Method for capturing sporadic abnormal signals in real time by intelligent oscilloscope

Abstract

The application discloses a kind of sporadic abnormal signal real-time capture method of intelligent oscilloscope, first by normal time-domain signal filtering denoising processing and signal decomposition starting point determination, obtain preprocessed data, then using empirical mode decomposition algorithm to normal signal decomposition, obtain the IMF component signal of normal signal, then the IMF component signal of normal signal is stored;Then when system has no signal input, according to the template of normal signal reconstructed from non-noise IMF component, then input the signal to be tested, calculate the distance of the signal to be tested with the reconstructed normal signal according to the distance sum of two signals to judge abnormality, when the degree of deviation is greater than the threshold value issued by software, the segment of the signal to be tested is detected as abnormal signal in real time.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of time domain signal testing, and more specifically relates to a method for real-time capture of sporadic abnormal signals of an intelligent oscilloscope. BACKGROUND

[0002] Time domain signal testing is a core component of signal processing and data analysis, mainly used for obtaining and analyzing the characteristics of signals changing with time. At present, with the rapid development of electronic systems, the form of time domain signals presents a trend of diversification and complication. Especially with the increase of acquisition rate, the form of time domain signals becomes more and more complex, and the transient characteristics increase sharply. Abnormal signals are sporadic and often cause equipment to work abnormally, system performance to drop sharply, etc. Therefore, the detection and real-time detection of abnormal signals are particularly important in the fields of electronic equipment maintenance and equipment health monitoring, wireless communication and radar, anti-jamming, etc.

[0003] In the field of time domain signal testing, the real-time detection of abnormal signals is based on time domain testing instruments such as oscilloscopes. The traditional oscilloscope captures specific abnormal signals by setting trigger conditions conforming to abnormal signals or adopts three-dimensional mapping, peak detection, etc. to capture abnormalities in normal signals. In fact, many abnormal signals are sporadic and transient signals with very short duration. It is difficult to rely on the form of signals to set specific capture conditions to capture abnormal signals when facing unknown signals, which is also a problem faced by traditional oscilloscopes. At the same time, in the actual signal acquisition process, there is a contradiction between large-capacity data and real-time storage and processing of signals. In order to cope with the non-missing detection of abnormal signals in high-speed data flow of tens of GSPS, traditional three-dimensional mapping, segmented storage, etc. are usually used in high-speed acquisition systems to effectively capture abnormal signals by improving the speed of waveform storage and processing. In the face of massive signal data, the signal is usually stored and processed uniformly without distinguishing between normal signals and abnormal signals. The time required for processing normal signals is relatively long and the resources are relatively large, resulting in a large dead time of the testing instrument, which causes abnormal signals to be often covered by normal signals, resulting in a small probability of identifying and capturing abnormal signals.

[0004] The existing time domain signal anomaly recognition method based on data analysis mainly includes two categories: one is to use traditional algorithm, obtain the statistical parameters of the signal through statistical analysis method, and compare with the set threshold value to judge whether the time domain signal is abnormal. Another is to use advanced machine learning algorithm and deep neural network model, according to a large number of normal signal history data to train the model, the model can learn the distribution of normal signal, and is used to distinguish abnormal signal, such as variational autoencoder, long short-term neural network and the like. However, in the actual application process of the existing time domain signal anomaly recognition method based on data analysis, the calculation amount is large, the model is trained and detected by computer offline, and it is difficult to detect the abnormal signal in real time. For the captured abnormal signal, the existing acquisition system does not process the abnormal signal, but needs to save the data for secondary processing, which brings great inconvenience to the user. SUMMARY

[0005] The present application aims to overcome the shortcomings of the prior art, and provides an occasional abnormal signal real-time capturing method of intelligent oscilloscope, which judges the abnormality by the deviation degree of the residual error of the signal to be tested and the residual error of the normal signal, solves the problems of low real-time capturing efficiency of signals with complex and unknown morphology, and low real-time detection rate of abnormal signals caused by traditional statistical analysis method or software implementation method, and improves the real-time detection rate of abnormal signals.

