from 85% to 90%). The time-frequency energy distribution matrix is ​​then projected onto the selected dominant modes to obtain the dimensionality-reduced mode coefficient matrix. Based on the dimensionality-reduced mode coefficient matrix, spatiotemporal calibration is performed in both time and space dimensions. The temporal offset and spatial distribution of the modes are linearly or nonlinearly adjusted according to known spatiotemporal reference information to align the dominant modes with standard spatiotemporal coordinates. Finally, the calibrated dominant modes are arranged in frequency channel order to form the time-frequency energy eigenvectors.

[0119] It should be noted that the safety threshold is obtained by extracting the eigenvalues ​​of the time-frequency energy distribution matrix through principal component analysis, using a cumulative contribution rate of 85% to 90% as the judgment boundary for signal feature integrity, and obtaining the safety threshold after hardware-in-the-loop verification.

[0120] It should be noted that the dominant mode refers to the mode corresponding to the top K eigenvectors that can explain most of the variance of the energy distribution, extracted by principal component analysis of the time-frequency energy distribution matrix.

[0121] Based on the time-frequency energy feature vector, the absolute deviation of the median is calculated using a robust statistical method, and then compared with the deviation safety threshold of the time-frequency energy to identify abnormal frequency points.

[0122] Furthermore, all elements in the time-frequency energy feature vector are considered as a set of time-frequency energy observations. The median is calculated using robust statistical methods, and the absolute value of the difference between each element and the median is obtained. Then, the median of the absolute deviation is calculated to obtain the median absolute deviation. The calculated median absolute deviation is compared with the deviation safety threshold of time-frequency energy (usually within ±5%). If the median absolute deviation is greater than the deviation safety threshold of time-frequency energy, it is determined that there is a significant deviation of energy distribution from the normal range in the time-frequency energy feature vector, and the corresponding abnormal frequency point is identified. The identification process locates the specific frequency point based on the position index of each element in the time-frequency energy feature vector.

[0123] It should be noted that the deviation safety threshold is set based on a comprehensive analysis of historical deviation statistics during normal operation, equipment accuracy requirements, environmental interference levels, and safety margins in practical applications.

[0124] S5. Compare and analyze the abnormal frequency points with historical benchmark data point by point, calculate the mean square error and maximum deviation, and generate the error frequency response curve and test report by piecewise polynomial fitting method.

[0125] The abnormal frequency points are compared with historical benchmark data point by point. The mean square error and maximum deviation of the abnormality are calculated by weighted least squares regression. The amplitude difference is obtained by the difference calculation function.

[0126] Furthermore, when performing point-by-point comparative analysis between abnormal frequency points and historical benchmark data, for each identified abnormal frequency point, the amplitude value in the current time-frequency energy feature vector is extracted, and the benchmark amplitude value corresponding to the same frequency point is obtained from the historical benchmark data to form paired data points. The paired data points are fitted using weighted least squares regression, with the benchmark amplitude value as the independent variable and the current amplitude value as the dependent variable. The weighted sum of squares of the fitting residuals is calculated to obtain the abnormal mean square error. At the same time, the value with the largest absolute value of the residuals among all paired data points is identified as the maximum deviation. The amplitude difference is obtained by subtracting the current amplitude value from the benchmark amplitude value for each abnormal frequency point using a difference calculation function.

[0127] It should be noted that the difference calculation function refers to the mathematical function used to calculate the difference between the amplitude value of the abnormal frequency point in the current time-frequency energy feature vector and the reference amplitude value of the corresponding frequency point in the historical reference data.

[0128] Based on the amplitude difference, an error frequency response curve is generated using a piecewise polynomial fitting method, and then integrated with the frequency response compensation curve and abnormal frequency points to generate a test report.

[0129] Furthermore, based on the amplitude difference, the amplitude difference corresponding to each abnormal frequency point is used as the function value and the frequency point frequency is used as the independent variable to construct a discrete data point sequence. A piecewise polynomial fitting method is adopted to divide the interval between adjacent frequency points. For the data points in each interval, the least squares method is used to fit a low-order polynomial to ensure that the polynomials of adjacent intervals are continuous at the connection point. Finally, the error frequency response curve covering the range of abnormal frequency points is spliced ​​to generate an error frequency response curve. The error frequency response curve and the frequency response compensation curve are aligned on the same frequency axis, and the positions of each abnormal frequency point are marked. The graphs are overlaid and the data is integrated in a unified coordinate system to generate a test report containing curve comparison, anomaly markings and numerical lists.

