A method and system for electromagnetic compatibility detection

Abstract

This invention provides an electromagnetic compatibility (EMC) testing method and system, relating to the field of EMC testing technology. The method includes: collecting background noise and constructing a background noise feature library when the device under test (DUT) is not powered on; collecting time-domain waveform data when the DUT is powered on and running, performing sliding window matching with the background noise feature library to filter strong background noise windows; constructing a reference signal using the strong background noise windows, extracting candidate interference waveforms after adaptive filtering to remove background noise, and then filtering steady-state interference waveforms through segmented energy accumulation and fluctuation amplitude analysis; comparing the steady-state interference waveforms with standard limit lines to mark the exceeding frequency bands and their time-domain waveform segments; finally, extracting time-frequency joint features and inputting them into a classification model to output interference source type identification results. This invention can effectively remove complex background noise in the field, accurately extract the device's own steady-state interference, and automatically identify the interference source type, improving the accuracy and intelligence level of EMC testing.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic compatibility testing technology, specifically to an electromagnetic compatibility testing method and system. Background Technology

[0002] Electromagnetic compatibility (EMC) testing is a crucial method for evaluating whether electrical and electronic equipment can function properly in an electromagnetic environment without causing unacceptable electromagnetic interference to other devices. Traditional EMC testing typically requires a standardized semi-anechoic chamber, where receiving antennas and spectrum analyzers positioned at specific locations are used to test the radiated and conducted emissions of the device under test according to preset test standards. However, for large electrical equipment (such as wind turbines, industrial frequency converters, and substation equipment), their size and weight prevent them from entering a standard anechoic chamber, necessitating on-site testing. While on-site testing solves the problem of large equipment being unable to enter the chamber, the background electromagnetic noise in the on-site environment is complex and variable, including various environmental radiation sources, interference from nearby equipment, and human electromagnetic activities. This background noise, combined with the electromagnetic radiation generated by the equipment itself, makes it difficult for the test results to accurately reflect the true EMC performance of the device under test.

[0003] In the prior art, an electromagnetic compatibility (EMC) testing method, system, device, and storage medium disclosed in CN117192273A acquires historical testing records of new energy vehicles, calculates a first testing threshold and the power-down time of mandatory components based on a fuzzy model, and detects a first electromagnetic value after the vehicle is fully powered on. If the first electromagnetic value exceeds the threshold, the mandatory components are powered down one by one, and the second electromagnetic value of each component after power-down is detected based on the power-down time, thereby calculating the EMC value of each mandatory component. This scheme indirectly obtains the EMC contribution of each component through power-down operations, achieving a certain degree of quantitative assessment of component-level EMC. However, this method mainly obtains changes in electromagnetic values ​​through power-up and power-down operations, failing to effectively isolate background noise interference in complex field environments. When the background noise fluctuates greatly or there is environmental interference in the same frequency band as the equipment, the accuracy of the test results will be significantly affected. In addition, this method can only assess whether there is excessive interference in the component, but cannot further identify the specific type of interference source, which is not conducive to subsequent interference investigation and rectification work.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide an electromagnetic compatibility testing method and system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An electromagnetic compatibility testing method, comprising the following steps: Step 1: With the device under test powered off, collect the first time-domain waveform data of the detection point within a preset time period and perform time-frequency analysis to generate a background noise frequency domain feature set. Calculate the energy distribution variance of each frequency point in the background noise frequency domain feature set and construct a background noise feature library. Step 2: When the device under test is powered on and in typical operating conditions, collect the second time domain waveform data of the detection point within the same time period and perform time-frequency analysis. Then, perform sliding window correlation matching with the background noise feature library, calculate the background noise dynamic matching degree index under each sliding window, and select the window with the matching degree index higher than the first threshold as the strong background noise window. Step 3: Construct a background noise reference signal based on the second time-domain waveform data of the strong background noise window. Subtract the background noise component from the background noise reference signal after adaptive filter processing from the second time-domain waveform data to extract the candidate device interference waveform data and segment it. Obtain the energy accumulation value of each segment at each frequency point based on the fast Fourier transform, and compare the fluctuation amplitude of the energy accumulation value at each frequency point with the preset threshold to select the steady-state interference waveform data. Step 4: Perform time-frequency conversion on the steady-state interference waveform data to generate the corresponding frequency domain spectrum, and compare the frequency domain spectrum with the standard limit line on a frequency-by-frequency basis to mark the frequency bands that exceed the standard and their corresponding time domain waveform segments. Step 5: Perform joint time-frequency feature extraction on the time-domain waveform segments corresponding to the frequency bands exceeding the standard, construct the feature vector of exceeding the standard, input the feature vector of exceeding the standard into the preset classification model, and output the interference source type identification result of the frequency band exceeding the standard.

[0007] Furthermore, the specific logic for generating the background noise frequency domain feature set is as follows: when the device under test is in a de-energized state, electromagnetic sensors arranged at the detection points are used to sample at a preset frequency. Continuous data collection for a preset duration The time-domain electromagnetic signal within the range is used to obtain the first time-domain waveform data. ,in The index of the sampling point; the first time-domain waveform data According to the preset window length and sliding step size Divided into The time-domain segments are divided into several time-domain segments, with a sliding step size smaller than the window length, resulting in overlap between adjacent time-domain segments. A Fast Fourier Transform is performed on each time-domain segment to obtain its frequency-domain amplitude spectrum, and the power spectral density of each time-domain segment is calculated. The power spectral densities of all segments are aggregated according to frequency points to obtain the background noise frequency-domain feature set. ,in Indicates the first One frequency point, For frequency point index, For the index of the time domain segment, Indicates the first The time-domain segment in the first Power spectral density values ​​at each frequency point; Furthermore, regarding the background noise frequency domain feature set For each frequency point in the time domain, calculate its energy distribution variance over all time domain segments, specifically as follows: First, calculate the mean power spectral density of each frequency point across all time-domain segments to characterize the average energy level of that frequency point in the background noise; then calculate the average of the squares of the deviations of the power spectral density of each time-domain segment from the mean, and use this average as the variance of the energy distribution of each frequency point to characterize the degree of fluctuation of the background noise at that frequency point across different time-domain segments. Finally, the mean power spectral density and the variance of energy distribution at each frequency point are used together as background noise feature parameters to construct a background noise feature library.

[0008] Furthermore, the device under test is powered on and placed under typical operating conditions, and electromagnetic sensors arranged at the same detection points are used at the same sampling frequency. Collect data for the same duration continuously The time-domain electromagnetic signal within is used to obtain the second time-domain waveform data. ; The second time-domain waveform data According to the preset window length and sliding step size Divided into One time-domain window; For the The second time-domain waveform data segment within a time-domain window Perform a Fast Fourier Transform to obtain the time-domain window at the [number]th [time domain]. Power spectral density at each frequency point ;in, For the index of the time domain window; When calculating the background noise dynamic matching degree index for each time-domain window across all frequency points, for each frequency point, the difference between the power spectral density of the measured signal within the time-domain window and the mean power spectral density of the corresponding frequency point in the background noise feature library is squared, and then divided by twice the energy distribution variance of that frequency point plus the regularization parameter to construct a normalized deviation metric. Subsequently, the deviation metric is negative and subjected to a natural exponential transformation to obtain the local matching degree at each frequency point. Finally, the arithmetic mean of the local matching degrees of all frequency points is calculated as the background noise dynamic matching degree index for that time-domain window. The calculated background noise dynamic matching degree index for each time domain window is compared with a preset first threshold. Compare and filter out those with a background noise dynamic matching degree index greater than or equal to the first threshold. The time-domain window is used as a strong background noise window.