[0006] To achieve the above application purpose, the occasional abnormal signal real-time capturing method of intelligent oscilloscope, characterized in that, comprising the following steps:

[0007] (1) The signal source generates a normal input signal input to the intelligent oscilloscope, and acquires a serial signal;

[0008] (2) The serial signal is converted into parallel multi-channel signal through serial-parallel conversion module, and expressed in matrix form as:

[0009]

[0010] Wherein, d ij represents the i-th sampling point under the j-th sampling clock, i=0,1,2,…,Q-1, Q represents the number of data channels for parallel processing, j=0,1,2,…,P-1, P represents the number of clock cycles;

[0011] (3) The parallel multi-channel signal is first filtered by N times sliding average, and then the N times filtering results are averaged to obtain the denoised parallel multi-channel signal;

[0012] (4) The period and amplitude of the parallel multi-channel signal are extracted, then the length L of EMD decomposition is set according to the period, and the trigger comparison level V o is set according to the amplitude;

[0013] (5) Extract the parallel multi-channel signals to obtain a normal signal of length L;

[0014] (6) Turn off the signal source, and then obtain the noise matrix D according to steps (1) to (2). * Traverse the noise matrix D * All elements in the array, then by the maximum value of the elements. minimum value of elements Calculate the range σ in ;

[0015]

[0016] (7) Perform parallel Q-path EMD decomposition on the normal signal to obtain the signal template;

[0017] (8) The signal source randomly generates the input signal, and then obtains the matrix of the signal to be tested according to steps (1) to (2).

[0018] (9) In the matrix of signals to be measured Find the starting point of EMD decomposition, and then take the starting point as the starting point and extract a vector of length L in the column direction, which is denoted as the signal to be measured Y.

[0019] (10) Calculate the L2 distance d between the signal Y to be measured and the signal template S. L2 Then based on the distance d L2 The anomaly detection threshold δ is used to detect anomalies in the signal under test. If d L2 If the value is greater than δ, the signal to be tested is determined to be an abnormal signal; otherwise, the signal to be tested is determined to be a normal signal.

[0020] The objective of this invention is achieved as follows:

[0021] This invention discloses a method for real-time acquisition of sporadic abnormal signals in an intelligent oscilloscope. First, preprocessed data is obtained by filtering and denoising the normal time-domain signal and determining the starting point of signal decomposition. Then, the normal signal is decomposed using an empirical mode decomposition algorithm to obtain the IMF component signal of the normal signal, which is then stored. Next, a template for reconstructing the normal signal based on the non-noise IMF component is obtained when there is no signal input to the system. After inputting the signal to be tested, the distance between the reconstructed normal signal and the signal to be tested is calculated. Anomalies are judged based on the sum of the distances between the two signals. When the deviation exceeds a threshold issued by the software, the segment of the signal to be tested is detected as an abnormal signal in real time.

[0022] Meanwhile, the real-time acquisition method for sporadic abnormal signals of a smart oscilloscope according to the present invention also has the following beneficial effects:

[0023] 1、The present application can be implemented on a platform with real-time digital signal processing, such as FPGA, with real-time performance, which can significantly improve the real-time detection efficiency of abnormal signals and the real-time abnormal capture rate.

[0024] 2、For the captured abnormal signals, the existing acquisition system does not process the abnormal signals, but needs to save the data for secondary processing, which brings great inconvenience to the user; and the present application can automatically and real-time detect the abnormal signals based on the empirical mode decomposition method, and improve the abnormal signal capture rate.

[0025] 3、Compared with the prior knowledge of traditional signal characteristics, the present application does not need to manually define abnormal parameters, and can avoid the complicated process of capturing specific signals by setting complex trigger conditions, greatly simplifying the use of time domain test instruments in abnormal signal detection.

[0026] 4、The present application can display the captured abnormal signals in multiple dimensions, compared with the original single abnormality observation, we can specify the abnormality, single frame, and superimposed display in three modes. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is a flow chart of the present application, a real-time capture method of sporadic abnormal signals of an intelligent oscilloscope;

[0028] Figure 2 is a schematic diagram of a normal input signal;

[0029] Figure 3 is a schematic diagram of a signal to be measured. DETAILED DESCRIPTION

[0030] The specific embodiments of the present application will be described below in conjunction with the accompanying drawings, so that those skilled in the art can better understand the present application. It should be particularly noted that in the following description, when the detailed description of known functions and designs may obscure the main content of the present application, these descriptions will be omitted here.