[0130] This embodiment also provides an automated testing system for high-frequency switching matrices, including:

[0131] The data acquisition module is used to collect compressed sensing parameter sets and environmental monitoring baseline data and perform preprocessing to generate spatiotemporal-environmental baseline data;

[0132] The compensation curve generation module is used to calculate the gain and phase compensation of synaptic weights under the current environment based on spatiotemporal-environmental baseline data, and generate frequency response compensation curves through a dynamic pulse weighted fusion algorithm.

[0133] The compressed sampling module is used to calculate the confidence spectrum based on the frequency response compensation curve, adjust the compressed sampling rate, and construct a random sampling mode.

[0134] The anomaly identification module is used to verify the full-band response characteristics of spatiotemporal-environmental baseline data based on random sampling patterns and to identify anomalous frequency points.

[0135] The report generation module is used to perform point-by-point comparison and analysis of abnormal frequency points with historical benchmark data, calculate the mean square error and maximum deviation, and generate error frequency response curves and test reports through piecewise polynomial fitting.

[0136] This embodiment also provides a computer device applicable to the automated testing method for high-frequency switching matrices, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the automated testing method for high-frequency switching matrices as proposed in the above embodiment.

[0137] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0138] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the automated testing method for high-frequency switching matrices as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0139] In summary, this invention achieves precise response to dynamic testing requirements in complex electromagnetic environments by: calculating the gain and phase compensation of synaptic weights under the current environment based on spatiotemporal-environmental baseline data; constructing a random sampling mode by adjusting the compressed sampling rate, thus realizing an adaptive adjustment mechanism for changes in characteristics of different frequency bands; and generating a sub-band sampling scheme with weighted labels by evenly dividing the confidence spectrum and calculating the confidence weight factor.

[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An automated testing method for a high-frequency switching matrix, characterized in that: include, Collect compressed sensing parameter sets and environmental monitoring baseline data and preprocess them to generate spatiotemporal-environmental baseline data; Based on the spatiotemporal-environmental baseline data, the gain and phase compensation of synaptic weights under the current environment are calculated, and a frequency response compensation curve is generated through a dynamic pulse weighted fusion algorithm. Based on the frequency response compensation curve, the confidence spectrum is calculated, and the compressed sampling rate is adjusted to construct a random sampling mode; Based on the random sampling pattern, the full-band response characteristics of the spatiotemporal-environmental baseline data are verified, and abnormal frequency points are identified. The abnormal frequency points are compared and analyzed point by point with historical benchmark data. The mean square error and maximum deviation are calculated. The error frequency response curve and test report are generated by piecewise polynomial fitting method.

2. The automated testing method for high-frequency switching matrices as described in claim 1, characterized in that: The compressed sensing parameter set includes a random measurement matrix, a sparse base dictionary, reconstruction algorithm parameters, and a dynamic sampling rate configuration. The environmental monitoring baseline data includes temperature field distribution, three-dimensional vibration spectrum, and near-electromagnetic environment; The preprocessing includes noise filtering, time alignment, normalization, and feature extraction.

3. The automated testing method for high-frequency switching matrices as described in claim 2, characterized in that: The specific steps for generating the spatiotemporal-environmental baseline data are as follows. The preprocessed compressed sensing parameter set and environmental monitoring baseline data are time-aligned and sorted according to high-precision timestamps to establish a unified time axis; Based on a unified time axis, the compressed sensing parameter set and environmental monitoring baseline data are gridded to generate a data cube; Spatially align the data cubes to generate spatiotemporal-environmental baseline data.

4. The automated testing method for high-frequency switching matrices as described in claim 3, characterized in that: The specific steps for calculating the gain and phase compensation of synaptic weights under the current environment based on spatiotemporal-environmental baseline data are as follows. Spatial correlation clustering is performed on the spatiotemporal-environmental baseline data to obtain the hierarchical relationship between the spatiotemporal-environmental baseline data nodes, and the spatiotemporal-environmental topological relationship is obtained through a dynamic spatiotemporal causal inference algorithm. Based on the spatiotemporal-environment topology relationship, a spiking neural network with a spatiotemporal-environment topology structure is constructed using the topology mapping method; Based on a spatiotemporal-environment topology spiking neural network, the gain and phase compensation of synaptic weights are calculated in the current environment.

