Real-time anomaly detection method based on time-frequency domain in intelligent oscilloscope

Abstract

The invention discloses a real-time anomaly detection method based on a time-frequency domain in an intelligent oscilloscope, and the method comprises the steps: carrying out the sampling of an input signal through a collection system, obtaining a digital sampling signal, and transmitting the digital sampling signal to a time-frequency domain anomaly detection module in an FPGA; in the time-frequency domain anomaly detection module, anomaly detection of signals is completed based on anomaly detection of spectrum distribution, channel division of data in a digital domain, adaptive downsampling, short-time Fourier transform with variable window length and anomaly detection based on time-frequency matrix representation. And after the detection is completed, the acquisition system stores the signal which is judged to be abnormal, and the signal is finally displayed on the upper computer.

Description

Technical Field

[0001] The present invention belongs to the technical field of abnormality detection of acquisition systems, and more specifically, relates to a real-time abnormality detection method based on time-frequency domain in an intelligent oscilloscope. Background Art

[0002] With the increasing complexity of electrical signals and the rapid development of 5G and 6G technologies, the requirements for observing high-speed electrical signals and acquisition systems are getting higher and higher. The acquisition systems are also becoming more and more complex, and higher requirements are placed on the acquisition system's ability to capture specific abnormal signals. Although the general trigger function can capture data waveforms, it can only capture abnormal signals with simple and known shapes. When faced with complex signals, especially when detecting anomalies in signals with regular frequency domain but complex time domain shapes, its ability to capture abnormal signals is insufficient. Due to the diversity and complexity of abnormal signals, it is difficult for users to capture abnormal signals by setting existing trigger conditions, and abnormal signals often contain the information that users want. Therefore, the necessity of the abnormal detection function for high-speed acquisition systems is obvious.

[0003] In most radar and communication signal fields, signals are different from simple sine and square waves. Many of them are modulated signals. It is often difficult to judge abnormalities only in the time domain or frequency domain. However, the time-frequency domain characteristics of these signals have obvious rules. Traditional acquisition systems do not have the ability to detect real-time abnormalities in the time-frequency domain, which makes it impossible for the acquisition system to effectively capture abnormalities of such signals. For example, if you need to analyze frequency modulated (FM) signals, the frequency changes with time. The general trigger function can only capture and display the waveform, and it is difficult to capture abnormalities. It is difficult to distinguish abnormalities through abnormality detection in the time domain, and it is also difficult to distinguish abnormalities through abnormality detection in the frequency domain. However, by performing short-time Fourier transform (STFT) on the signal and analyzing its time-frequency domain characteristics, it is possible to determine whether the signal has abnormalities.

[0004] For a broadband acquisition system, if the full bandwidth signal is calculated and analyzed, the amount of calculation is extremely large, the resource occupancy rate is large, the throughput is low, and the system processing time is too long. There will be a long period of time when abnormal signals cannot be detected, resulting in poor real-time performance of the system and failure to meet the abnormality detection requirements of the high-speed acquisition system. Therefore, a time-frequency domain anomaly detection method that can extract the time-frequency domain features of the signal in real time is needed. By analyzing the time-frequency domain features of the signal, the anomalies in the acquisition can be captured to meet the real-time requirements of the high-speed acquisition system. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope, which determines whether a signal is abnormal by comparing the time-frequency domain characteristics and displays the captured abnormal signal in real time.

[0006] To achieve the above-mentioned object of the invention, the present invention provides a real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope, characterized in that it comprises the following steps:

[0007] (1) The digital oscilloscope is powered on and initialized, and then the input signal is connected;

[0008] (2) In the normal detection mode, the input signal is sampled by the acquisition system in the digital oscilloscope to obtain a digital sampling signal;

[0009] (3) Input the digital sampling signal into the FFT unit and the WOLA structure respectively;

[0010] (3.1) Perform FFT operation on the digital sampling signal in the FFT unit to obtain the spectrum distribution feature X norm ={X norm [0],X norm [1],…,X norm [n],…,X norm [N-1]}, N is the number of points of FFT operation;