[0009] Furthermore, the second time-domain waveform data segments corresponding to each strong background noise window are extracted, and the second time-domain waveform data segments of all strong background noise windows are spliced ​​together in time order to construct a background noise reference signal; Based on the statistical correlation between the background noise reference signal and the second time-domain waveform data, an adaptive filter is constructed. The coefficients of the adaptive filter are iteratively updated using the minimum mean square error criterion. The updated adaptive filter is used to filter the background noise reference signal to obtain the estimated background noise component. The background noise component is subtracted from the second time-domain waveform data to obtain the candidate device interference waveform data. The interference waveform data of the candidate device is divided into several time segments of equal length according to the time series. For any segment, a fast Fourier transform is performed on the time domain signal in the segment to obtain the spectral distribution of the signal at each frequency point in the segment. Then, the modulus square operation is performed on the spectral components corresponding to each frequency point to obtain the cumulative energy value of the frequency point in the segment.

[0010] Furthermore, for each frequency point, the fluctuation amplitude of the cumulative energy value across all segments is calculated, specifically as follows: First, calculate the average cumulative energy value of the frequency point across all segments as the baseline energy level. Then, calculate the dispersion of the cumulative energy value of each segment relative to the baseline energy level, i.e., calculate the square root of the sum of squares of the deviations of each segment divided by the number of segments minus one, and obtain the standard deviation. Finally, divide the standard deviation by the baseline energy level, and the resulting ratio is the fluctuation amplitude of the frequency point.

[0011] The fluctuation amplitude is compared with a preset second threshold, and frequency points that meet the condition that the fluctuation amplitude is less than or equal to the second threshold are selected as steady-state interference frequency points. The candidate device interference waveform data corresponding to the steady-state interference frequency points are extracted to obtain steady-state interference waveform data.

[0012] Furthermore, the steady-state interference waveform data is converted from time to frequency to generate the corresponding frequency domain spectrum. The frequency domain spectrum is then compared with the standard limit line on a frequency-by-frequency basis to mark the frequency bands exceeding the limit and their corresponding time domain waveform segments. The specific logic behind this is as follows: The steady-state interference waveform data is subjected to time-frequency conversion processing. The steady-state interference waveform data is subjected to fast Fourier transform according to a preset transform window length to obtain the frequency domain spectrum of the steady-state interference waveform data. The frequency domain spectrum contains the power spectral density value corresponding to each frequency point. Obtain standard limit lines corresponding to the type of equipment under test and the test standard, wherein the standard limit lines include electromagnetic radiation limits corresponding to each frequency point; The frequency domain spectrum is compared with the standard limit line at each frequency point. For each frequency point, the power spectral density value is compared with the electromagnetic radiation limit. When the power spectral density value is greater than the electromagnetic radiation limit, the frequency point is marked as an out-of-limit frequency point. A frequency band consisting of consecutive out-of-standard frequency points is defined as an out-of-standard frequency band, and the time-domain waveform segment corresponding to the out-of-standard frequency band is determined based on the inverse transformation relationship of time-frequency conversion.

[0013] Furthermore, time-frequency joint features are extracted from the time-domain waveform segments corresponding to the frequency bands exceeding the standard. The time-frequency joint features include time-domain feature parameters and frequency-domain feature parameters. The time-domain feature parameters include peak amplitude, pulse width, rise time, fall time and zero-crossing rate. The frequency-domain feature parameters include center frequency, bandwidth, harmonic order distribution and spectral envelope shape features. The extracted time-domain feature parameters and frequency-domain feature parameters are fused to construct the super-standard feature vector; The excess feature vector is input into a preset classification model, which is a pre-trained machine learning classification model. The machine learning classification model is trained by a supervised learning algorithm using historical excess feature vectors labeled with interference source type as training samples. The interference source types include switching power supply interference type, motor drive interference type, wireless communication interference type and electrostatic discharge interference type. After processing by the classification model, the results of identifying the interference source type in the frequency band exceeding the standard are output, thus completing the electromagnetic compatibility test.

[0014] The present invention also provides an electromagnetic compatibility testing system for performing the above-described electromagnetic compatibility testing method, comprising: The background noise library module is used to collect the first time-domain waveform data of the detection point within a preset time period when the device under test is not powered on, perform time-frequency analysis, generate a background noise frequency domain feature set, calculate the energy distribution variance of each frequency point in the background noise frequency domain feature set, and construct a background noise feature library. The strong noise window matching module is used to collect second time-domain waveform data of the detection point within the same time period and perform time-frequency analysis when the device under test is powered on and in typical operating conditions. Then, it performs sliding window correlation matching with the background noise feature library, calculates the background noise dynamic matching degree index under each sliding window, and selects windows with matching degree index higher than the first threshold as strong background noise windows. The steady-state interference extraction module is used to construct a background noise reference signal based on the second time-domain waveform data of the strong background noise window, subtract the background noise component in the background noise reference signal after adaptive filter processing from the second time-domain waveform data, extract the candidate device interference waveform data and segment it; obtain the energy accumulation value of each segment at each frequency point based on fast Fourier transform, and compare the fluctuation amplitude of the energy accumulation value at each frequency point with a preset threshold to filter out the steady-state interference waveform data. The out-of-standard frequency band marking module is used to perform time-frequency conversion on steady-state interference waveform data, generate the corresponding frequency domain spectrum, and compare the frequency domain spectrum with the standard limit line on a frequency-by-frequency basis to mark the out-of-standard frequency band and its corresponding time domain waveform segment. The interference source identification module is used to extract time-frequency joint features from the time-domain waveform segments corresponding to the frequency bands exceeding the standard, construct the feature vector exceeding the standard, input the feature vector exceeding the standard into the preset classification model, and output the interference source type identification result of the frequency band exceeding the standard.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention constructs a background noise feature library containing mean and variance, and introduces a background noise dynamic matching degree index to accurately filter strong background noise windows, thereby achieving adaptive identification and stripping of time-varying background noise on site. This solves the problem of inaccurate background noise subtraction in traditional methods and improves detection accuracy. 2. This invention uses adaptive filtering to remove background noise and combines segmented energy accumulation and fluctuation amplitude analysis to quantify the time-varying characteristics of energy at each frequency point with the coefficient of variation, thereby screening out steady-state interference waveform data and achieving effective separation of the device's own steady-state interference from external sporadic interference, thus improving the reliability of detection. 3. This invention extracts time-frequency joint features, integrates time-domain feature parameters and frequency-domain feature parameters to construct a multi-dimensional over-standard feature vector, inputs it into a machine learning classification model, and outputs the interference source type identification result, thereby realizing the automatic classification and identification of over-standard interference sources and providing clear guidance for electromagnetic compatibility rectification. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 For frequency point Dual Y-axis images of power spectral density exceeding standard limits; Figure 3 This is a schematic diagram of the overall system modules of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0018] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] Example: Please see Figures 1-2 The present invention provides a technical solution: An electromagnetic compatibility testing method, comprising the following steps: Step 1: With the device under test powered off, collect the first time-domain waveform data of the detection point within a preset time period and perform time-frequency analysis to generate a background noise frequency domain feature set. Calculate the energy distribution variance of each frequency point in the background noise frequency domain feature set and construct a background noise feature library. In this embodiment, the specific logic for generating the background noise frequency domain feature set is as follows: when the device under test is not powered on and the on-site environment is consistent with the actual test conditions, electromagnetic sensors arranged at the detection points are used to sample at a preset frequency. Continuous data collection for a preset duration The time-domain electromagnetic signal within the range is used to obtain the first time-domain waveform data. ,in The sampling frequency is set according to the electromagnetic interference frequency range of the device under test. The typical value is more than 2.5 times the highest frequency of interest of the device under test. Preferably, the sampling frequency is set to 10 GHz or higher to ensure that high-frequency components do not alias. The preset duration is determined according to the time-varying characteristics of the background noise at the site. The typical duration is not less than 10 seconds to ensure the sufficiency of the statistical samples. The first time-domain waveform data According to the preset window length and sliding step size Divided into The time-domain segment consists of several time-domain segments. The window length is determined based on the required frequency resolution, typically 1024, 2048, or 4096 sampling points, corresponding to the sampling frequency divided by the window length. The sliding step size is set to 50% to 75% of the window length, resulting in a 25% to 50% overlap between adjacent time-domain segments to enhance the continuity and statistical validity of the time-domain analysis. It should be noted that if the sliding step size is smaller than the window length, there will inevitably be overlap between adjacent segments.

