CN116359851A - Radar active interference detection and identification method and device based on converged network - Google Patents
Radar active interference detection and identification method and device based on converged network Download PDFInfo
- Publication number
- CN116359851A CN116359851A CN202210009977.4A CN202210009977A CN116359851A CN 116359851 A CN116359851 A CN 116359851A CN 202210009977 A CN202210009977 A CN 202210009977A CN 116359851 A CN116359851 A CN 116359851A
- Authority
- CN
- China
- Prior art keywords
- interference
- frequency
- signal
- time
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 54
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000001228 spectrum Methods 0.000 claims abstract description 67
- 238000012545 processing Methods 0.000 claims abstract description 43
- 230000004927 fusion Effects 0.000 claims abstract description 42
- 238000013135 deep learning Methods 0.000 claims abstract description 31
- 238000001914 filtration Methods 0.000 claims abstract description 17
- 238000004088 simulation Methods 0.000 claims abstract description 11
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 9
- 238000012549 training Methods 0.000 claims description 47
- 239000011159 matrix material Substances 0.000 claims description 43
- 238000010586 diagram Methods 0.000 claims description 27
- 238000006243 chemical reaction Methods 0.000 claims description 23
- 238000007781 pre-processing Methods 0.000 claims description 19
- 238000009826 distribution Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 13
- 238000012567 pattern recognition method Methods 0.000 claims description 10
- 238000005520 cutting process Methods 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 9
- 238000005516 engineering process Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000004519 manufacturing process Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 4
- 238000003909 pattern recognition Methods 0.000 claims description 4
- 230000000737 periodic effect Effects 0.000 claims description 4
- 230000007613 environmental effect Effects 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 2
- 238000009499 grossing Methods 0.000 claims description 2
- 239000000758 substrate Substances 0.000 claims description 2
- 230000007246 mechanism Effects 0.000 abstract description 2
- 238000000605 extraction Methods 0.000 description 5
- 238000012935 Averaging Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- RVRCFVVLDHTFFA-UHFFFAOYSA-N heptasodium;tungsten;nonatriacontahydrate Chemical compound O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.[Na+].[Na+].[Na+].[Na+].[Na+].[Na+].[Na+].[W].[W].[W].[W].[W].[W].[W].[W].[W].[W].[W] RVRCFVVLDHTFFA-UHFFFAOYSA-N 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 230000008707 rearrangement Effects 0.000 description 2
- 238000007711 solidification Methods 0.000 description 2
- 230000008023 solidification Effects 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035485 pulse pressure Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/36—Means for anti-jamming, e.g. ECCM, i.e. electronic counter-counter measures
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/40—Means for monitoring or calibrating
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention relates to a radar active interference detection, type and pattern intelligent identification method and device based on a secondary deep learning fusion network, and belongs to the technical field of signal processing. The invention adopts a method combining the image and the signal processing, and detects the interference signal by adopting an average filtering algorithm and threshold signal detection to obtain whether interference exists or not and the multi-domain parameters of the interference signal, thereby not only increasing the robustness of interference identification, but also providing a basis for anti-interference decision. Through a fusion network of a residual error network (ResNet) and a long-short-time memory network (Attention-LSTM) based on an Attention mechanism, multidimensional features are automatically extracted from a time-frequency image and a frequency spectrum of an interference signal, so that the accurate identification of the interference type and pattern is realized, and the simulation result proves the advancement and feasibility of the method.
Description
Technical Field
The invention relates to the technical field of signal processing, in particular to the technical field of radars, and particularly relates to a method and a device for radar active interference signal detection, parameter extraction and interference type and pattern recognition.
Background
According to an interference mechanism, radar active interference can be divided into two types of suppression interference and deception interference, wherein the suppression interference mainly utilizes strong noise interference to submerge a target echo signal, and the radar receiver cannot normally detect the target echo signal through a large interference signal ratio; the deception jamming is to construct an interfering signal strongly related to the radar signal, and after pulse pressure processing, a plurality of false targets which cannot be distinguished by the radar system are generated, so that the radar cannot distinguish between true targets and false targets.
With the increasing complexity of electronic countermeasure, especially the development of DRFM technology, the capability of active interference is greatly enhanced. However, the radar can only make a corresponding anti-interference strategy based on the accurate identification of the interference signal. At present, the traditional method for identifying the active interference mainly designs characteristic parameters with obvious distinction from the fields of time domain, frequency domain, time-frequency domain, transform domain and the like, and classifies the characteristic parameters through a preset classifier. The method has the problems of large difficulty in designing characteristic parameters, general recognition effect, difficulty in adapting to the situation of large data volume, large difference of recognition results of different interference patterns and the like, and further development is limited to a great extent.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides a radar active interference signal detection, interference type and pattern recognition method and device which can improve the accuracy of interference signal recognition and are provided with interference detection and interference parameter extraction.