[0031] EMBODIMENT

[0032] In this embodiment, as shown in Figure 1 The present application, a real-time capture method of sporadic abnormal signals of an intelligent oscilloscope, comprises the following steps:

[0033] (1) The signal source generates a normal input signal input to the intelligent oscilloscope, and then acquires a serial signal;

[0034] (2) The serial signal is converted into a parallel multi-channel signal by a serial-parallel conversion module, and is represented in matrix form as:

[0035]

[0036] wherein d ij represents the i-th sampling point under the j-th sampling clock, i=0, 1, 2, …, Q-1, Q represents the number of parallel processing data channels, j=0, 1, 2, …, P-1, P represents the number of clock cycles;

[0037] In the embodiment, it is assumed that the sampling rate of the intelligent oscilloscope is f acq , the processing clock frequency is f dsp , and the number of parallel processing data channels Q=f acq / f dsp after conversion by the serial-parallel conversion module.

[0038] (3) The parallel multi-channel signal is first subjected to N times of sliding average filtering, N being determined according to the noise of the signal itself. Then, the N times of filtering results are averaged to obtain the denoised parallel signal.

[0039] In the embodiment, N is determined according to the noise of the signal itself. If the signal noise is too large, N is relatively large, and if the signal noise is small, N is relatively small.

[0040] (4) The period and amplitude of the parallel multi-channel signal are extracted.

[0041] In the embodiment, the length L of EMD decomposition is set according to the period, and the length L of EMD decomposition is at least greater than one complete period of the parallel multi-channel signal. The amplitude is used to set the size of the trigger comparison level V o , and generally V o is smaller than the amplitude.

[0042] (5) The parallel multi-channel signal is intercepted.

[0043] In the embodiment, the interception of the signal refers to intercepting a segment of the signal from the trigger point to a segment of the signal satisfying the length L of EMD decomposition, and the specific process is as follows:

[0044] (5.1) The trigger comparison level V o is set to be smaller than the period of the parallel multi-channel signal. The parallel multi-channel signal is input to the trigger comparison module, each element d i,j in the matrix D is compared with the trigger comparison level V o , and each element c i,j in the matrix C is obtained.

[0045]

[0046] (5.2) The element c 0,0As a starting point, traverse the elements in matrix C by column, and when the first element with a value of 1 is found, record the position of the element as the starting point position of EMD decomposition;

[0047] (5.3) The length of EMD decomposition is at least greater than one complete cycle by adaptive extraction and interpolation of the signal.

[0048] In this embodiment, since EMD decomposition requires at least one complete waveform, extraction and interpolation are used to expand and compress the waveform to meet the length requirement of decomposition.

[0049] (5.4) Find the starting point position of EMD decomposition in matrix D, and then take a vector with a length of L as the starting point position, denoted as normal signal Y0;

[0050] (6) Turn off the signal source, and then obtain noise matrix D according to steps (1) to (2) * ; Traverse all elements in noise matrix D * , and then calculate the range σ by the maximum value of the elements and the minimum value of the elements in ;

[0051]

[0052] In this embodiment, noise matrix D * is the noise of the oscilloscope itself when there is no signal input.

[0053] (7) Real-time parallel EMD decomposition and reconstruction of normal signal

[0054] (7.1) Perform Q-path real-time parallel EMD decomposition on normal signal Y0 to obtain multiple IMF components;

[0055] In this embodiment, due to the limitation of hardware resources, the number of iterations of Q-path real-time parallel EMD decomposition is limited, and the number of generated IMF components depends on the characteristics of the signal itself.

[0056] (7.2) Calculate the range σimfk of each IMF component, which is similar to the calculation in matrix D * , and the range is calculated by the difference between the maximum value and the minimum value.