[0011] Traverse the spectrum distribution feature X norm Each spectral component X in norm [n], mark the spectral component X norm The spectrum components whose [n] is greater than the threshold δ are valid spectrum components;

[0012] (3.2) In the WOLA structure, the digital sampling signal is channelized digitally filtered by the channelized digital filter in each channel and then filtered with the frequency The digital single tone signal is digitally mixed, and finally digital anti-image filtering is performed through a digital anti-image filter, so that a low-frequency digital signal is obtained in each channel; where i represents the channel number, i = 0, 1, ..., m-1, m is the number of channels, and f s is the real-time sampling rate of the acquisition system;

[0013] Finally, the D-rate adaptive extraction of time domain data is performed on each low-frequency digital signal, and the output signal of each channel is finally obtained;

[0014] (4) Determine the channel where the effective spectrum component is located in the WOLA structure;

[0015] Traverse the spectrum distribution feature X norm Each effective spectral component in the norm [k], then the corresponding frequency component is the effective component, where k is the number of the effective spectral component, and f sis the real-time sampling rate of the acquisition system, if Then the i-th channel is the effective spectral component X norm [k] the channel it is located on;

[0016] (5) According to the channel determined in step (4), the output signal of the corresponding channel i in the WOLA structure is selected for adaptive short-time Fourier transform, and then down-sampled. Finally, the down-sampled data is converted into a time-frequency matrix P norm [i];

[0017] (6) The spectrum distribution feature X obtained in the normal detection mode norm And the time-frequency matrix P norm [i] As standard spectrum distribution template and time-frequency matrix template and deployed to FPGA;

[0018] (7) Set the digital oscilloscope to the abnormality detection mode, and then repeat steps (2) to (5) to obtain the frequency spectrum distribution feature X of the input signal in the abnormality detection mode. in ={X in [0],X in [1],…,X in [n],…,X in [N-1]} and the time-frequency matrix P in [i];

[0019] (8) Abnormal judgment of input signal;

[0020] (8.1), traverse the spectrum distribution feature X norm and X in For each spectral component in X norm and X in As long as there is a set of spectral components that satisfy the relationship: |X in [n]-X norm [n]|>X set , X set To set the threshold, the input signal is determined to be an abnormal signal, and then go to step (8.3); otherwise, go to step (8.2);

[0021] (8.2), traverse the time-frequency matrix P norm [i] and P in [i], compare whether the channel numbers i are the same one-to-one. If the channel numbers are different, the input signal is determined to be an abnormal signal, and then go to step (8.3); if the channel numbers are the same one-to-one, use the starting point search and cyclic comparison method to compare all the time-frequency matrices P with the same number norm [i] and P inFor each element in [i], if the comparison results of the elements at a certain position are different, the input signal is determined to be an abnormal signal, and then the process goes to step (8.3); if the comparison results of the time-frequency matrices corresponding to the channel numbers are the same, the input signal is determined to be a normal signal, and then the process goes to step (8.3);

[0022] (8.3) The judged signal is input to the host computer for real-time display.

[0023] The object of the invention of the present invention is achieved in this way:

[0024] The present invention discloses a real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope. The method comprises the following steps: firstly, an input signal is sampled by an acquisition system to obtain a digital sampling signal and send the digital sampling signal to a time-frequency domain anomaly detection module in an FPGA; in the time-frequency domain anomaly detection module, anomaly detection of the signal is completed based on anomaly detection of spectrum distribution, channel division of data in the digital domain, adaptive downsampling, short-time Fourier transform of variable window length and anomaly detection based on time-frequency matrix representation; after the detection is completed, the acquisition system stores the signal judged to be abnormal and finally displays it on a host computer.