In order to achieve the above object, the radar active interference signal detection, parameter extraction and interference type and pattern recognition method of the present invention comprises the steps of:
(1) The down-conversion processing module demodulates the digital intermediate frequency signal into a zero intermediate frequency signal by using a quadrature demodulation technology;
(2) Carrying out frequency domain processing on the zero intermediate frequency signal, obtaining whether interference exists or not and parameters of interference signal frequency domain, airspace and energy domain through mean value filtering and interference detection, and giving an enabling signal of a preprocessing module;
(3) The zero intermediate frequency signal is subjected to frequency domain transformation on the first path on the basis of receiving an enabling signal to obtain a frequency spectrum, and normalization and standardization are carried out; performing time-frequency transformation on the second path to obtain a time-frequency matrix, performing gray level discretization on the time-frequency matrix, and adaptively cutting the time-frequency matrix to a proper size; outputting the two paths of results to a secondary deep learning fusion network;
(4) The two-stage deep learning fusion network mainly comprises a ResNet network for identifying images, an Attention-LSTM network for identifying sequences and a multi-layer perceptron network for fusing the outputs of the two networks. Training the secondary deep learning fusion network through the training set to obtain a training model and solidification parameters thereof. And finally realizing the real-time identification of the type and the style of the interference signal through the training model by using the frequency spectrum and the time-frequency diagram of the preprocessing link.
In the radar active interference detection, type and pattern intelligent identification method, the step (1) specifically comprises the following steps:
the down-conversion processing module mixes the input digital intermediate frequency signals with cosine signals and sine signals respectively, and then obtains IQ two paths of quadrature signals through a low-pass filter.
In the radar active interference detection, type and pattern intelligent identification method, the step (2) comprises the following steps:
(21) Processing the zero intermediate frequency signal by adopting a periodic graph method to obtain an interference signal power spectrum;
(22) The interference signal spectrum is subjected to mean value filtering treatment, environmental noise in the signal is filtered, spectrum jitter in a signal passband is reduced, and a smooth interference signal power spectrum is obtained;
(23) Processing the smooth interference signal power spectrum by adopting a double-threshold signal detection algorithm, and providing a judgment for the existence of an interference signal;
(24) If no interference signal exists, returning to the power spectrum estimation module to process the next pulse; if a plurality of interference exists, the interference signal energy domain and frequency domain parameters are calculated, the direction of the interference source is measured, the interference source azimuth information is obtained, and meanwhile an enabling signal of the preprocessing module is given.
In the intelligent identification method for the radar active interference detection, type and pattern, the real-time power spectrum estimation method in the step (21) is as follows: by means ofFFT conversion to obtain interference signal spectrum, squaring the modulus value, and averaging to obtain interference signal power spectrum P w (n)。
In the radar active interference detection, type and pattern intelligent identification method, the step (22) specifically comprises the following steps: setting an average filtering window T f For P w (n) performing an average filtering operation to obtain a smoothed interference signal spectrum
In the intelligent identification method for the radar active interference detection, type and pattern, the double-threshold detection algorithm of the step (23) comprises the following steps:
(A1) Determining an upper threshold factor and a lower threshold factor according to the false alarm rate and the ground clutter distribution;
(A2) Acquiring initial threshold power according to the noise substrate;
(A3) Updating the upper threshold power and the lower threshold power in real time by adopting a recursion method;
(A4) Searching a sequence, and considering that interference occurs if the continuous H points are larger than an upper threshold value;
(A5) Continuing searching the sequence, and when the continuous H points are smaller than a lower threshold value, considering that the interference is ended;
in the intelligent radar active interference detection, type and pattern recognition method, the parameter extraction in the step (24) is specifically as follows:
(B1) Calculation of P w The maximum and average values of (n) as the interference signal peak power and average power;
(B2) The upper and lower threshold powers given by the double threshold detection are used for calculating frequency domain parameters such as bandwidth, center frequency and the like of the interference signal;
in the radar active interference detection, type and pattern intelligent identification method, the step (3) comprises the following steps:
(31) FFT transforming the orthogonal zero intermediate frequency interference signal to obtain the frequency spectrum of the signal, normalizing the frequency spectrum, and executing
(x (n) is a normalized spectrum, mu is the mean value of x (n), sigma is the standard deviation of x (n), and the normalized spectrum is normalized to the same distribution and is input into an LSTM branch network in a fusion network;
(32) Performing STFT (standard time-shift transform) on orthogonal zero intermediate frequency interference signals, obtaining a signal time-frequency matrix by taking a module value of the signal time-frequency matrix, linearly dispersing the signal time-frequency matrix into 0-255 to obtain a gray image, adaptively cutting the gray image into 128 pixels by 128 pixels, performing normalization processing, and inputting the 128 pixels into a ResNet branch network in a fusion network;
in the radar active interference detection, type and pattern intelligent identification method, normalization processing in the steps (31) and (32) is specifically as follows: and (3) obtaining the maximum value and the minimum value of the sequence or the matrix, and obtaining the normalized value of the (sequence (matrix) -minimum value)/(maximum value-minimum value).