[0057] (7.3) If the range of an IMF component satisfies: σ imf < σ in + ε, ε represents the noise of the input signal itself, then the IMF component is considered to be a noise component, otherwise, it is considered to be a normal component;

[0058] (7.4), remove the noise component, and then the remaining K IMF components constitute the signal template;

[0059]

[0060] (8), the signal source randomly generates the input signal, and then acquires the to-be-tested signal matrix according to steps (1)-(2)

[0061] (9), in the to-be-tested signal matrix find the starting point position of the EMD decomposition, and then take a vector with a length of L as the starting point of the starting point position, and mark it as the to-be-tested signal Y;

[0062] (10), calculate the L2 distance d between the to-be-tested signal Y and the signal template S L2 ;

[0063] (10.1), respectively, the to-be-tested signal Y and the signal template S are expressed in the form of a matrix, and the size of the matrix is QXP, and the missing element position in the matrix is filled with 0;

[0064] For example, when taking a vector with a length of L0 as the starting point, all the sampling points after the starting point position are taken, and then when expressing the matrix Y and the matrix S, the sampling points before the starting point position are all filled with 0.

[0065] (10.2), calculate the L2 distance between the matrix Y and the matrix S;

[0066]

[0067] Where, y i,j represents the element of the i-th row and the j-th column of the matrix Y, and s i,j represents the element of the i-th row and the j-th column of the matrix S.

[0068] (11), to-be-tested signal anomaly detection;

[0069] Set and issue the abnormality judgment threshold δ; if d L2 > δ, it is judged that the to-be-tested signal is an abnormal signal; otherwise, it is judged that the to-be-tested signal is a normal signal.

[0070] In this embodiment, the abnormality judgment threshold δ can be σ in + ε.

[0071] When an abnormal signal is detected, the amplitude waveform is stored, and the abnormal signal segment is stored in the RAM. If no abnormality is detected, the system detects in real time, does not store the signal, and finally uploads the abnormal data to the upper computer for display and observation.

[0072] Examples

[0073] As shown in Figure 2 , the sinusoidal signal with length L1 as a normal time-domain signal, as shown in Figure 3 , the amplitude modulation signal with length L2 as the abnormal real-time identification step of the time-domain signal to be tested, the used FPGA is xcku115-flva1517-2-i of Xilinx company, and the specific process is as follows: the signal source input and then the serial time-domain signal is collected;

[0074] The sampling rate of the intelligent oscilloscope is f acq =20Gsps, the processing clock frequency is f dsp =250Mhz, so the data lane number Q of parallel processing after the conversion of the serial-parallel conversion module is f acq / f dsp =80.

[0075] In matrix form:

[0076]

[0077] Where d ij represents the i-th sampling point under the j-th sampling clock, i=0,1,2,…,Q-1, Q represents the data lane number of parallel processing, i.e. 80, j=0,1,2,…,P-1, P represents the clock period number;

[0078] The parallel time-domain signal is first filtered by 1-time moving average filter, and then the filtered result is averaged to obtain the denoised parallel signal;

[0079] The normal signal feature is extracted from the parallel time-domain signal, the period feature is 200μs per period, and the amplitude is 4000.

[0080] The trigger level is set to 3100, and after comparing with the starting point, the first trigger position is located at d 28,0 of the matrix, the length of EMD decomposition is set to 400, which has met the length of EMD decomposition at least greater than one complete period. Therefore, there is no need to extract and interpolate. Turn off the signal source, and then obtain the noise matrix D * ; traverse all elements in the noise matrix D * , and then calculate the range σ in =112 by the maximum value of the element and the minimum value of the element .

[0081]

[0082] To Figure 2The signal is decomposed by 80 parallel EMD to obtain three IMF components. The range σimfk of each IMF component is calculated; the range of the first IMF component satisfies: σ imf <σ in +ε, ε represents the noise of the input signal itself, and the IMF component is considered as a noise component, and the other two are considered as normal components; the remaining two are reconstructed. The signal source input Figure 3 The signal is decomposed by 80 parallel EMD to obtain three IMF components. The range σimfk of each IMF component is calculated; the range of the first IMF component satisfies: σ L2 .

[0083] The L2 distance between matrix Y and matrix S is calculated;

[0084]

[0085] The software sets and issues an abnormal threshold value δ=400;

[0086] Because d L2 >δ, the test signal is determined as an abnormal signal; the abnormal data is uploaded to the upper computer, and the specified frame of the observed abnormality can be displayed and superimposed.

