Method and device for fast identification of electromagnetic interference sources based on spectral joint image analysis

Abstract

The present application relates to the field of electromagnetic interference source identification, and particularly relates to a kind of electromagnetic interference source fast identification method and device based on spectrum joint image analysis, greatly improve the identification efficiency of electromagnetic interference source, improve the identification accuracy of interference source.The present application provides a kind of electromagnetic interference source fast identification method based on spectrum joint image analysis, the identification method includes: according to electromagnetic data acquisition time-frequency domain joint image;The time-frequency domain joint image obtained is processed to extract multi-dimensional features;The multi-dimensional features extracted are fused to obtain a two-dimensional data matrix;Two-dimensional data matrix is transformed into image, and data-image sample is obtained;Data-image sample is placed in classifier and trained, and electromagnetic emission interference source identification model is obtained;Electromagnetic emission interference source in electromagnetic emission data is identified by electromagnetic emission interference source identification model.The present application is suitable for electromagnetic interference source identification.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic interference source identification, and specifically to a method and apparatus for rapid identification of electromagnetic interference sources based on spectrum joint image analysis. Background Technology

[0002] Modern electronic information systems are becoming increasingly complex and integrated, and electronic devices are exhibiting diversified functions. Compared to earlier, independent individual devices, modern electronic devices can combine and reuse multiple functional units. Due to differences in functional architecture, electrical signals change constantly during transmission. Furthermore, different circuit structures pose varying potential electromagnetic radiation threats. Therefore, when measuring the electromagnetic emissions of electronic devices from the outside, the acquired signal contains multiple components, which are often mixed together and difficult to distinguish upon reception.

[0003] When electromagnetic emissions are acquired using electromagnetic compatibility (EMC) testing methods, it is observed that the received time-domain signals exhibit characteristics such as time-varying, unstable waveforms, and waveform distortion. The frequency-domain signals contain numerous spectral components and suffer from issues like increased noise floor and spectral overlap. EMC test data encompasses multiple dimensions of EMC characteristics, including temporal, spatial, and vector characteristics. This multi-dimensional data is analyzed, synthesized, manipulated, and used according to certain criteria to form a consistent description of the tested object. These measurement data cover various forms of electromagnetic signals. Different EMC measurement standards have simple output requirements for measurement results, often including only voltage, current, and field strength data, which cannot reflect the inherent EMC characteristics. Individual test data is insufficient to characterize complete EMC properties. Faced with the challenge of receiving signals with diverse characteristics and the inability to directly identify signal components through observation, electromagnetic interference (EMI) source identification (EMI) techniques have emerged to distinguish signal types within electromagnetic signals. The main purpose of EMI source identification is to identify specific signal characteristics in time-varying or even nonlinear time / frequency domain data, and to classify and identify test data based on these characteristics, thereby identifying potential sources of electromagnetic interference emissions.

[0004] Electromagnetic interference (EMI) source identification primarily involves processing time / frequency domain data acquired through testing. Feature extraction and classification methods are used to identify the signal types present and associate them with specific interference sources. Currently, interference source identification methods mainly fall into two categories: time-domain and frequency-domain. Time-domain analysis utilizes an oscilloscope to acquire time-domain waveforms, while frequency spectrum analysis uses a spectrum analyzer to acquire frequency spectra as analytical data. Time-domain analysis mainly considers the amplitude and phase changes of the waveform, further considering the correlation between signals to extract signal features, and then combines this with signal characteristics within the circuit structure to identify the electromagnetic emission signal. Frequency-domain analysis extracts frequency components from the spectrum, further considering characteristics such as spectral rise, noise floor changes, and phase jitter to extract frequency domain features, and then combines this with the frequency response characteristics within the circuit structure to identify the electromagnetic emission signal.