The STFT conversion in the step (32) specifically includes: and intercepting and multiplying the zero intermediate frequency interference J (n) by adopting a window function with a set length, and then carrying out one-dimensional FFT (fast Fourier transform). A series of one-dimensional cut-off frequency spectrums are obtained through sliding of the window function, and the one-dimensional cut-off frequency spectrums are arranged according to a time axis, so that a two-dimensional time-frequency diagram matrix can be obtained. The calculation formula is as follows:
where STFT (·) is the time-frequency plot matrix, h (·) is the selected window function,
is a twiddle factor.
The self-adaptive cutting in the step (32) is specifically as follows:
(C1) Smoothing and filtering the time-frequency diagram matrix, and filtering background noise to obtain a smoothed time-frequency diagram matrix;
(C2) Determining a self-adaptive clipping lower threshold T according to the calculated time-frequency matrix mean value image ；
(C3) Searching from two ends to the center of the image in the time dimension and the frequency dimension respectively to find outThe first is greater than T image The rows and columns are marked as w1, w2, h1 and h2 and are used as self-adaptive clipping areas;
(C4) Adjusting the adaptive clipping region to 128 pixels by 128 pixels using bicubic interpolation;
in the radar active interference detection, type and pattern intelligent identification method, the step (4) comprises the following steps:
(41) And setting the interference type and pattern to be identified, and designing and manufacturing training and testing simulation data set samples.
(42) And adjusting the depth and the breadth of the ResNet network and the attribute-LSTM parameter scale according to the interference pattern class number and the training set sample number so as to match the ResNet network depth and the breadth with each other.
(43) The training set and the testing set are input into a two-stage deep learning fusion network, network super parameters are set, weight parameters in the network are adjusted through a gradient descent method, and finally a training model is obtained.
(44) The radar receives the interference signal in real time, converts the interference signal to an intermediate frequency, and outputs the type and the style of the interference signal through a preprocessing link and a training model after down-conversion processing.
In the radar active interference detection, type and pattern intelligent identification method, the simulation data set in the step (41) is specifically as follows: by setting parameters such as different bandwidths, duty ratios, dry-to-noise ratios, temperatures, slice numbers, false target numbers, sweep frequency speeds, frequency modulation slopes, amplitude modulation slopes and the like, training and testing simulation data sets are manufactured, and the sample number proportion is 18000:3600.
the depth and breadth of the ResNet network in the step (42) are respectively: depth, namely the stacking number of residual modules, and selecting proper number according to task requirements; the breadth, i.e., the number of ResNet network feature maps, characterizes the ability to extract image features. The parameter scale is specifically as follows: the total number of units and the number of memory units of the Attention-LSTM, each unit calculates the output of the current moment by calculating the input of the current moment and combining the results of all the effective memory units in the past.
The network super parameters in the step (43) are specifically as follows: learning rate, lot number, training period, loss function, regularization parameters, etc. affect parameters of the network training. The training model is specifically as follows: after training the training set through the secondary deep learning fusion network, determining the trained network of all weight parameters in the network.
The invention also provides a radar active interference detection, type and pattern intelligent identification device for realizing the method, which comprises the following steps:
the down-conversion processing module is used for demodulating the digital intermediate frequency signal to a zero intermediate frequency signal by using a quadrature demodulation technology;
the interference detection module is used for carrying out frequency domain processing on the zero intermediate frequency signal, detecting whether interference exists or not, and obtaining parameters such as interference power, azimuth and bandwidth;
the preprocessing module is used for carrying out FFT conversion and STFT conversion on the zero intermediate frequency signal to obtain a frequency spectrum and a time-frequency matrix of the zero intermediate frequency signal; and respectively carrying out normalization and standardization processing on the frequency spectrum, and carrying out linear graying processing and self-adaptive clipping processing on the time-frequency matrix.
And the second-level deep learning fusion network module automatically extracts the interference signal characteristics from the output of the preprocessing module through a training model and outputs the type and the style of interference.