[0025] The real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope of the present invention also has the following beneficial effects:

[0026] (1) Adopting adaptive downsampling technology based on WOLA structure, this structure can divide the full bandwidth signal into multiple narrow bandwidth digital channels in the digital domain, and then move the sub-signal of each channel to the low frequency band through digital domain signal processing. At the same time, by real-time adaptive multi-rate extraction of the signal time domain waveform in the digital domain, the equivalent sampling rate of the signal is reduced, thereby greatly reducing the amount of calculation of subsequent real-time anomaly detection, thereby realizing real-time calculation of signal time-frequency characteristics and anomaly detection;

[0027] (2) In the WOLA structure, the time domain data of the mixed low-frequency signal is adaptively extracted, the equivalent sampling rate of the signal is adaptively reduced, and the amount of subsequent short-time Fourier transform calculations is reduced, thereby improving the real-time performance of the system and increasing the efficiency of the system in capturing abnormalities.

[0028] (3) When the user needs to capture an abnormal signal, it is often necessary to perform complex settings on the acquisition system. The detection method of the present invention can automatically detect the abnormality of the input signal without the user being familiar with the various functions of the acquisition system, thereby improving the oscilloscope's ability to automatically detect abnormal signals;

[0029] (4) It is difficult to detect abnormalities in some signals only by using abnormality detection methods in the time domain or frequency domain. The detection method of the present invention can solve this problem and greatly improve the ability of the acquisition system to accurately identify and capture abnormal signals; BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of a real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope of the present invention;

[0031] Figure 2 is a schematic diagram of the input signal;

[0032] Figure 3 This is a partial screenshot of the FFT transformation of the sampling signal in normal detection mode;

[0033] Figure 4 This is the signal time-frequency diagram obtained by channel 4 in normal detection mode;

[0034] Figure 5 This is the signal time-frequency diagram obtained on channel 5 in normal detection mode;

[0035] Figure 6 This is a partial screenshot of the sampled signal after FFT transformation in the anomaly detection mode;

[0036] Figure 7 It is the signal time-frequency diagram obtained in the 4th channel in the anomaly detection mode;

[0037] Figure 8 This is the signal time-frequency diagram obtained in the 5th channel in the anomaly detection mode;

[0038] Fig. 9 It is a schematic diagram of starting point search and loop comparison. DETAILED DESCRIPTION

[0039] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0040] Example

[0041] In this embodiment, if Figure 1 As shown, the present invention provides a real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope, comprising the following steps:

[0042] S1. The digital oscilloscope is powered on and initialized, and then the input signal is connected. The input signal can be as follows Figure 2 The signal shown; in this embodiment, the digital oscilloscope works in the normal detection mode by default when it is powered on and initialized;

[0043] S2. In the normal detection mode, the input signal is sampled by the acquisition system in the digital oscilloscope to obtain a digital sampling signal. In this embodiment, it is necessary to collect sampling data of sufficient length, at least including a complete cycle of the input signal;

[0044] S3, inputting the digital sampling signal into the FFT unit and the WOLA structure respectively;

[0045] S3.1. Perform FFT operation on the digital sampling signal in the FFT unit to obtain the spectrum distribution feature X norm ={X norm [0],X norm [1],…,X norm [n],…,X norm [N-1]}, N is the number of points for FFT operation. Here N is 4096, and a 4096-point FFT is performed. The FFT result is as follows Figure 3 As shown, only a part of the signal in FFT is intercepted;

[0046] Traverse the spectrum distribution feature X norm Each spectral component X in norm [n], mark the spectral component X norm [n] The spectrum components greater than the threshold δ = 150 are valid spectrum components; in this example, the effective frequency range of the signal in the FFT of the signal corresponds to the number of FFT points from 103 to 143, and the system sampling rate is 20GHz, so its effective frequency components are

[0047] S3.2, WOLA structure divides the full bandwidth of the acquisition system into m = 160 narrow bandwidth channels in the digital domain. The WOLA structure is a 1-input 160-output structure. The input is the full bandwidth digital sampling signal, and the corresponding output is 160 narrow bandwidth sampling data;

[0048] In this embodiment, according to the real-time sampling rate f of the acquisition system s = 20G and the digital signal processing frequency f in the digital domain fpga = 250MHz to decide to divide the full bandwidth into m channels, where the number of channels m satisfies:

[0049]

[0050] The bandwidth of each channel is:

[0051] The i-th channel M i The frequency range is:

[0052] M i={125MHz*i, 125MHz*(i+1)},0≤i≤m-1

[0053] In the WOLA structure, the digital sampling signal is channelized digitally filtered by the channelized digital filter in each channel, so that the input full-bandwidth sampling data is evenly distributed in m=160 channels; in this embodiment, each channel has a corresponding channelized digital filter with a bandwidth of 125MHz, and the corresponding channelized digital filter of each channel is a bandpass digital filter, and its passband range is the frequency range M of the channel. i ={125MHz*i, 125MHz*(i+1)},0≤i≤m-1;

[0054] In each channel, the channelized digital filtering result is digitally mixed with a digital single-tone signal with a frequency of 125MHz*i, and then the high-frequency component after mixing is filtered out by a digital anti-image filter, so that the signal frequency is moved to a low frequency band to achieve spectrum shifting, and finally a low-frequency digital signal is obtained in each channel; wherein, i represents the channel number, i=0,1,…,m-1, and m is the number of channels; in this embodiment, the anti-image filter is a low-pass filter, and the cutoff frequency of the filter is 125MHz. The operations of digital mixing and digital anti-image filtering are completed in the FPGA. After the digital anti-image filtering, the signal of the channel is moved to the frequency band of 0 to 125MHz, so the signals of the m channels are spectrum shifted to obtain m low-frequency signals;

[0055] Finally, the D-rate adaptive extraction of time domain data is performed on each low-frequency digital signal, and the extraction rate and channel bandwidth are And the real-time sampling rate of the acquisition system f s The extraction ratio D satisfies the relationship: In this embodiment, the D value is 80, so that the data after the spectrum shifting is adaptively extracted, and the final output of m=160 channels in the WOLA structure is obtained;

[0056] S4, determining that the effective spectrum component is located in the channel of the WOLA structure;

[0057] Traverse the spectrum distribution feature X norm Each effective spectral component in the norm [k], then the corresponding frequency component is the effective component, where k is the number of the effective spectral component, and f s is the real-time sampling rate of the acquisition system, if Then the i-th channel is the effective spectral component X norm [k] The channel where it is located; in this example, from S3.1, it can be known that the signal is located in the two channels i=4 and i=5;

[0058] S5, according to the channel determined in step S4, select the output signals of the two channels corresponding to i=4 and i=5 in the WOLA structure to perform adaptive short-time Fourier transform, then perform downsampling processing, and finally convert the downsampled data into a time-frequency matrix P norm [i]; In this embodiment, the time-frequency matrix P of the normal signal obtained by the 4th and 5th channels is norm [4] and P norm [5] Figure 4 and Figure 5 As shown;

[0059] S6. The spectrum distribution feature X obtained in the normal detection mode norm And the time-frequency matrix P norm [i] As standard spectrum distribution template and time-frequency matrix template and deployed to FPGA;

[0060] S7, set the digital oscilloscope to abnormality detection mode, and then repeat steps (2) to (5) to obtain the frequency spectrum distribution feature X of the input signal in abnormality detection mode. in ={X in [0],X in [1],…,X in [n],…,X in [N-1]} and the time-frequency matrix P in [i];

[0061] In this embodiment, the frequency spectrum distribution feature X of the input signal in the abnormal detection mode is in like Figure 6 As shown, the time-frequency matrix P of the input signal is obtained for the 4th and 5th channels. in [4] and P in [5] Figure 7 and Figure 8 As shown;

[0062] S8, abnormal judgment of input signal;

[0063] S8.1. Traverse the spectrum distribution feature X norm and X in For each spectral component in X norm and X in As long as there is a set of spectral components that satisfy the relationship: |X in [n]-X norm [n]|>X set , X set=20 is the set threshold, then the input signal is determined to be an abnormal signal, and then the process goes to step S8.3; otherwise, the process goes to step S8.2; in the above example, the signal is determined not to be an abnormal signal in this step, and the process goes to step S8.2 for judgment;

[0064] S8.2. Traversing the time-frequency matrix P norm [i] and P in [i], compare whether the channel numbers i are the same in one-to-one correspondence. If the channel numbers are different, it is determined that the input signal is an abnormal signal, and then the process goes to step S8.3. In this embodiment, the two channel numbers in the WOLA structure in the normal detection mode and the abnormal detection mode are both 4 and 5, so the channel numbers are the same;