[0005] However, current methods all have certain limitations. Time-domain analysis, due to its focus on local features, often fails to extract characteristics related to power, energy, and instantaneous frequency changes in the circuit's signal. Frequency-domain analysis, on the other hand, focuses on frequency and energy, neglecting the correlation and memory effects of signal changes, thus also resulting in some feature loss. Neither method can fully represent all the characteristics of an electrical signal. Therefore, while these methods may perform well for specific circuits or objects, their identification errors can be significant if the application conditions are not met. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and device for rapid identification of electromagnetic interference sources based on spectrum joint image analysis, which greatly improves the identification efficiency of electromagnetic interference sources, enhances the identification accuracy of interference sources, and makes the analysis of electromagnetic emission characteristics more intuitive.

[0007] The present invention achieves the above objectives by adopting the following technical solution: the present invention provides a rapid identification method for electromagnetic interference sources based on spectrum joint image analysis, the identification method comprising:

[0008] Obtain a joint time-frequency domain image based on electromagnetic data;

[0009] Multidimensional features are extracted from the acquired time-frequency domain joint image through data processing.

[0010] The extracted multidimensional features are fused to obtain a two-dimensional data matrix;

[0011] The two-dimensional data matrix is ​​transformed into an image to obtain data-image samples;

[0012] Data-image samples are fed into a classifier for training to obtain an electromagnetic emission interference source identification model.

[0013] Electromagnetic interference sources in electromagnetic emission data are identified using an electromagnetic emission interference source identification model.

[0014] Furthermore, obtaining the joint time-frequency domain image based on electromagnetic data specifically includes:

[0015] The time-domain data in the electromagnetic data is processed to obtain its time-frequency domain data, and the frequency-domain data in the electromagnetic data is obtained by inverse Fourier transform.

[0016] The time-domain data is represented as a waveform with time on the horizontal axis and amplitude on the vertical axis; the frequency-domain data is represented as amplitude-frequency and phase-frequency diagrams with frequency on the horizontal axis and power and phase on the vertical axes, respectively; the time-frequency-domain data is represented as a time-frequency-domain waterfall diagram with time on the horizontal axis and frequency on the vertical axis.

[0017] The waveform corresponding to the time domain data, the amplitude-frequency diagram and phase-frequency diagram corresponding to the frequency domain data, and the time-frequency waterfall diagram corresponding to the time-frequency domain data are combined to form a time-frequency domain joint image.

[0018] Furthermore, the data processing of the acquired time-frequency domain joint image to extract multidimensional features specifically includes:

[0019] The time-frequency data is trained using a neural network, and then fused and extracted through intermediate convolutional layers to obtain the time-frequency data features from multiple samplings.

[0020] Furthermore, the data processing and multi-dimensional feature extraction of the acquired time-frequency domain joint image also includes: periodically filtering the sampled features at different levels to obtain intermediate layer features abstracted from sampling on the network.

[0021] Furthermore, the data processing and multidimensional feature extraction from the acquired time-frequency domain joint image specifically includes:

[0022] For the original time-spectrum image data, the high-amplitude region in the foreground of the time-spectrum image is separated, and the foreground features are combined with the fine-grained features of the image to extract the time-spectrum image features using image processing algorithms.

[0023] Furthermore, the data-image samples are fed into a classifier for training, resulting in an electromagnetic emission interference source identification model, which specifically includes:

[0024] The antenna is loaded with a set interference source to conduct radiation emission. The corresponding data-image samples are obtained through the test device for training. Based on the four basic waveforms of electromagnetic emission, a large number of feature images are obtained by changing the amplitude, phase, duty cycle, modulation method, rising edge, pulse width, repetition period parameter and signal combination form of the signal. These images are used as training sample sets to train the classifier and obtain the electromagnetic emission interference source identification model.