[0087] Although the above describes the specific embodiments of the present application in order to facilitate the understanding of the present application by those skilled in the art, it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, it is obvious that various changes are within the spirit and scope of the present application defined and determined by the appended claims, and all the inventions using the concept of the present application are within the scope of protection.

Claims

1. A method for real-time capturing of sporadic abnormal signals of an intelligent oscilloscope, characterized in that, The method comprises the following steps: (1) a signal source generates a normal input signal input to an intelligent oscilloscope, and a serial signal is acquired; (2) the serial signal is converted into a parallel multi-channel signal through a serial-parallel conversion module, and is expressed in a matrix form as: wherein d ij represents the ith sampling point under the jth sampling clock, i = 0, 1, 2, …, Q-1, Q represents the number of data lanes for parallel processing, j = 0, 1, 2, …, P-1, P represents the number of clock cycles; (3) the parallel multi-channel signal is first subjected to N times of sliding average filtering, and then the N times of filtering results are averaged to obtain a denoised parallel multi-channel signal; (4), extract the period and amplitude of the parallel multi-channel signal, and then set the length of EMD decomposition as L according to the period, and set the trigger comparison level V according to the amplitude o ; (5) the parallel multi-channel signal is intercepted to obtain a normal signal with a length of L; (6) Turn off the signal source, and then acquire the noise matrix D according to steps (1)-(2) * ; traverse all elements in the noise matrix D * , and then calculate the range σ by the maximum value of the elements and the minimum value of the elements in ; (7) the normal signal is subjected to parallel Q-path EMD decomposition to obtain a signal template; (8) The signal source randomly generates the input signal, and then acquires the to-be-tested signal matrix according to steps (1)-(2) (9), in the signal matrix to be measured The starting point position of the EMD decomposition is found, and then a vector with a length of L is intercepted in the column direction from the starting point position as the starting point, and is recorded as the signal Y to be measured; (10) Calculate the L2 distance d between the signal Y to be measured and the signal template S. L2 Then based on the distance d L2 The anomaly detection threshold δ is used to detect anomalies in the signal under test. If d L2 If the value is greater than δ, the signal to be tested is determined to be an abnormal signal; otherwise, the signal to be tested is determined to be a normal signal.

2. The method for capturing the sporadic abnormal signal of the intelligent oscilloscope in real time according to claim 1, characterized in that, The method for intercepting the normal signal in the step (5) is: (2.1), inputting the parallel multi-channel signal into the trigger comparison module, and obtaining each element c in matrix C by traversing each element d in matrix D i,j and the trigger comparison level V o comparison, obtaining each element c in matrix C i,j ; (2.2), taking element c 0,0 As a starting point, traverse each element in matrix C by column, when traverse to the first element point whose value is 1, then record the position of the point as the starting point position of EMD decomposition. (2.3) the parallel multi-channel signal is subjected to adaptive decimation or interpolation, so that the length L of the EMD decomposition is at least greater than one complete period of the parallel multi-channel signal; (2.4) a starting point position of the EMD decomposition is found in the matrix D, and then a vector with the length L is taken in the column direction from the starting point position as a starting point, and is recorded as the normal signal Y0.

3. The method of claim 1, wherein the method comprises: The method for obtaining the signal template through the EMD decomposition in the step (7) is: (3.1) the normal signal Y0 is subjected to parallel Q-path EMD decomposition to obtain a plurality of IMF components; (3.2) calculating the range σ of each IMF component imfk ; (3.3) If the range of a certain IMF component satisfies: σ imf <σ in + ε, ε represents the noise of the input signal itself, then the IMF component is considered as a noise component, otherwise, it is considered as a normal component; (3.4) noise components are removed, and then the remaining K IMF components constitute the signal template; Wherein, K represents the number of the remaining IMF components after the noise components are removed.

4. The method of claim 1, wherein the method comprises: The step (10) calculates the L2 distance d between the to-be-tested signal Y and the signal template S L2 The method is as follows: (4.1) the abnormal signal Y and the signal template S are respectively expressed in a matrix form, and the matrix sizes are both QxP, and the missing element positions in the matrix are filled with 0; (4.2) an L2 distance between the matrix Y and the matrix S is calculated; where y i,j denotes the element of matrix Y in the i-th row and j-th column, s i,j denotes the element of matrix S in the i-th row and j-th column.

Images