[0020] After applying a Hanning or Hamming window to each time-domain segment, a Fast Fourier Transform is performed to obtain the frequency domain amplitude spectrum of each time-domain segment. The power spectral density of each time-domain segment is then calculated. The power spectral densities of all segments are aggregated according to frequency points to obtain the background noise frequency domain feature set. This background noise frequency domain feature set contains power spectral density information for each time domain segment at each frequency point, comprehensively characterizing the energy distribution characteristics of background noise in the time and frequency domain, and providing a data foundation for subsequent statistical modeling of background noise; in Indicates the first One frequency point, For frequency point index, For the index of the time domain segment, Indicates the first The time-domain segment in the first Power spectral density values ​​at each frequency point; The requirement that the on-site environment be kept consistent with the actual test conditions means that the average power spectral density of the electromagnetic noise in the on-site environment does not differ from the background noise acquisition period to the equipment power-on test period. Furthermore, the operating status of the main environmental interference sources (such as nearby operating equipment and power grid harmonic sources) remains consistent. Before collecting background noise, a 5-minute environmental noise pre-scan is performed, and the environmental noise spectrum is recorded as a reference. If the power spectral density of any frequency point changes more than 3 dB relative to the reference during the formal acquisition period, the data for that period is marked as invalid and re-acquired.

[0021] For the background noise frequency domain feature set For each frequency point in the time domain, calculate its energy distribution variance over the entire time domain segment. The calculation formula is as follows: in, For the first frequency points The average power spectral density over all time domain segments is used to characterize the average energy level of background noise at that frequency. For the first frequency points The energy distribution variance is used to characterize the degree of fluctuation of the background noise at this frequency point in different time domain segments; This represents the number of time-domain segments. The energy distribution variance is used to characterize the degree of energy fluctuation of background noise at a given frequency. A smaller energy distribution variance indicates that the background noise energy at that frequency is more stable and fluctuates less across different time segments, resulting in better consistency of the background noise at that frequency. Conversely, a larger energy distribution variance indicates that the background noise energy at that frequency changes more drastically over time, resulting in poorer consistency of the background noise at that frequency. Based on this characteristic, the energy distribution variance is used as the normalized denominator in the subsequent calculation of the background noise dynamic matching degree index. For frequencies with smaller energy distribution variance (i.e., frequencies with stable noise), the matching degree calculation is more sensitive to the deviation between the measured signal and the background noise, enabling more accurate identification of weak interference. For frequencies with larger energy distribution variance (i.e., frequencies with large noise fluctuations), the matching degree calculation automatically reduces its sensitivity to avoid misjudgment due to the drastic fluctuations of the background noise itself. This formula is the core of constructing the background noise feature library. It provides statistical features in two dimensions—mean and variance—for each frequency point, ensuring that the background noise feature library includes not only average energy information but also time-varying fluctuation information, thus providing a quantified noise benchmark for subsequent sliding window matching and adaptive filtering.

[0022] The mean power spectral density and the variance of energy distribution at each frequency point are used together as background noise feature parameters to construct a background noise feature library. , This represents the number of frequency points.

[0023] Step 2: When the device under test is powered on and in typical operating conditions, collect the second time domain waveform data of the detection point within the same time period and perform time-frequency analysis. Then, perform sliding window correlation matching with the background noise feature library, calculate the background noise dynamic matching degree index under each sliding window, and select the window with the matching degree index higher than the first threshold as the strong background noise window. In this embodiment, the device under test is powered on and placed under typical operating conditions. Electromagnetic sensors arranged at the same detection points are used to sample at the same frequency. Collect data for the same duration continuously The time-domain electromagnetic signal within is used to obtain the second time-domain waveform data. ; The typical operating conditions refer to the operation of the device under test (DUT) under its rated operating conditions as specified in the design or under its most representative operating conditions in actual use. Specifically, this includes: for frequency converters, the rated speed and rated load conditions; for switching power supplies, the rated input voltage and rated output current conditions; and for communication equipment, the maximum transmit power conditions. The sampling frequency is consistent with that of the background noise acquisition phase, i.e., not less than 2.5 times the highest frequency of interest of the DUT, preferably 10 GHz or higher, to ensure that high-frequency interference components do not aliasing. The acquisition duration is the same as that of the background noise acquisition phase, with a typical duration of not less than 10 seconds, to ensure that the background noise variation pattern and the device's operating cycle are fully covered.

[0024] The second time-domain waveform data According to the preset window length and sliding step size Divided into There are several time-domain windows; the window length and sliding step size are set exactly the same as in the background noise acquisition stage: the window length is determined according to the required frequency resolution, with typical values ​​of 1024, 2048 or 4096 sampling points; the sliding step size is set to 50% to 75% of the window length, so that there is an overlap rate of 25% to 50% between adjacent time-domain windows. The purpose of keeping the window length and sliding step size consistent with the background noise stage is to ensure that the frequency domain resolution of each time-domain window corresponds one-to-one with the frequency points in the background noise feature library during subsequent matching calculations, avoiding frequency point misalignment or resolution mismatch due to different window parameters.

[0025] For the The second time-domain waveform data segment within a time-domain window Perform a Fast Fourier Transform to obtain the time-domain window at the [number]th [time domain]. Power spectral density at each frequency point ;in, For the index of the time domain window; For each second time-domain waveform data segment within a time-domain window, before performing the Fast Fourier Transform (FFT), a window function of the same type as that used in the background noise acquisition stage (preferably a Hanning or Hamming window) is applied to reduce spectral leakage. Subsequently, the FFT is performed to obtain the frequency domain amplitude spectrum of that time-domain window, and the power spectral density at each frequency point is calculated accordingly. The power spectral density is calculated by squaring the amplitude value at each frequency point in the frequency domain amplitude spectrum and dividing by the ratio of the window length to the sampling frequency.