[0025] This invention also provides a rapid electromagnetic interference source identification device based on spectrum joint image analysis, used to implement the aforementioned rapid electromagnetic interference source identification method based on spectrum joint image analysis, the identification device comprising:

[0026] The image acquisition module is used to acquire a joint time-frequency domain image based on electromagnetic data;

[0027] The feature extraction module is used to process the acquired time-frequency domain joint image to extract multidimensional features;

[0028] The feature processing module is used to fuse the extracted multidimensional features to obtain a two-dimensional data matrix;

[0029] The model acquisition module is used to feed data-image samples into a classifier for training to obtain an electromagnetic emission interference source identification model.

[0030] The interference source rapid identification module is used to identify electromagnetic interference sources in electromagnetic emission data using an electromagnetic emission interference source identification model.

[0031] Furthermore, the image acquisition module is specifically used to process the time-domain data in the electromagnetic data to obtain its time-frequency domain data, and to use the inverse Fourier transform to obtain its time-frequency domain data in the electromagnetic data.

[0032] The time-domain data is represented as a waveform with time on the horizontal axis and amplitude on the vertical axis; the frequency-domain data is represented as amplitude-frequency and phase-frequency diagrams with frequency on the horizontal axis and power and phase on the vertical axes, respectively; the time-frequency-domain data is represented as a time-frequency-domain waterfall diagram with time on the horizontal axis and frequency on the vertical axis.

[0033] The waveform corresponding to the time domain data, the amplitude-frequency diagram and phase-frequency diagram corresponding to the frequency domain data, and the time-frequency waterfall diagram corresponding to the time-frequency domain data are combined to form a time-frequency domain joint image.

[0034] Furthermore, the feature extraction module is specifically used to extract time-frequency data from multiple samples by fusing and extracting time-frequency data through intermediate convolutional layers when training the neural network with spectral data.

[0035] Alternatively, data processing can be performed on the acquired time-frequency domain joint image to extract multi-dimensional features. Specifically, this includes periodically filtering the sampled features at different levels to obtain intermediate layer features abstracted from sampling on the network.

[0036] Alternatively, for the original time-spectrum image data, the high-amplitude region of the foreground of the time-spectrum image can be separated, and the foreground features can be combined with the fine-grained features of the image to extract the time-spectrum image features using image processing algorithms.

[0037] Furthermore, the model acquisition module is specifically used to use a set interference source loaded antenna to radiate emissions, and to acquire corresponding data-image samples through a test device for training. Based on the four basic waveforms of electromagnetic emission, a large number of feature images are acquired by changing the amplitude, phase, duty cycle, modulation method, rising edge, pulse width, repetition period parameters and signal combination form of the signal. These images are then used as training sample sets to train the classifier and obtain an electromagnetic emission interference source identification model.

[0038] The beneficial effects of this invention are as follows:

[0039] Instead of analyzing individual domain data, this method uses data visualization to collect and characterize electromagnetic data by visualizing time / frequency / time-frequency domain data and then outputting the combined data as a set of samples. It enables multi-dimensional data representation and analysis, and the data in the image samples also possess potential characteristics such as time duration, frequency bandwidth, and time-frequency domain correlation. These characteristics are more intuitive and easily obtained in image descriptions, offering more visualization features compared to traditional analysis methods.

[0040] In the traditional time-frequency domain data feature extraction process, image feature extraction has been introduced. By using image processing techniques to extract the cumulative effect, transient effect, and other correlation expressions exhibited after data visualization, the expressive features of a set of data samples can be more fully represented.

[0041] Replacing data samples with data-image samples enriches the form of data representation. Image samples can be further segmented using segmentation techniques, resulting in more training samples compared to traditional test data acquisition. This allows for more efficient training of classification models and further improves the accuracy of electromagnetic interference source identification. Attached Figure Description

[0042] Figure 1 This is a flowchart of a method for rapid identification of electromagnetic interference sources based on joint spectrum image analysis provided by an embodiment of the present invention;

[0043] Figure 2 This is a flowchart of the process for establishing an image multidimensional feature electromagnetic interference source identification model for electromagnetic emission test data, provided in an embodiment of the present invention.