The intelligent identification method and device for the radar active interference detection, type and pattern adopt a method of combining image and signal processing, and the interference signal is detected by adopting an average filtering algorithm and threshold signal detection, so that the existence of interference and multi-domain parameters of the interference signal are obtained. The method separates training and testing of the fusion network, is more beneficial to engineering realization, and has the capability of detecting and identifying multiple interferences in real time.
Drawings
Fig. 1 is a schematic diagram of a radar active interference detection, type and pattern intelligent recognition method of the present invention.
Fig. 2 is a schematic diagram of an intelligent radar active interference detection, type and pattern recognition device provided by the invention.
Fig. 3 is a simulation diagram of comb spectrum interference double threshold detection in an embodiment provided by the invention.
Fig. 4 is a schematic diagram of a time-frequency branch preprocessing according to the present invention.
Fig. 5 is a block-interference signal time domain waveform, spectrum and time-frequency distribution diagram according to an embodiment of the present invention.
Fig. 6 is a time domain waveform, spectrum and time-frequency distribution diagram of a swept interference signal according to an embodiment of the invention.
Fig. 7 is a time domain waveform, spectrum and time-frequency distribution diagram of an aiming interference signal according to an embodiment of the present invention.
Fig. 8 is a time domain waveform, spectrum and time-frequency distribution diagram of a comb spectrum interference signal according to an embodiment of the present invention.
Fig. 9 is a time domain waveform, spectrum and time-frequency distribution diagram of dense decoy interference signals according to an embodiment of the present invention.
Fig. 10 is a time domain waveform, spectrum and time-frequency distribution diagram of a slice interference signal according to an embodiment of the present invention.
Fig. 11 is a diagram of a network structure of a res net in an embodiment provided by the present invention.
Fig. 12 is a schematic diagram of an attribute-LSTM network in an embodiment provided by the present invention.
Fig. 13 is an overall structure diagram of a two-level deep learning fusion network in an embodiment provided by the invention.
Fig. 14 is a performance diagram of two-level deep learning fusion network simulation recognition in an embodiment provided by the invention.
Detailed Description
In order to make the technical contents of the present invention more clearly understood, the following examples are specifically described.
Fig. 1 is a schematic diagram of a radar active interference detection, type and pattern intelligent recognition method based on a two-level deep learning fusion network.
In one embodiment, the radar active interference detection, type and pattern intelligent identification method comprises the following steps:
(1) The down-conversion processing module demodulates the digital intermediate frequency signal into a zero intermediate frequency signal by using a quadrature demodulation technology;
(2) Carrying out frequency domain processing on the zero intermediate frequency signal, obtaining whether interference exists and parameters of interference signal frequency domain, airspace and energy domain through mean value filtering and interference detection, and giving an enabling signal of the work of the preprocessing module;
(3) The zero intermediate frequency signal is subjected to frequency domain transformation on the first path on the basis of receiving an enabling signal to obtain a frequency spectrum of the zero intermediate frequency signal, and normalization and standardization processing are carried out; performing time-frequency transformation on the second path to obtain a time-frequency matrix, performing gray level discretization on the time-frequency matrix, and adaptively cutting the time-frequency matrix to a proper size; outputting the two paths of results to a secondary deep learning fusion network;
(4) The two-level deep learning fusion network comprises a ResNet network for identifying images, an Attention-LSTM network for identifying sequences and a multi-layer perceptron network for fusing the outputs of the two networks. Training the secondary deep learning fusion network through the training set to obtain a training model and solidification parameters thereof. And finally realizing intelligent identification of the type and the style of the interference signal through the training model by using the frequency spectrum and the time-frequency diagram of the preprocessing link.
The invention also provides a corresponding radar active interference detection, type and pattern intelligent identification device for realizing the method, as shown in fig. 2, the device comprises:
the receiving equipment is used for receiving radio frequency signals of radar interference and converting the radio frequency signals into digital intermediate frequency signals;
the down-conversion processing module is used for demodulating the digital intermediate frequency signal to a zero intermediate frequency signal by using a quadrature demodulation technology;
the interference detection module is used for carrying out frequency domain processing on the zero intermediate frequency signal, detecting whether interference exists or not, and obtaining parameters such as interference power, azimuth and bandwidth;
the preprocessing module is used for carrying out FFT conversion and STFT conversion on the zero intermediate frequency signal to obtain a frequency spectrum and a time-frequency matrix of the zero intermediate frequency signal; and respectively carrying out normalization and standardization processing on the frequency spectrum, and carrying out linear graying processing and self-adaptive clipping processing on the time-frequency matrix.
And the second-level deep learning fusion network module automatically extracts the interference signal characteristics from the output of the preprocessing module through a training model and outputs the type and the style of interference.