[0026] It should be noted that for any given time-domain window, its time domain window is... The physical meaning of the power spectral density at each frequency point is: the signal's power spectral density during that time period. Average power within a unit bandwidth in the vicinity.

[0027] Calculate the first Background noise dynamic matching index at all frequency points for each time window The background noise dynamic matching degree index is used to characterize the first The overall similarity in the frequency domain between the measured signal and the background noise feature library within each time window, with a range of values. The closer the exponent value is to 1, the closer the frequency domain characteristics of the measured signal within the window are to the background noise, meaning that the influence of the equipment's own interference is smaller and the proportion of background noise is higher during this period. The specific calculation process is as follows: For each frequency point, firstly, the power spectral density of the measured signal within the time-domain window is subtracted from the mean power spectral density of the corresponding frequency point in the background noise feature library to obtain the difference; then, the difference is squared to obtain the energy deviation square term at that frequency point; the energy deviation square term reflects the degree of deviation of the measured signal's power spectral density at that frequency point from the average level of the background noise. The larger the deviation, the more significant the influence of the frequency point on the interference of the equipment itself or the sudden change in the environment.

[0028] Simultaneously, the energy distribution variance at this frequency point is multiplied by 2 and then a preset regularization parameter is added as the normalized denominator. The energy distribution variance characterizes the inherent fluctuation of background noise at this frequency point: the smaller the variance, the more stable the background noise at that frequency point, and the higher the sensitivity to deviation should be; the larger the variance, the more drastic the fluctuation of the background noise itself, and the lower the sensitivity to deviation should be accordingly. The regularization parameter is a positive decimal greater than zero, serving two purposes: first, to prevent the denominator from being zero when the energy distribution variance is zero; second, to adjust the overall sensitivity of the matching degree calculation. The larger the regularization parameter value, the smoother the decay of the matching degree index to deviation, and vice versa. The typical range of the regularization parameter value is 1% to 10% of the mean background noise power spectral density. Dividing the squared energy deviation term by the normalized denominator yields the normalized deviation metric at that frequency. Its physical meaning is: using the fluctuation of background noise as a scale, it measures the relative degree to which the measured signal deviates from the average level of background noise. Then, the normalized deviation metric is negatively evaluated and subjected to a natural exponential transform to obtain the local matching degree at that frequency. The natural exponential transform transforms the normalized deviation metric from... Mapped to Interval: When the deviation is zero, the exponent term is 1, indicating a perfect match; as the deviation increases, the exponent term rapidly decays and approaches 0, indicating an extremely low degree of match. After obtaining all After determining the local matching degree at each frequency point, the arithmetic mean of the local matching degrees at all frequency points is calculated to obtain the background noise dynamic matching degree index for that time-domain window. The purpose of using the arithmetic mean instead of the geometric mean or other weighting methods is to avoid excessively lowering the overall matching degree due to large deviations at individual frequency points, and to ensure that windows dominated by background noise can be effectively identified.

[0029] The calculation formula is: In the formula, These are preset regularization parameters used to prevent the denominator from being zero and to adjust the sensitivity of the matching degree calculation; It is a natural exponential function; The calculated background noise dynamic matching degree index for each time domain window is compared with a preset first threshold. Compare and filter out those with a background noise dynamic matching degree index greater than or equal to the first threshold. The time-domain window is used as a strong background noise window.

[0030] The first threshold is an empirical value between 0 and 1, with a typical range of 0.6 to 0.8. The specific value depends on the complexity of the background noise and the required detection accuracy: the more complex and fluctuating the background noise, the lower the threshold can be, for example, to 0.6, in order to retain more background noise samples; the purer the background noise, the higher the threshold can be, for example, to 0.8, in order to improve the strictness of the screening.

[0031] The measured signal within the strong background noise window is highly similar to the background noise characteristics, indicating that the electromagnetic interference energy generated by the device itself during this period is much lower than the background noise, and the main signal is still the ambient noise. Therefore, it is suitable as a background noise reference signal source for subsequent adaptive filtering.

[0032] Step 3: Construct a background noise reference signal based on the second time-domain waveform data of the strong background noise window. Subtract the background noise component from the background noise reference signal after adaptive filter processing from the second time-domain waveform data to extract the candidate device interference waveform data and segment it. Obtain the energy accumulation value of each segment at each frequency point based on the fast Fourier transform, and compare the fluctuation amplitude of the energy accumulation value at each frequency point with the preset threshold to select the steady-state interference waveform data. In this embodiment, the second time-domain waveform data segments corresponding to each strong background noise window are extracted. It should be noted that these strong background noise windows may not be continuous on the time axis, and there are intervals between windows (i.e., they are separated by non-strong background noise windows). During extraction, only the waveform data segments within the window are extracted, and the interval data between windows are not used. The second time-domain waveform data segments of all strong background noise windows are spliced ​​together in chronological order to form a continuous time-series signal, which is the background noise reference signal. An adaptive filter is constructed based on the statistical correlation between the background noise reference signal and the second time-domain waveform data. This adaptive filter employs a transverse filter structure, and its order is determined by the correlation length of the background noise, typically ranging from 32 to 256. A higher order enhances the filter's ability to model complex noise, but also increases the computational load. The filter's input is the background noise reference signal, and the target output is the optimal estimate of the background noise components in the second time-domain waveform.

[0033] The coefficients of the adaptive filter are iteratively updated using the minimum mean square error criterion. Specifically, the second time-domain waveform data is used as the desired response. The output of the adaptive filter (i.e., the estimated background noise component) is subtracted from the desired response to obtain the error signal. This error signal reflects the unfiltered components in the second time-domain waveform, which theoretically should mainly be electromagnetic interference generated by the equipment itself. The goal of the minimum mean square error criterion is to minimize the mean square value of the error signal, that is, to continuously adjust the filter coefficients so that the estimated background noise component approximates as closely as possible to the actual background noise component contained in the second time-domain waveform.

[0034] The filter coefficients are iteratively updated using the least mean square algorithm, with the update formula: New coefficient = Old coefficient + Step size parameter × Error signal × Current input signal. The step size parameter is a positive decimal, typically ranging from 0.001 to 0.01, used to control the speed and stability of coefficient updates: a larger step size results in faster convergence but decreased stability; a smaller step size results in slower convergence but smaller steady-state error. To ensure the stability and convergence of the filter, the step size parameter must satisfy the constraint that it is less than the reciprocal of the input signal power. The iterative update process continues until the filter coefficients converge (i.e., the coefficient change between two adjacent iterations is less than a preset threshold, typically 0.001) or the preset maximum number of iterations is reached (typically 2000). The background noise reference signal is filtered using an updated adaptive filter to obtain an estimated background noise component. The background noise component is then subtracted from the second time-domain waveform data to obtain candidate device interference waveform data. In this candidate device interference waveform data, the main component of the on-site background noise has been removed, and the remaining part mainly includes the electromagnetic interference signal generated by the device under test itself, as well as a small amount of residual background noise that could not be completely removed due to the non-ideal characteristics of the filter. The candidate device interference waveform data will be used for subsequent segmented energy accumulation and steady-state interference screening analysis.