[0065] If the channel numbers are the same, the starting point search and cyclic comparison method are used to compare all time-frequency matrices P with the same number. norm [i] and P in For each element in [i], if the comparison results of the elements at a certain position are different, the input signal is determined to be an abnormal signal, and then the process goes to step S8.3; if the comparison results of the time-frequency matrices corresponding to the channel numbers are all the same, the input signal is determined to be a normal signal, and then the process goes to step S8.3;

[0066] Among them, the time-frequency matrix P norm [i] and P in [i] Using the method of starting point search and cyclic comparison, such as Fig. 9 As shown, assuming P norm [i] and P in The size of [i] is J×T. We first compare P norm [4] and P in [4], then compare P norm [5] and P in [5], if the comparison results of the two sets of time-frequency matrices are the same, the input signal is judged to be a normal signal, otherwise the signal is judged to be abnormal;

[0067] S8.3. Input the judged signal to the host computer for real-time display.

[0068] We use P norm [4] and P in [4] is taken as an example, and the specific comparison process is as follows:

[0069] 1) Time-frequency matrix P norm [4] and P in [4] starting point search;

[0070] The time-frequency matrix P in [i] in the first column P in [i] 1And the time-frequency matrix P norm Compare each column of [4] to find the time-frequency matrix P in All elements in the first column of [4] satisfy The column t 0 As the starting point, then enter step 2), otherwise, the input signal is determined to be an abnormal signal, and the comparison ends; where P set To set the threshold, the value is 20, P in [4] 1,v Represents the time-frequency matrix P in [4] The vth element in the first column. In this embodiment, by comparison, it can be found that P in The first column in [4] is the starting point;

[0071] 2) Perform cyclic comparison based on the starting point;

[0072] P in [i] in column 1 and P norm [4] 0 Column comparison, P in The second column and P in [i] norm [4] 0 +1 column for comparison, and so on, when the comparison reaches P norm [i] and then return to P norm Continue comparing the first column of [4] until the time-frequency matrix P norm [4] and P in All columns of [4] are compared. During the comparison process, if any column has an element that satisfies |P in [4] 1,v -P norm [4] t,v |>P set When P in [4] and P norm [4] is different. In this example, P in [4] and P norm The second half of [4] is different, so the signal is judged to be an abnormal signal.