[0044] Figure 3 These are schematic diagrams of the four basic waveforms of electromagnetic emission provided in the embodiments of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0046] This invention presents a rapid electromagnetic interference source identification method based on joint electromagnetic emission time-spectrum image analysis. Electromagnetic data is represented graphically, and features are extracted from the data and images to generate a feature database. Then, data fusion and dimensionality reduction techniques are used to process the data features. Finally, data-image samples are fed into a classifier for training, resulting in an electromagnetic emission interference source identification model. In practice, test data is fed into the model to obtain a matching output, identifying the type of interference source present in the electromagnetic emission data.

[0047] This invention provides a method for rapid identification of electromagnetic interference sources based on joint spectrum image analysis, such as... Figure 1 As shown, multiple sets of electromagnetic test data are first represented as time-domain waveforms, frequency-domain amplitude-frequency diagrams, frequency-domain phase-frequency diagrams, and time-frequency waterfall plots using a joint time-frequency domain transformation method. This integrates the relevant data into a set of image data, enabling the acquisition of large amounts of data by describing the electromagnetic data frame by frame. Further, data feature extraction and image feature extraction methods are used to extract features from the test data and images, achieving a unified representation of the test data from the time domain, frequency domain, time-frequency domain, and composite domain. Then, M groups of objects are iterated, and feature fusion is used to achieve a normalized expression of multi-dimensional features. The fused samples are then subjected to dimensionality reduction to remove redundant features. The fused two-dimensional feature data is then output as a feature image. Finally, each set of test data and feature image is used as a sample and fed into a classifier for training, ultimately applied to the requirement of identifying interference sources in electromagnetic emissions. By feeding the test data into the model, the corresponding interference source type can be obtained.

[0048] Specifically, such as Figure 2 As shown, experimental data was first acquired using acquisition equipment such as an oscilloscope, receiver, and real-time spectrum analyzer. Then, short-time Fourier transform, wavelet transform, and Hilbert-Huang transform were used to process the time-domain data to obtain their corresponding time-frequency data. The time-domain data was then processed to obtain its time-frequency domain data, and the frequency-domain data was processed using inverse Fourier transform to obtain its time-frequency domain data. The time-domain data was represented as a waveform with time on the horizontal axis and amplitude on the vertical axis; the frequency-domain data was represented as amplitude-frequency and phase-frequency diagrams with frequency on the horizontal axis and power and phase on the vertical axes, respectively; and the time-frequency domain data was represented as a time-frequency waterfall plot with time on the horizontal axis and frequency on the vertical axis. All images from a set of data samples were combined to form a set of image samples.

[0049] The acquired time-spectrum joint image was then processed in three parts: First, a neural network was used, employing a variational autoencoder to train the time-spectrum data. Windowing and fusion extraction of the time-frequency data were performed through intermediate convolutional layers to obtain multiple-sample time-frequency features. Second, the sampled features at different levels were periodically filtered to obtain intermediate-layer features abstracted from the network sampling. For intermediate-layer features, due to their varying dimensions, data at different sampling intervals were periodically filtered to obtain the network's upsampled abstract features. Third, for the original time-spectrum image, since the amplitude distinctions are quite clear (i.e., there are response regions and non-response regions), an unsupervised clustering algorithm was used to separate the time-spectrum image into two parts: a foreground electromagnetic response region and a background non-response region. Post-processing was then performed on the separated foreground region.

[0050] For fine-grained image features, image processing methods are employed to extract edge envelope features from the time-frequency spectrogram, shape features resulting from transient gradient changes in the time-frequency spectrum, and local texture features caused by data jitter within a short window. Further, secondary frequency domain features resulting from joint time-frequency domain changes can be extracted from the time-frequency domain data. When extracting image edge features, the edge characteristics in the X and Y directions of the time-frequency spectrogram can be extracted, i.e., the envelope characteristics of the time-frequency spectrogram. By combining the envelope characteristics in these two directions, the overall envelope of the time-frequency spectrogram can be obtained. Simultaneously, a histogram based on image gradient directions can be created. Compared to edge features that can only handle regions with relatively obvious edges, this method can handle edge detection in situations with low contrast between foreground and background, effectively extracting the shape characteristics caused by transient changes in the spectrogram, and also characterizing the direction of change for each pixel.