[0035] The candidate device interference waveform data is divided into time series. A series of equal-length segments, the first... The time intervals corresponding to each segment are: , Represents a continuous-time variable, and For any segment Perform a Fast Fourier Transform to calculate the cumulative energy value of this segment at each frequency point. The calculation formula is as follows: in, Indicates the first The frequency point at the first The cumulative energy value across each segment; and They represent the first The start and end times of each segment; This represents the modulo-square operation; This represents the interference waveform data of the candidate device.

[0036] This formula is used to calculate the cumulative energy value of the interference waveform data of the candidate device at each frequency point within each equal time segment. Specifically, for the _____, The process involves dividing the signal into time segments. First, the time-domain interference signal within each segment is multiplied by a complex exponential kernel function. Then, integration is performed from the start to the end of each segment. This integration is essentially a finite-time continuous Fourier transform, decomposing the time-domain signal into different frequency components. Finally, the integral result is squared modulo-based to obtain the cumulative energy value of that segment at that frequency. The physical meaning of this cumulative energy value is: the interference waveform of the candidate device at the [missing information - likely a specific time interval or frequency]... Frequency within a time period The signal energy within a unit bandwidth is a local measure of interference energy in the time-frequency plane. The larger the value, the stronger the electromagnetic radiation energy generated by the device under test at that time period and frequency, and the more likely it is to become a major source of electromagnetic interference or a contributor exceeding the standard; the smaller the value, the lower the electromagnetic emission level of the device in that local time-frequency region.

[0037] The logic of this formula is as follows: First, by dividing the signal into equal-length segments and calculating the energy accumulation separately, the time-varying characteristics of the interference signal energy can be captured, avoiding the masking of transient or intermittent interference due to long-term averaging. Second, the use of modulo-square operation to convert the frequency domain complex result into a real energy value conforms to the physical convention of using power or energy as an evaluation index in electromagnetic compatibility testing. Third, the complex exponential kernel function in the integral transform ensures the orthogonal decomposition capability of the energy accumulation value for different frequency components, allowing the energy contribution at different frequency points to be analyzed independently. Fourth, this energy accumulation value directly provides a quantitative basis for subsequent fluctuation amplitude calculation and steady-state interference screening. That is, if the energy accumulation value of a certain frequency point remains stable in each segment (i.e., the fluctuation amplitude is small), it indicates that the interference at that frequency point is a continuous steady-state interference, which is worth further analysis; if the fluctuation is severe, it may be an occasional interference or residual background noise.

[0038] For each frequency point, calculate the fluctuation range of the cumulative energy value across all segments. The calculation formula is as follows: in, Indicates the first frequency points The fluctuation range, For the first frequency points The average cumulative energy over the entire segment For the first frequency points The cumulative fluctuation range of energy; the larger the value, the more drastic the change in interference energy over time at that frequency. This formula is used to calculate the fluctuation amplitude of the cumulative energy value of each frequency point across all segments. Essentially, it is a quantitative representation of the normalized time-varying dispersion of the interference energy at each frequency point. The specific calculation process consists of two steps: First, for the... Each frequency point is used to calculate its position in the entire... The arithmetic mean of the cumulative energy values ​​over several time segments is used as the baseline energy level for interference at that frequency. Next, the dispersion of the cumulative energy values ​​for each segment relative to this baseline level is calculated. This involves first calculating the deviation of each segment's cumulative energy value from the mean and squaring it. Then, the squares of the squared deviations from all segments are summed and divided by the number of segments minus one to obtain the variance. The square root of the variance is then taken to obtain the standard deviation. Finally, this standard deviation is divided by the baseline energy level to obtain a dimensionless ratio, which serves as the fluctuation amplitude at that frequency. The physical meaning of this fluctuation amplitude is: in the form of a relative coefficient of variation, it reflects the stability of the candidate device's interference waveform energy over time at that frequency. The larger the value, the more drastic the fluctuations in the interference energy at that frequency across different time segments, indicating that the electromagnetic emission at that frequency exhibits intermittent, sudden, or highly time-varying characteristics, possibly originating from transient interference, unsteady operating conditions, or residual background noise fluctuations. The smaller the value, the more consistent the interference energy at that frequency across segments, indicating that the electromagnetic emission at that frequency has continuity and stability, belonging to typical steady-state interference.

[0039] The logic of this formula is as follows: First, by using the ratio of standard deviation to mean (i.e., coefficient of variation), the influence of the absolute magnitude difference in interference energy between different frequency points on the evaluation of fluctuation level is eliminated, allowing the time-varying characteristics of strong and weak energy frequency points to be compared and thresholded on the same scale. Second, the denominator uses the mean instead of a fixed constant, ensuring the normalization of the dimensions of the fluctuation amplitude index, which is consistent with the actual scenario in electromagnetic compatibility testing where the interference levels at different frequency points may differ by several orders of magnitude. Third, this fluctuation amplitude directly serves the purpose of screening steady-state interference. Frequency points with small fluctuation amplitudes correspond to continuous steady-state interference, which is the electromagnetic radiation continuously generated under the normal operating conditions of the equipment itself, and should be the main targets for subsequent over-limit judgment and interference source identification. Frequency points with large fluctuation amplitudes are excluded to avoid misjudging external sporadic interference or residual background noise as the steady-state emission of the equipment itself.

[0040] Will With the preset second threshold Compare and filter those that meet the requirements. The frequency point is used as the steady-state interference frequency point, and the candidate device interference waveform data corresponding to the steady-state interference frequency point is extracted to obtain the steady-state interference waveform data.

[0041] Second threshold The first threshold is a dimensionless positive number used to determine the time-varying stability of interference energy at each frequency point, with a typical value range of [0.1, 0.3]. The specific value of the second threshold is determined based on the type of device under test and the requirements of the on-site test: for scenarios requiring high stability testing (such as product type certification testing), the threshold can be set to a smaller value, such as 0.1, to filter out the most stable interference frequency points; for scenarios with more complex on-site environments and larger fluctuations in background noise, the threshold can be appropriately relaxed to 0.2 or 0.3 to retain more effective interference information. The second threshold can be pre-calibrated by the following methods: in a standard anechoic chamber environment, test a standard device with known interference characteristics, calculate the fluctuation amplitude distribution of each frequency point, and take the 90th percentile of the fluctuation amplitude of all frequency points as the reference threshold; or use the empirical default value of 0.2, which has been verified by a large number of on-site tests and can effectively distinguish between steady-state interference and non-steady-state interference in most scenarios.

[0042] The physical meaning of this screening condition is that only those frequency points with relatively stable energy accumulation values ​​and fluctuation amplitudes not exceeding the preset tolerance limit in each time segment are considered as continuous steady-state interference generated by the device itself; frequency points with fluctuation amplitudes greater than the second threshold are considered as non-steady-state interference or residual background noise and are excluded.