5. The automated testing method for high-frequency switching matrices as described in claim 4, characterized in that: The specific steps for generating the frequency response compensation curve using the dynamic pulse weighted fusion algorithm are as follows: Based on the gain and phase compensation of the synaptic weights, the synaptic connection strength between neurons in a spatiotemporal-environment topology spiking neural network is obtained through a dynamic pulse weighted fusion algorithm, and then weighted fusion is performed to obtain the compensated frequency response. The frequency response is smoothed to generate a frequency response compensation curve.

6. The automated testing method for high-frequency switching matrices as described in claim 5, characterized in that: The specific steps for calculating the confidence spectrum based on the frequency response compensation curve are as follows. The frequency response compensation curve is divided into frequency bands according to a fixed frequency to generate the frequency response interval of each frequency point, and the amplitude and phase values ​​of each frequency point are extracted by the orthogonal matching pursuit method. The amplitude and phase values ​​at each frequency point are defined as state variables to construct the process noise matrix; Based on the process noise matrix, the state estimate and covariance values ​​of the frequency points are calculated, and principal component fusion analysis is performed to generate the state-covariance matrix. Calculate the trace of the state-covariance matrix and perform spectral analysis to generate the frequency confidence spectrum.

7. The automated testing method for high-frequency switching matrices as described in claim 6, characterized in that: The specific steps for adjusting the compression sampling rate and constructing the random sampling pattern are as follows. The frequency confidence spectrum is evenly divided to obtain equally divided frequency confidence sub-spectrums, and the confidence weight factor is calculated by average weighting. By fusing and comparing the confidence weight factor with historical confidence-sampling rate mapping data, a sub-band sampling scheme with weighted labels is generated, and the compressed sampling rate is adjusted to construct a random sampling mode.

8. The automated testing method for high-frequency switching matrices as described in claim 7, characterized in that: The process of verifying the full-band response characteristics of the spatiotemporal-environment baseline data based on a random sampling pattern and identifying abnormal frequency points involves the following specific steps: Based on the random sampling pattern, wavelet packet decomposition is performed on the spatiotemporal-environment baseline data to generate an energy distribution matrix, and discrete wavelet transform is performed to obtain the approximate coefficients and detail coefficients of the energy distribution. Based on the approximation coefficient and detail coefficient of the energy distribution, the wavelet packet energy value of each node is calculated, and the time-frequency energy distribution matrix is ​​generated by the time-frequency gridding algorithm; Principal component analysis is performed on the time-frequency energy distribution matrix to extract the dominant modes, and spatiotemporal calibration is performed to generate time-frequency energy feature vectors; Based on the time-frequency energy feature vector, the median absolute deviation is calculated and compared with the time-frequency energy deviation safety threshold to identify abnormal frequency points.

9. The automated testing method for high-frequency switching matrices as described in claim 8, characterized in that: The process involves comparing abnormal frequency points with historical benchmark data point by point, calculating the mean square error and maximum deviation, and generating an error frequency response curve and test report using a piecewise polynomial fitting method. The specific steps are as follows: The abnormal frequency points are compared and analyzed point by point with historical benchmark data to calculate the abnormal mean square error and maximum deviation, and the amplitude difference is obtained through the difference calculation function. Based on the amplitude difference, an error frequency response curve is generated using a piecewise polynomial fitting method, and then integrated with the frequency response compensation curve and abnormal frequency points to generate a test report.

10. An automated testing system for a high-frequency switching matrix, based on the automated testing method for a high-frequency switching matrix according to any one of claims 1 to 9, characterized in that: include, The data acquisition module is used to collect compressed sensing parameter sets and environmental monitoring baseline data and perform preprocessing to generate spatiotemporal-environmental baseline data; The compensation curve generation module is used to calculate the gain and phase compensation of synaptic weights under the current environment based on spatiotemporal-environmental baseline data, and generate frequency response compensation curves through a dynamic pulse weighted fusion algorithm. The compressed sampling module is used to calculate the confidence spectrum based on the frequency response compensation curve, adjust the compressed sampling rate, and construct a random sampling mode. The anomaly identification module is used to verify the full-band response characteristics of spatiotemporal-environmental baseline data based on random sampling patterns and to identify anomalous frequency points. The report generation module is used to perform point-by-point comparison and analysis of abnormal frequency points with historical benchmark data, calculate the mean square error and maximum deviation, and generate error frequency response curves and test reports through piecewise polynomial fitting.

Images