[0073] Although the above describes the illustrative specific embodiments of the present invention to facilitate those skilled in the art to understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope, characterized in that: The following steps are involved: (1) The digital oscilloscope is powered on and initialized, and then the input signal is connected; (2) In the normal detection mode, the input signal is sampled by the acquisition system in the digital oscilloscope to obtain a digital sampling signal; (3) Input the digital sampling signal into the FFT unit and the WOLA structure respectively; (3.1) Perform FFT operation on the digital sampling signal in the FFT unit to obtain the spectrum distribution feature X norm ={X norm [0],X norm [1],…,X norm [n],…,X norm [N-1]}, N is the number of points of FFT operation; Traverse the spectrum distribution feature X norm Each spectral component X in norm [n], mark the spectral component X norm [n] The spectrum components that are greater than the threshold δ are valid spectrum components; (3.2) In the WOLA structure, the digital sampling signal is channelized digitally filtered by the channelized digital filter in each channel and then filtered with the frequency The digital single tone signal is digitally mixed, and finally digital anti-image filtering is performed through a digital anti-image filter, so that a low-frequency digital signal is obtained in each channel; where i represents the channel number, i = 0, 1, ..., m-1, m is the number of channels, and f s is the real-time sampling rate of the acquisition system; Finally, the D-rate adaptive extraction of time domain data is performed on each low-frequency digital signal, and the output signal of each channel is finally obtained; (4) Determine the channel where the effective spectrum component is located in the WOLA structure; Traverse the spectrum distribution feature X norm Each effective spectral component in the norm [k], then the corresponding frequency component is the effective component, where k is the number of the effective spectral component, and f s is the real-time sampling rate of the acquisition system, if Then the i-th channel is the effective spectral component X norm [k] the channel it is located on; (5) According to the channel determined in step (4), the output signal of the corresponding channel i in the WOLA structure is selected for adaptive short-time Fourier transform, and then down-sampled. Finally, the down-sampled data is converted into a time-frequency matrix P norm [i]; (6) The spectrum distribution feature X obtained in the normal detection mode norm And the time-frequency matrix P norm [i] As standard spectrum distribution template and time-frequency matrix template and deployed to FPGA; (7) Set the digital oscilloscope to the abnormality detection mode, and then repeat steps (2) to (5) to obtain the frequency spectrum distribution feature X of the input signal in the abnormality detection mode. in ={X in [0],X in [1],…,X in [n],…,X in [N-1]} and the time-frequency matrix P in [i]; (8) Abnormal judgment of input signal; (8.1), traverse the spectrum distribution feature X norm and X in For each spectral component in X norm and X in As long as there is a set of spectral components that satisfy the relationship: |X in [n]-X norm [n]|>X set , X set To set the threshold, the input signal is determined to be an abnormal signal, and then go to step (8.3); otherwise, go to step (8.2); (8.2), traverse the time-frequency matrix P norm [i] and P in [i], compare whether the channel numbers i are the same one-to-one. If the channel numbers are different, the input signal is determined to be an abnormal signal, and then go to step (8.3); if the channel numbers are the same one-to-one, use the starting point search and cyclic comparison method to compare all the time-frequency matrices P with the same number norm [i] and P in For each element in [i], if the comparison results of the elements at a certain position are different, the input signal is determined to be an abnormal signal, and then the process goes to step (8.3); if the comparison results of the time-frequency matrices corresponding to the channel numbers are all the same, the input signal is determined to be a normal signal, and then the process goes to step (8.3); (8.3) The judged signal is input to the host computer for real-time display.

2. The method for real-time anomaly detection based on time-frequency domain in an intelligent oscilloscope according to claim 1, characterized in that: The WOLA structure divides the full bandwidth of the acquisition system into m narrow bandwidth channels in the digital domain. The WOLA structure is a 1-input and m-output structure. The input is a full-bandwidth digital sampling signal, and the corresponding output is m narrow-bandwidth sampling data. The number of channels m satisfies: Among them, f s is the real-time sampling rate of the acquisition system, f fpga is the digital signal processing frequency in the digital domain; The bandwidth of each channel is: The i-th channel M i The frequency range is:

3. The real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope according to claim 1, characterized in that: The bandwidth corresponding to the channelized digital filter in each channel is: The channelized digital filter corresponding to each channel is a bandpass digital filter, and its passband range is the frequency range of the channel.

4. The real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope according to claim 1, characterized in that: The anti-imaging filter is a low-pass filter with a cut-off frequency of 5. The real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope according to claim 1, characterized in that: The extraction ratio D satisfies:

6. The real-time anomaly detection method based on time-frequency domain in an intelligent oscilloscope according to claim 1, characterized in that: The time-frequency matrix P norm [i] and P in [i] The method of using the starting point search and cyclic comparison is: (6.1), the time-frequency matrix P norm [i] and P in [i] start point search; The time-frequency matrix P in [i] in the first column P in [i]1 and time-frequency matrix P norm Compare each column of [i] to find the time-frequency matrix P in All elements in the first column of [i] satisfy The column t0 of is taken as the starting point, and then enter step (6.2), otherwise, the input signal is determined to be an abnormal signal, and the comparison ends; where P set To set the threshold, P in [i] 1,v Represents the time-frequency matrix P in [i] the vth element in the first column; (6.2) Perform cyclic comparison based on the starting point; P in [i] in column 1 and P norm [i] t0 column, P in The second column and P in [i] norm [i] is compared with the t0+1th column, and so on. norm [i] and then return to P norm The first column of [i] continues to be compared until the time-frequency matrix P norm [i] and P in All columns of [i] are compared. If any column has an element that satisfies When P in [i] and P norm [i] is different, the input signal is determined to be an abnormal signal, and the comparison ends.

Images