[0043] Step 4: Perform time-frequency conversion on the steady-state interference waveform data to generate the corresponding frequency domain spectrum, and compare the frequency domain spectrum with the standard limit line on a frequency-by-frequency basis to mark the frequency bands that exceed the standard and their corresponding time domain waveform segments. In this embodiment, the steady-state interference waveform data is converted from time to frequency to generate a corresponding frequency domain spectrum. The frequency domain spectrum is then compared with the standard limit line on a frequency-by-frequency basis to mark the frequency bands exceeding the limit and their corresponding time domain waveform segments. The specific logic behind this is as follows: The steady-state interference waveform data undergoes time-frequency conversion. Before the time-frequency conversion, the transformation window length is preset according to the requirements of the test standard and the type of the device under test. The typical value of the transformation window length is consistent with the window length in step 1, i.e., 1024, 2048, or 4096 sampling points, to ensure the consistency of frequency domain resolution. After applying the same type of window function (preferably Hanning or Hamming window) to the steady-state interference waveform data as in the previous steps, a fast Fourier transform is performed to convert the time-domain signal into a frequency-domain signal, obtaining the frequency domain spectrum of the steady-state interference waveform data. This frequency domain spectrum contains the power spectral density values ​​corresponding to each frequency point. Obtain standard limit lines corresponding to the type of equipment under test and the test standard, wherein the standard limit lines include electromagnetic radiation limits corresponding to each frequency point; Obtain the standard limit curves corresponding to the type of device under test and the test standard. The standard limit curves are reference curves generated based on the radiated or conducted emission limits specified in national electromagnetic compatibility standards, international standards, or industry-specific standards. The standard limit curves contain the electromagnetic radiation limits corresponding to each frequency point, typically expressed in logarithmic coordinates, with units of [unit missing]. (Radiation emission) or (Conducted emission). Different equipment types and test levels correspond to different limit lines. For example, information technology equipment is subject to the GB / T 9254 standard, industrial, scientific and medical equipment is subject to the GB 4824 standard, and electric vehicles are subject to the GB / T 18387 standard. The standard limit lines can be obtained by looking up a table, that is, interpolating the frequency band segmented limit values ​​given in the standard onto the same frequency point grid as the frequency domain spectrum to form a frequency point-by-frequency limit sequence.

[0044] The frequency domain spectrum is compared with the standard limit line frequency by frequency. Specifically, for each frequency point, the power spectral density value of the steady-state interference waveform at that frequency point is read, and the corresponding standard electromagnetic radiation limit value is also read. The magnitudes of the two are compared. During the comparison, attention must be paid to the consistency of units. If the power spectral density value is in linear power units (e.g., ...), ... The standard limit is expressed in logarithmic units (e.g., logarithmic units). If the expression is not provided, then a unit conversion must be performed beforehand to unify the two into the same physical quantity and the same unit before comparison. The unit conversion formula is determined based on parameters such as antenna factor, test distance, and measurement bandwidth. The specific conversion method is a well-known technology in this field and will not be elaborated here.

[0045] When the power spectral density value of the steady-state interference waveform at a certain frequency point is greater than the electromagnetic radiation limit corresponding to that frequency point, that frequency point is marked as an out-of-standard frequency point. A frequency band consisting of consecutive out-of-range frequencies is defined as an out-of-range frequency band. The criterion for determining consecutive out-of-range frequencies is that the frequency interval between two adjacent out-of-range frequencies does not exceed twice the frequency resolution. If the out-of-range frequencies are discretely distributed, each isolated out-of-range frequency constitutes a separate out-of-range frequency band. For each out-of-range frequency band, its start frequency, end frequency, and center frequency are recorded. Based on the inverse time-frequency transformation relationship, the time-domain waveform segment corresponding to the exceeding frequency band is determined. Specifically, the method is as follows: First, based on the start and end frequencies of the exceeding frequency band, and combined with the phase information within that frequency band in the frequency domain spectrum (this phase information is retained during the Fast Fourier Transform), an inverse Fast Fourier Transform is performed on the frequency domain data within that band to obtain the time-domain waveform segment corresponding to the exceeding frequency band. If the original steady-state interference waveform data is divided into multiple segments in the time domain, the time-domain waveform segment corresponding to the exceeding frequency band may be distributed across multiple time segments. In this case, the time-domain waveform segments of all relevant segments are extracted and spliced ​​together in chronological order to form a complete time-domain waveform segment corresponding to the exceeding frequency band, which is used for subsequent time-frequency joint feature extraction and interference source type identification.

[0046] Step 5: Perform joint time-frequency feature extraction on the time-domain waveform segments corresponding to the frequency bands exceeding the standard, construct the feature vector of exceeding the standard, input the feature vector of exceeding the standard into the preset classification model, and output the identification result of the interference source type of the frequency band exceeding the standard. In this embodiment, time-frequency joint features are extracted from the time-domain waveform segments corresponding to the frequency exceeding the standard band. The time-frequency joint features include time-domain feature parameters and frequency-domain feature parameters. Among them, the time-domain feature parameters include peak amplitude, pulse width, rise time, fall time, and zero-crossing rate. Peak amplitude refers to the maximum amplitude value in the time-domain waveform segment, which is used to characterize the strength of the interference signal. Pulse width refers to the time interval between the 50% amplitude point of the rising edge and the 50% amplitude point of the falling edge of the interference pulse, which is used to distinguish between narrow pulse interference and wide pulse interference. Rise time refers to the time required for the pulse leading edge to rise from 10% amplitude to 90% amplitude, and fall time refers to the time required for the pulse trailing edge to fall from 90% amplitude to 10% amplitude. Both reflect the steepness of the interference signal. Zero-crossing rate refers to the ratio of the number of times the time-domain waveform crosses the zero level to the time length, which is used to characterize the oscillation frequency characteristics of the signal. Frequency domain characteristic parameters include center frequency, bandwidth, harmonic order distribution, and spectral envelope shape characteristics. Center frequency refers to the geometric center frequency or energy centroid frequency of the out-of-standard frequency band, used to locate the main frequency position of interference. Bandwidth refers to the frequency width of the out-of-standard frequency band, used to distinguish between narrowband interference and broadband interference. Harmonic order distribution refers to the integer multiple relationship between the out-of-standard frequency point and the fundamental frequency, used to identify harmonic interference generated by switching power supply or frequency converter drive. Spectral envelope shape characteristics refer to the overall contour shape of the out-of-standard frequency band in the frequency domain spectrum, including single-peak, multi-peak, flat, or attenuated shapes, used to distinguish the spectral characteristics of different interference sources.

[0047] The extracted time-domain and frequency-domain feature parameters are fused to construct a super-standard feature vector. The fusion method involves arranging all the aforementioned time-domain and frequency-domain feature parameters in a fixed order to form a multi-dimensional numerical vector. Each feature parameter occupies one dimension in the vector, and the total number of dimensions of the feature vector equals the number of extracted feature parameters. Before constructing the feature vector, each feature parameter can be normalized by subtracting the mean of that feature from the training sample set and then dividing by the standard deviation. This eliminates differences in units and orders of magnitude between different features, enabling the classification model to utilize the information from each feature in a balanced manner.

[0048] The excess feature vector is input into a preset classification model; The classification model is a pre-trained machine learning classification model. Its training process is as follows: First, a large number of historical out-of-specification feature vectors of known interference source types are collected as training samples. Each training sample is labeled with a corresponding interference source type. Then, supervised learning algorithms are used to train these training samples. Commonly used supervised learning algorithms include support vector machines, random forests, gradient boosting decision trees, or lightweight gradient boosting machines. After training, a mapping relationship from feature vectors to interference source types is established within the classification model. The interference source types include switching power supply interference, motor drive interference, wireless communication interference, and electrostatic discharge interference. Switching power supply interference typically exhibits spectral characteristics with periodic spike pulses and high-order harmonics; motor drive interference typically exhibits modulation sidebands and broadband noise characteristics related to motor speed; wireless communication interference typically exhibits spectral characteristics of a single carrier frequency or narrowband modulation signal; and electrostatic discharge interference typically exhibits transient characteristics with extremely short rise times, wide bandwidth, and rapid energy decay.