The step (1) specifically comprises the following steps: the down-conversion process respectively combines the digital intermediate frequency signal with the cosine signal cos (2pi f 0 t) and a sine signal sin (2 pi f) 0 t) multiplying to mix frequency, and then passing through a low-pass filter H L (x) Obtaining IQ two paths of orthogonal signals;
as shown in fig. 3, the step (2) specifically includes:
(21) The periodic diagram method is adopted to carry out power spectrum estimation on the zero intermediate frequency signal, namely
Where J (n) is the interfering signal, FFT (& gt) represents FFT transform, P w (n) is an estimated power spectrum.
(22) For said P w (n) setting the average filter window size T f The average filtering process is carried out to filter the environmental noise in the signal, reduce the power spectrum jitter in the signal passband and obtain the smooth interference signal power spectrum
(23) And processing the smooth interference signal power spectrum by adopting a double-threshold signal detection algorithm, detecting whether an interference signal exists, and calculating frequency domain parameters such as bandwidth, center frequency and the like of the interference signal by using the double-threshold.
The double threshold detection algorithm of the step (23) comprises the following steps:
(A1) Determining a lower threshold factor T according to the false alarm probability requirement low And an upper threshold factor T high ；
Taking the example that the ground clutter amplitude accords with Rayleigh distribution, wherein T is a threshold factor and sigma 2 Is the variance of the pure noise spectrum, P fa Is the required false alarm probability.
(A2) Taking the mean value of the interference-free smooth power spectrum of the previous group (or the initial section of the group) as an initial threshold P E ；
(A3) Will be
Less than T low ×P E To obtain a new threshold P 'by averaging the data of (2)' E Repeating until the new threshold is equal to the original threshold or the maximum iteration number is reached, wherein the lower threshold is T low ×P′ E The upper threshold is T high ×P′ E ；
(A4) Searching a sequence, and considering that interference occurs if the continuous H points are larger than an upper threshold value;
(A5) Continuing searching the sequence, and when the continuous H points are smaller than a lower threshold value, considering that the interference is ended;
(24) If no interference signal exists, returning to the power spectrum estimation module to process the next interference signal; if a plurality of interference exists, energy domain and frequency domain parameters of the interference signals are calculated, direction finding is carried out on the interference sources, the azimuth information of the interference sources is obtained, and meanwhile an enabling signal of the preprocessing module is given.
The extracting parameters in the step (24) comprises the following steps:
(B1) From interference signal power spectrum P w (n) calculating the peak power and average power thereof;
(B2) Searching
Calculate->
Middle is greater than->
Record the initial frequency f 1 And termination frequency f 2 From B J ＝f 2 -f 1 And->
The bandwidth and center frequency of the interference signal can be calculated;
the step (3) is specifically as follows:
(31) FFT conversion is carried out on the orthogonal zero intermediate frequency interference signals to obtain the frequency spectrum of the signals, and after the normalization processing is carried out on the frequency spectrum, the method is carried out
(x (n) is a normalized spectrum, mu is the mean value of x (n), sigma is the standard deviation of x (n), and the normalized spectrum is normalized to the same distribution and is input into an LSTM branch network in a fusion network;
(32) As shown in fig. 4, taking comb spectrum interference as an example, performing STFT conversion on orthogonal zero intermediate frequency interference signals, taking a module value of the signal time-frequency matrix, linearly dispersing the signal time-frequency matrix into 0-255 to obtain a gray image, adaptively cutting the gray image into 128 pixels by 128 pixels, performing normalization processing, and inputting the 128 pixels into a ResNet branch network in a fusion network;
in the radar active interference detection, type and pattern intelligent identification method, normalization in the steps (31) and (32) is specifically as follows:
wherein X is the frequency spectrum of the interference signal or the time-frequency diagram after clipping, max is the maximum value of X, min is the minimum value of X, and Y is normalized output.
The STFT conversion in the step (32) specifically includes: and intercepting and multiplying the zero intermediate frequency interference J (n) by adopting a window function with a set length, and then carrying out one-dimensional FFT (fast Fourier transform). A series of one-dimensional cut-off frequency spectrums are obtained through sliding of the window function, and the one-dimensional cut-off frequency spectrums are arranged according to a time axis, so that a two-dimensional time-frequency diagram matrix can be obtained. The calculation formula is as follows:
where STFT (·) is the time-frequency plot matrix, h (·) is the selected window function,
is a twiddle factor.