[0051] Window feature extraction is defined as comparing the grayscale values ​​of the eight adjacent pixels within a 3x3 window, using the center pixel as a threshold, to generate an 8-bit binary number, thereby obtaining the texture features of data jitter within a short window. Finally, the time-frequency domain data can be converted to the frequency domain to obtain a spectral image, representing the frequency domain features of the data. The image feature information is then stored in a database.

[0052] To address the electromagnetic emission time-frequency variation features extracted by different methods from the above, all features are fused. Features of different dimensions are downsampled to ensure all features have the same size. Then, data dimensionality reduction techniques are used to remove noise from the fused features. The feature fusion result is still expressed as a two-dimensional data matrix, therefore, image transformation is required. The acquired dimensionality-reduced features are divided into several patches to obtain training samples—image data—for subsequent training. Finally, a classifier model is built so that the low-dimensional parameter distribution feature map of the combined signal can be described by the feature map corresponding to the low-dimensional parameter distribution of the underlying single signal, and the single signal behind the combined signal can be analyzed. Radiation emission is performed using an antenna loaded with a specific interference source, and corresponding data is acquired through a testing device for training. Based on the four basic waveforms of electromagnetic emission, such as... Figure 3 As shown, a large number of feature images are obtained by changing parameters such as signal amplitude, phase, duty cycle, modulation method, rising edge, pulse width, and repetition period, as well as signal combinations. These images are then used as a training sample set to train the classifier. Finally, the correlation and accuracy of the basic waveform feature domain representation are verified using electromagnetic emission data from typical test samples.

[0053] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A rapid identification method for electromagnetic interference sources based on joint spectrum image analysis, characterized in that, Includes the following steps: Obtain a joint time-frequency domain image based on electromagnetic data; Multidimensional features are extracted from the acquired time-frequency domain joint image through data processing. The extracted multidimensional features are fused to obtain a two-dimensional data matrix; The two-dimensional data matrix is ​​transformed into an image to obtain data-image samples; Data-image samples are fed into a classifier for training to obtain an electromagnetic emission interference source identification model. Electromagnetic interference sources in electromagnetic emission data are identified using an electromagnetic emission interference source identification model. The acquisition of a joint time-frequency domain image based on electromagnetic data specifically includes: The time-domain data in the electromagnetic data is processed to obtain its time-frequency domain data, and the frequency-domain data in the electromagnetic data is obtained by inverse Fourier transform. The time-domain data is represented as a waveform with time on the horizontal axis and amplitude on the vertical axis; the frequency-domain data is represented as amplitude-frequency and phase-frequency diagrams with frequency on the horizontal axis and power and phase on the vertical axes, respectively; the time-frequency-domain data is represented as a time-frequency-domain waterfall diagram with time on the horizontal axis and frequency on the vertical axis. The waveform corresponding to the time domain data, the amplitude-frequency diagram and phase-frequency diagram corresponding to the frequency domain data, and the time-frequency waterfall diagram corresponding to the time-frequency domain data are combined to form a time-frequency domain joint image; The data processing and multidimensional feature extraction of the acquired time-frequency domain joint image also includes: For the original time-spectrum image data, the high-amplitude region in the foreground of the time-spectrum image is separated, and the foreground features are combined with the fine-grained features of the image to extract the time-spectrum image features using image processing algorithms.

2. The method for rapid identification of electromagnetic interference sources based on joint spectrum image analysis according to claim 1, characterized in that, The data processing and multidimensional feature extraction of the acquired time-frequency domain joint image specifically includes: The time-frequency data is trained using a neural network, and then fused and extracted through intermediate convolutional layers to obtain the time-frequency data features from multiple samplings.