[0049] After processing by the classification model, the system outputs the identification results of the interference source types in the frequency bands exceeding the standard, thus completing the electromagnetic compatibility test. This identification result represents the most likely source type of electromagnetic interference exceeding the standard in the tested equipment, providing a clear directional basis for subsequent electromagnetic compatibility rectification.

[0050] Table 1: Statistical Results of Key Parameters for Electromagnetic Compatibility Testing at Different Frequency Points This study is based on electromagnetic compatibility test data from 15 frequency points (30-310MHz) in Table 1, combined with... Figure 2 Analysis showed that the mean background noise was between The variance ranged from 0.79 to 2.13, showing a stable distribution with minimal fluctuations, indicating stable background noise characteristics and a well-established background noise feature library. The background noise dynamic matching index ranged from 0.65 to 0.88, with high matching degrees at most frequencies, effectively filtering out strong background noise windows and providing a reliable basis for adaptive filtering to remove environmental noise. The energy fluctuation amplitude at each frequency ranged from 0.08 to 0.26, all below the steady-state interference threshold, indicating that the extracted interference waveforms were all steady-state interference generated by the tested equipment itself, effectively eliminating the influence of transient interference and residual noise, making the interference extraction results accurate and reliable. Comparison with standard limits reveals that the power spectral density of the tested device exceeds the standard limits at frequency points of 30, 50, 90, 130, 150, 190, 210, 230, 270, 290, and 310 MHz. The most serious exceedance reached [a certain level]. The frequency bands exceeding the standard are mainly concentrated in and In the mid-to-high frequency range, the exceeding of standards truly reflects the electromagnetic compatibility issues of the equipment itself. It also verifies that this detection method can effectively remove background noise, accurately extract steady-state interference, and precisely locate the exceeding frequency band in complex field environments, demonstrating high detection accuracy and practicality.

[0051] Please see Figure 3 An electromagnetic compatibility testing system, comprising: The background noise library module is used to collect the first time-domain waveform data of the detection point within a preset time period when the device under test is not powered on, perform time-frequency analysis, generate a background noise frequency domain feature set, calculate the energy distribution variance of each frequency point in the background noise frequency domain feature set, and construct a background noise feature library. The strong noise window matching module is used to collect second time-domain waveform data of the detection point within the same time period and perform time-frequency analysis when the device under test is powered on and in typical operating conditions. Then, it performs sliding window correlation matching with the background noise feature library, calculates the background noise dynamic matching degree index under each sliding window, and selects windows with matching degree index higher than the first threshold as strong background noise windows. The steady-state interference extraction module is used to construct a background noise reference signal based on the second time-domain waveform data of the strong background noise window, subtract the background noise component in the background noise reference signal after adaptive filter processing from the second time-domain waveform data, extract the candidate device interference waveform data and segment it; obtain the energy accumulation value of each segment at each frequency point based on fast Fourier transform, and compare the fluctuation amplitude of the energy accumulation value at each frequency point with a preset threshold to filter out the steady-state interference waveform data. The out-of-standard frequency band marking module is used to perform time-frequency conversion on steady-state interference waveform data, generate the corresponding frequency domain spectrum, and compare the frequency domain spectrum with the standard limit line on a frequency-by-frequency basis to mark the out-of-standard frequency band and its corresponding time domain waveform segment. The interference source identification module is used to extract time-frequency joint features from the time-domain waveform segments corresponding to the frequency bands exceeding the standard, construct the feature vector exceeding the standard, input the feature vector exceeding the standard into the preset classification model, and output the interference source type identification result of the frequency band exceeding the standard.

[0052] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0053] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0054] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0055] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An electromagnetic compatibility testing method, characterized in that, The specific steps include: Step 1: With the device under test powered off, collect the first time-domain waveform data of the detection point within a preset time period and perform time-frequency analysis to generate a background noise frequency domain feature set. Calculate the energy distribution variance of each frequency point in the background noise frequency domain feature set and construct a background noise feature library. Step 2: When the device under test is powered on and in typical operating conditions, collect the second time domain waveform data of the detection point within the same time period and perform time-frequency analysis. Then, perform sliding window correlation matching with the background noise feature library, calculate the background noise dynamic matching degree index under each sliding window, and select the window with the matching degree index higher than the first threshold as the strong background noise window. Step 3: Construct a background noise reference signal based on the second time-domain waveform data of the strong background noise window. Subtract the background noise component from the background noise reference signal after adaptive filter processing from the second time-domain waveform data to extract the candidate device interference waveform data and segment it. Obtain the energy accumulation value of each segment at each frequency point based on the fast Fourier transform, and compare the fluctuation amplitude of the energy accumulation value at each frequency point with the preset threshold to select the steady-state interference waveform data. Step 4: Perform time-frequency conversion on the steady-state interference waveform data to generate the corresponding frequency domain spectrum, and compare the frequency domain spectrum with the standard limit line on a frequency-by-frequency basis to mark the frequency bands that exceed the standard and their corresponding time domain waveform segments. Step 5: Perform joint time-frequency feature extraction on the time-domain waveform segments corresponding to the frequency bands exceeding the standard, construct the feature vector of exceeding the standard, input the feature vector of exceeding the standard into the preset classification model, and output the interference source type identification result of the frequency band exceeding the standard.

2. The electromagnetic compatibility testing method according to claim 1, characterized in that: The specific logic underlying the generation of the background noise frequency domain feature set is as follows: when the device under test is in a de-energized state, electromagnetic sensors arranged at the detection points are used to sample at a preset frequency. Continuous data collection for a preset duration The time-domain electromagnetic signal within the range is used to obtain the first time-domain waveform data. ,in The index of the sampling point; the first time-domain waveform data According to the preset window length and sliding step size Divided into The time-domain segments are divided into several time-domain segments, with a sliding step size smaller than the window length, resulting in overlap between adjacent time-domain segments. A Fast Fourier Transform is performed on each time-domain segment to obtain its frequency-domain amplitude spectrum, and the power spectral density of each time-domain segment is calculated. The power spectral densities of all segments are aggregated according to frequency points to obtain the background noise frequency-domain feature set. ,in Indicates the first One frequency point, For frequency point index, For the index of the time domain segment, Indicates the first The time-domain segment in the first Power spectral density values ​​at each frequency point.

3. The electromagnetic compatibility testing method according to claim 2, characterized in that: For the background noise frequency domain feature set For each frequency point in the time domain, calculate its energy distribution variance over all time domain segments, specifically as follows: First, calculate the mean power spectral density of each frequency point across all time-domain segments to characterize the average energy level of that frequency point in the background noise; then calculate the average of the squares of the deviations of the power spectral density of each time-domain segment from the mean, and use this average as the variance of the energy distribution of each frequency point to characterize the degree of fluctuation of the background noise at that frequency point across different time-domain segments. Finally, the mean power spectral density and the variance of energy distribution at each frequency point are used together as background noise feature parameters to construct a background noise feature library.