In the step (32), the self-adaptive cutting is specifically as follows:
(C1) Adopting a two-dimensional uniform filtering operator to carry out smooth filtering on the time-frequency matrix, and mainly filtering background noise to obtain a smooth time-frequency graph matrix;
(C2) Calculating the maximum value M of the time-frequency matrix max And mean value M E Determining the threshold value of the adaptive clipping as T image ＝max(α×(1+e -JNR )×M max ,β×(1+e -JNR )×M E ) Where α=0.27, β=1.75;
(C3) Searching from two ends to the center of the image in the time dimension and the frequency dimension respectively to find out that the first is larger than T image The rows and columns are marked as w1, w2, h1 and h2 and are used as self-adaptive clipping areas;
(C4) Adjusting the adaptive clipping region to 128 pixels by 128 pixels using bicubic interpolation;
the step (4) specifically comprises the following steps:
(41) Setting interference types and patterns to be identified, and designing and manufacturing training data set samples;
(42) According to the interference pattern class number and the training set sample number, the depth and the breadth of the ResNet network and the attribute-LSTM network parameter scale are adjusted to be matched with each other;
(43) Inputting the manufactured training set and test set into a secondary deep learning fusion network, setting network super-parameters, adjusting the weights of all layers in the network by a gradient descent method, and finally obtaining a training model;
(44) The radar receives the interference signal in real time, converts the interference signal to an intermediate frequency, performs down-conversion processing, and outputs the type and the style of the interference signal through a preprocessing link and a training model.
The designing and manufacturing the data set in the step (41) specifically includes: taking blocking interference, aiming interference, sweep interference, dense false target interference, comb spectrum interference and slicing interference as examples, training and testing data set samples are designed and manufactured, and part of the samples are shown in fig. 5-10.
The depth and breadth of the ResNet in step (42) are as shown in FIG. 11, specifically: depth refers to the number of convolution layers and fully connected layers, characterizing the nonlinear fitting capability of the network, while breadth refers to the number and size of convolution kernels, characterizing the feature extraction capability. The attribute-LSTM network parameter scale, as shown in fig. 12, is specifically: the total number of units and the number of effective memory units of the attribute-LSTM, each unit calculates the output of the current moment by calculating the input of the current moment and combining the results of all the effective memory units in the past.
In the second-stage fusion network in the step (43), as shown in fig. 13, the first-stage is respectively used for extracting time-frequency image features by the ResNet network and spectrum features by the Attention-LSTM network, splicing the features at the full-connection layer, further extracting the features by the second-stage multi-layer perceptron network, and finally converting the features into probability distribution of an interference pattern by a softmax function, thereby completing intelligent identification of active interference. In addition, the network super parameters are specifically: parameters affecting the network training, such as learning rate, number of lots, training period, loss function, regularization parameters, etc., are shown in table 1 below.
| Super parameter | Numerical value | Super parameter | Numerical value |
| Learning rate | 0.005 | Learning rate decline speed | 0.5/ |
| Batch number | |||
| 60 | |
100 | |
| Loss function | RMSprop | Rearrangement mode | Periodic rearrangement |
Table 1 converged network superparameter settings
The simulation data set was trained and tested according to the network structure of fig. 12 and the super parameter settings of table 1, and the simulation results are shown in fig. 14. Where fig. 14 (a) is a graph of recognition accuracy of the training set and the test set, and fig. 14 (b) is a sample recognition confusion matrix of the test set. From the accuracy curve, the two can be well matched, which shows that the depth and the breadth of the selected fusion network are good in matching degree with the data set. The test set identification accuracy reaches 99.94%, which shows that the invention has excellent identification performance. The numerical value of the confusion matrix indicates that the fusion network can well identify each interference pattern.
The intelligent identification method and the intelligent identification device for the radar active interference detection, type and pattern adopt a double-threshold signal detection algorithm to accurately detect the interference signal in real time and extract the parameters of the signal frequency domain, the signal airspace and the signal energy domain; the method of combining the image and the signal processing is adopted, the time-frequency distribution map and the interference sequence characteristics are automatically extracted by a secondary deep learning fusion network, the robustness of interference identification is ensured under the early stage of improving the interference identification rate, and the feasibility of the method is proved.
In this specification, the invention has been described with reference to specific embodiments thereof. It will be apparent that various modifications and variations can be made without departing from the spirit and scope of the invention, such as the use of LSTM variant networks, such as pepole LSTM, GRU, bi-LSTM, multilayer LSTM and variant networks like Attention-LSTM, etc.; or varying ResNet network depth, feature pattern numbers and employing variant structures such as ResNet-50, resNet101, resNeXt, resNeSt, SE-ResNet, etc. The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense.