3. The method for rapid identification of electromagnetic interference sources based on joint spectrum image analysis according to claim 1, characterized in that, The data processing and multi-dimensional feature extraction of the acquired time-frequency domain joint image also includes: periodically filtering the sampled features at different levels to obtain intermediate layer features abstracted from sampling on the network.

4. The method for rapid identification of electromagnetic interference sources based on joint spectrum image analysis according to claim 1, characterized in that, The data-image samples are fed into a classifier for training to obtain an electromagnetic emission interference source identification model, which specifically includes: The antenna is loaded with a set interference source to conduct radiation emission. The corresponding data-image samples are obtained through the test device for training. Based on the four basic waveforms of electromagnetic emission, a large number of feature images are obtained by changing the amplitude, phase, duty cycle, modulation method, rising edge, pulse width, repetition period parameter and signal combination form of the signal. These images are used as training sample sets to train the classifier and obtain the electromagnetic emission interference source identification model.

5. A rapid electromagnetic interference source identification device based on spectrum joint image analysis, used to implement the rapid electromagnetic interference source identification method based on spectrum joint image analysis as described in any one of claims 1-4, characterized in that, The identification device includes: The image acquisition module is used to acquire a joint time-frequency domain image based on electromagnetic data; The feature extraction module is used to process the acquired time-frequency domain joint image to extract multidimensional features; The feature processing module is used to fuse the extracted multidimensional features to obtain a two-dimensional data matrix; The model acquisition module is used to feed data-image samples into a classifier for training to obtain an electromagnetic emission interference source identification model. The interference source rapid identification module is used to identify electromagnetic interference sources in electromagnetic emission data using an electromagnetic emission interference source identification model.

6. The rapid electromagnetic interference source identification device based on spectrum joint image analysis according to claim 5, characterized in that, The image acquisition module is specifically used to process the time-domain data in the electromagnetic data to obtain its time-frequency domain data, and to use the inverse Fourier transform to obtain its time-frequency domain data in the electromagnetic data. The time-domain data is represented as a waveform with time on the horizontal axis and amplitude on the vertical axis; the frequency-domain data is represented as amplitude-frequency and phase-frequency diagrams with frequency on the horizontal axis and power and phase on the vertical axes, respectively; the time-frequency-domain data is represented as a time-frequency-domain waterfall diagram with time on the horizontal axis and frequency on the vertical axis. The waveform corresponding to the time domain data, the amplitude-frequency diagram and phase-frequency diagram corresponding to the frequency domain data, and the time-frequency waterfall diagram corresponding to the time-frequency domain data are combined to form a time-frequency domain joint image.

7. The rapid electromagnetic interference source identification device based on joint spectrum image analysis according to claim 5, characterized in that, The feature extraction module is specifically used to extract time-frequency data from multiple samples by fusing and extracting time-frequency data through intermediate convolutional layers using spectral data trained on a neural network. Alternatively, data processing can be performed on the acquired time-frequency domain joint image to extract multi-dimensional features. Specifically, this includes periodically filtering the sampled features at different levels to obtain intermediate layer features abstracted from sampling on the network. Alternatively, for the original time-spectrum image data, the high-amplitude region in the foreground of the time-spectrum image can be separated, and the foreground features can be combined with the fine-grained features of the image to extract the time-spectrum image features using image processing algorithms.

8. The rapid electromagnetic interference source identification device based on spectrum joint image analysis according to claim 5, characterized in that, The model acquisition module is specifically used to load an antenna with a set interference source to radiate emissions, acquire corresponding data-image samples through a test device for training, and obtain a large number of feature images by changing the amplitude, phase, duty cycle, modulation method, rising edge, pulse width, repetition period parameters and signal combination forms of the four basic waveforms of electromagnetic emission based on the four basic waveforms of electromagnetic emission. These feature images are then used as training sample sets to train the classifier and obtain an electromagnetic emission interference source identification model.

Images