4. The electromagnetic compatibility testing method according to claim 3, characterized in that: The device under test is powered on and placed under typical operating conditions. Electromagnetic sensors arranged at the same detection points are used to sample at the same frequency. Collect data for the same duration continuously The time-domain electromagnetic signal within is used to obtain the second time-domain waveform data. ; The second time-domain waveform data According to the preset window length and sliding step size Divided into One time-domain window; For the The second time-domain waveform data segment within a time-domain window Perform a Fast Fourier Transform to obtain the time-domain window at the [number]th [time domain]. Power spectral density at each frequency point ;in, For the index of the time domain window; When calculating the background noise dynamic matching degree index for each time-domain window across all frequency points, for each frequency point, the difference between the power spectral density of the measured signal within the time-domain window and the mean power spectral density of the corresponding frequency point in the background noise feature library is squared, and then divided by twice the energy distribution variance of that frequency point plus the regularization parameter to construct a normalized deviation metric. Subsequently, the deviation metric is negative and subjected to a natural exponential transformation to obtain the local matching degree at each frequency point. Finally, the arithmetic mean of the local matching degrees of all frequency points is calculated as the background noise dynamic matching degree index for that time-domain window. The calculated background noise dynamic matching degree index for each time domain window is compared with a preset first threshold. Compare and filter out those with a background noise dynamic matching degree index greater than or equal to the first threshold. The time-domain window is used as a strong background noise window.

5. The electromagnetic compatibility testing method according to claim 4, characterized in that: Extract the second time-domain waveform data segments corresponding to each strong background noise window, and splice the second time-domain waveform data segments of all strong background noise windows in time order to construct a background noise reference signal. Based on the statistical correlation between the background noise reference signal and the second time-domain waveform data, an adaptive filter is constructed. The coefficients of the adaptive filter are iteratively updated using the minimum mean square error criterion. The updated adaptive filter is used to filter the background noise reference signal to obtain the estimated background noise component. The background noise component is subtracted from the second time-domain waveform data to obtain the candidate device interference waveform data. The interference waveform data of the candidate device is divided into several time segments of equal length according to the time series. For any segment, a fast Fourier transform is performed on the time domain signal in the segment to obtain the spectral distribution of the signal at each frequency point in the segment. Then, the modulus square operation is performed on the spectral components corresponding to each frequency point to obtain the cumulative energy value of the frequency point in the segment.

6. The electromagnetic compatibility testing method according to claim 5, characterized in that: For each frequency point, calculate the fluctuation range of the cumulative energy value across all segments, specifically: First, calculate the average cumulative energy value of the frequency point across all segments as the baseline energy level; then, calculate the dispersion of the cumulative energy value of each segment relative to the baseline energy level, that is, calculate the square root of the sum of squares of the deviations of each segment divided by the number of segments minus one, and obtain the standard deviation; finally, divide the standard deviation by the baseline energy level, and the resulting ratio is the fluctuation amplitude of the frequency point. The fluctuation amplitude is compared with a preset second threshold, and frequency points that meet the condition that the fluctuation amplitude is less than or equal to the second threshold are selected as steady-state interference frequency points. The candidate device interference waveform data corresponding to the steady-state interference frequency points are extracted to obtain steady-state interference waveform data.

7. The electromagnetic compatibility testing method according to claim 1, characterized in that: The steady-state interference waveform data is converted from time to frequency to generate the corresponding frequency domain spectrum. The frequency domain spectrum is then compared with the standard limit line on a frequency-by-frequency basis to mark the frequency bands that exceed the limit and their corresponding time domain waveform segments. The specific logic behind this is as follows: The steady-state interference waveform data is subjected to time-frequency conversion processing. The steady-state interference waveform data is subjected to fast Fourier transform according to a preset transform window length to obtain the frequency domain spectrum of the steady-state interference waveform data. The frequency domain spectrum contains the power spectral density value corresponding to each frequency point. Obtain standard limit lines corresponding to the type of equipment under test and the test standard, wherein the standard limit lines include electromagnetic radiation limits corresponding to each frequency point; The frequency domain spectrum is compared with the standard limit line at each frequency point. For each frequency point, the power spectral density value is compared with the electromagnetic radiation limit. When the power spectral density value is greater than the electromagnetic radiation limit, the frequency point is marked as an out-of-limit frequency point. A frequency band consisting of consecutive out-of-standard frequency points is defined as an out-of-standard frequency band, and the time-domain waveform segment corresponding to the out-of-standard frequency band is determined based on the inverse transformation relationship of time-frequency conversion.

8. The electromagnetic compatibility testing method according to claim 7, characterized in that: Time-frequency joint features are extracted from the time-domain waveform segments corresponding to the frequency bands exceeding the standard. The time-frequency joint features include time-domain feature parameters and frequency-domain feature parameters. The time-domain feature parameters include peak amplitude, pulse width, rise time, fall time and zero-crossing rate. The frequency-domain feature parameters include center frequency, bandwidth, harmonic order distribution and spectral envelope shape features. The extracted time-domain feature parameters and frequency-domain feature parameters are fused to construct the super-standard feature vector; The excess feature vector is input into a preset classification model, which is a pre-trained machine learning classification model. The machine learning classification model is trained by a supervised learning algorithm using historical excess feature vectors labeled with interference source type as training samples. The interference source types include switching power supply interference type, motor drive interference type, wireless communication interference type and electrostatic discharge interference type. After processing by the classification model, the results of identifying the interference source type in the frequency band exceeding the standard are output, thus completing the electromagnetic compatibility test.

9. An electromagnetic compatibility testing system for performing an electromagnetic compatibility testing method according to any one of claims 1-8, characterized in that, include: The background noise library module is used to collect the first time-domain waveform data of the detection point within a preset time period when the device under test is not powered on, perform time-frequency analysis, generate a background noise frequency domain feature set, calculate the energy distribution variance of each frequency point in the background noise frequency domain feature set, and construct a background noise feature library. The strong noise window matching module is used to collect second time-domain waveform data of the detection point within the same time period and perform time-frequency analysis when the device under test is powered on and in typical operating conditions. Then, it performs sliding window correlation matching with the background noise feature library, calculates the background noise dynamic matching degree index under each sliding window, and selects windows with matching degree index higher than the first threshold as strong background noise windows. The steady-state interference extraction module is used to construct a background noise reference signal based on the second time-domain waveform data of the strong background noise window, subtract the background noise component in the background noise reference signal after the adaptive filter is processed from the second time-domain waveform data, and extract the candidate device interference waveform data and segment it. The cumulative energy value of each segment at each frequency point is obtained based on the fast Fourier transform, and the fluctuation amplitude of the cumulative energy value at each frequency point is compared with a preset threshold to filter out steady-state interference waveform data. The out-of-standard frequency band marking module is used to perform time-frequency conversion on steady-state interference waveform data, generate the corresponding frequency domain spectrum, and compare the frequency domain spectrum with the standard limit line on a frequency-by-frequency basis to mark the out-of-standard frequency band and its corresponding time domain waveform segment. The interference source identification module is used to extract time-frequency joint features from the time-domain waveform segments corresponding to the frequency bands exceeding the standard, construct the feature vector exceeding the standard, input the feature vector exceeding the standard into the preset classification model, and output the interference source type identification result of the frequency band exceeding the standard.

Images