Claims (10)
1. A radar active interference detection, type and pattern intelligent identification method based on a two-level deep learning fusion network is characterized by comprising the following steps:
(1) The down-conversion processing module demodulates the digital intermediate frequency signal into a zero intermediate frequency signal by using a quadrature demodulation technology;
(2) Carrying out frequency domain processing on the zero intermediate frequency signal, obtaining whether interference exists or not and parameters of interference signal frequency domain, airspace and energy domain through mean value filtering and interference detection, and giving an enabling signal of a preprocessing module;
(3) The zero intermediate frequency signal is subjected to frequency domain transformation on the first path on the basis of receiving an enabling signal to obtain a frequency spectrum of the zero intermediate frequency signal, and normalization and standardization processing are carried out; performing time-frequency transformation on the second path to obtain a time-frequency matrix, performing gray level discretization on the time-frequency matrix, and adaptively cutting the time-frequency matrix to a proper size; outputting the two paths of results to a secondary deep learning fusion network;
(4) The secondary deep learning fusion network mainly comprises a ResNet network for identifying images, an Attention-LSTM network for identifying sequences and a multi-layer perceptron network for fusing the outputs of the two networks, wherein the secondary deep learning fusion network is trained through a training set, model parameters are solidified to obtain a training model, and the frequency spectrum and time-frequency diagram of a preprocessing link are subjected to intelligent identification of the type and the style of an interference signal through the training model.
2. The intelligent radar active interference detection, type and pattern recognition method based on the two-level deep learning fusion network according to claim 1, wherein the step (1) is specifically as follows:
the down-conversion processing module mixes the input digital intermediate frequency signals with cosine signals and sine signals respectively, and then obtains IQ two paths of quadrature signals through a low-pass filter.
3. The intelligent radar active interference detection, type and pattern recognition method based on the two-level deep learning fusion network according to claim 1, wherein the step (2) comprises the following steps:
(21) Processing the zero intermediate frequency signal by adopting a periodic graph method to obtain an interference signal power spectrum;
(22) The average value filtering processing is carried out on the interference signal power spectrum, the environmental noise in the signal is filtered, the jitter in the signal passband is reduced, and the smooth interference signal power spectrum is obtained;
(23) Processing the smooth interference signal power spectrum by adopting a double-threshold signal detection algorithm to provide a decision for whether an interference signal exists or not;
(24) If no interference signal exists, returning to the power spectrum estimation module to process the next pulse; if a plurality of interference exists, interference power and bandwidth parameters are calculated, direction finding is carried out on an interference source, interference source azimuth information is obtained, and an enabling signal of the work of the preprocessing module is given.
4. The method for intelligent radar active interference detection, type and pattern recognition based on the two-level deep learning fusion network according to claim 3,
the double threshold signal detection algorithm in the step (23) comprises the following steps:
(A1) Determining an upper threshold factor and a lower threshold factor according to the false alarm rate and the ground clutter distribution;
(A2) Acquiring initial threshold power according to the noise substrate;
(A3) Updating the upper threshold power and the lower threshold power in real time by adopting a recursion method;
(A4) Searching a sequence, and considering that interference occurs if the continuous H points are larger than an upper threshold value;
(A5) Continuing the search sequence, when the continuous H points are smaller than the lower threshold value, the interference is considered to be ended,
the interference power and bandwidth parameters calculated in the step (24) are specifically:
(B1) Calculation of P w The maximum and average values of (n) as the interference signal peak power and average power;
(B2) And calculating frequency domain parameters such as bandwidth, center frequency and the like of the interference signal by using the upper and lower threshold powers given by the double threshold detection.
5. The intelligent radar active interference detection, type and pattern recognition method based on the two-level deep learning fusion network according to claim 1, wherein the step (3) comprises the following steps:
(31) Performing Fast Fourier Transform (FFT) on the orthogonal zero intermediate frequency interference signals to obtain the frequency spectrum of the signals, normalizing the frequency spectrum, and executing
Wherein x (n) is a normalized spectrum, mu is the mean value of x (n), sigma is the standard deviation of x (n), and the normalized spectrum is normalized to the same distribution and is input into an LSTM branch network in a fusion network;
(32) Performing short-time Fourier time-frequency transformation (STFT) on the orthogonal zero intermediate frequency interference signals, taking a signal time-frequency matrix of a modular value of the signal time-frequency transformation, linearly dispersing the signal time-frequency matrix into 0-255 to obtain a gray image, adaptively cutting the gray image into 128 pixels multiplied by 128 pixels, performing normalization processing, and inputting the 128 pixels into a ResNet branch network in a fusion network.
6. The intelligent radar active interference detection, type and pattern recognition method based on the secondary deep learning fusion network according to claim 5, wherein,
the normalization processing in the step (31) and the step (32) is specifically as follows: solving the maximum value and the minimum value of the sequence or the matrix, and obtaining the normalized value of the (sequence (matrix) -minimum value)/(maximum value-minimum value);
the short-time fourier time-frequency transformation in the step (32) specifically includes: intercepting and multiplying the zero intermediate frequency interference J (n) by adopting a window function with a set length, and then performing one-dimensional fast Fourier transform; a series of one-dimensional cut-off frequency spectrums are obtained through sliding of the window function, the one-dimensional cut-off frequency spectrums are arranged according to a time axis, and a two-dimensional time-frequency diagram matrix is obtained, wherein the calculation formula is as follows:
where STFT (·) is the time-frequency plot matrix, h (·) is the selected window function,
is a twiddle factor.
7. The intelligent radar active interference detection, type and pattern recognition method based on the secondary deep learning fusion network according to claim 6, wherein,
the self-adaptive cutting in the step (32) is specifically as follows:
(C1) Smoothing and filtering the time-frequency diagram matrix, and filtering background noise to obtain a smoothed time-frequency diagram matrix;
(C2) Determining a self-adaptive clipping lower threshold T according to the calculated time-frequency matrix mean value image ；
(C3) Searching from two ends to the center of the image in the time dimension and the frequency dimension respectively to find out that the first is larger than T image The rows and columns are marked as w1, w2, h1 and h2 and are used as self-adaptive clipping areas;
(C4) The adaptive clipping region is adjusted to 128 pixels by 128 pixels using bicubic interpolation.
8. The intelligent radar active interference detection, type and pattern recognition method based on the two-level deep learning fusion network according to claim 1, wherein the step (4) comprises the following steps:
(41) Setting interference types and patterns to be identified, and designing and manufacturing training and testing simulation data set samples;
(42) According to the interference pattern category number and the training set sample number, the depth and the breadth of ResNet and the Attention-LSTM parameter scale are adjusted to be matched with each other;
(43) Inputting the manufactured training set and test set into a secondary deep learning fusion network, setting network super-parameters, adjusting weight parameters in the network by a gradient descent method, and finally obtaining a training model;
(44) The radar receives the interference signal in real time, converts the interference signal to an intermediate frequency, and outputs the type and the style of the interference signal through a preprocessing link and a training model after down-conversion processing.
9. The intelligent radar active interference detection, type and pattern recognition method based on the secondary deep learning fusion network according to claim 8, wherein,
the design and manufacture training and test simulation data set sample in the step (41) specifically comprises the following steps: by setting different bandwidth, duty ratio, dry-to-noise ratio, temperature, slice number, false target number, sweep frequency speed, frequency modulation slope, amplitude modulation slope parameters, a training and testing simulation data set is manufactured, and the sample number proportion is 18000:3600;
the ResNet depth in the step (42) refers to the stacking number of residual modules, and proper number is selected according to task requirements; resNet breadth refers to the number of ResNet network feature graphs, and the capability of extracting image features is represented; the attribute-LSTM parameter scale is specifically: the total unit number and the memory unit number of the Attention-LSTM, each unit calculates the output of the current moment by calculating the input of the current moment and combining the results of all the effective memory unit numbers in the past;
the network super parameters in the step (43) are specifically as follows: the learning rate, the batch number, the training period, the loss function and the regularization parameter of the network training are affected, and the training model is specifically: after training the training set through the secondary deep learning fusion network, determining the trained network of all weight parameters in the network.
10. An intelligent radar active interference detection, type and pattern recognition device for implementing the method, the device comprising:
the down-conversion processing module is used for demodulating the digital intermediate frequency signal to a zero intermediate frequency signal by using a quadrature demodulation technology;
the interference detection module is used for carrying out frequency domain processing on the zero intermediate frequency signal, detecting whether interference exists or not, and obtaining parameters such as interference power, azimuth and bandwidth;
the preprocessing module is used for carrying out FFT conversion and STFT conversion on the zero intermediate frequency signal to obtain a frequency spectrum and a time-frequency matrix of the zero intermediate frequency signal; respectively carrying out normalization and standardization treatment on the frequency spectrum, and carrying out linear graying treatment and self-adaptive clipping treatment on the time frequency matrix;
and the second-level deep learning fusion network module automatically extracts the interference signal characteristics from the output of the preprocessing module through a training model and outputs the type and the style of interference.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210009977.4A CN116359851A (en) | 2022-01-06 | 2022-01-06 | Radar active interference detection and identification method and device based on converged network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210009977.4A CN116359851A (en) | 2022-01-06 | 2022-01-06 | Radar active interference detection and identification method and device based on converged network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116359851A true CN116359851A (en) | 2023-06-30 |
Family
ID=86940445
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210009977.4A Pending CN116359851A (en) | 2022-01-06 | 2022-01-06 | Radar active interference detection and identification method and device based on converged network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116359851A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116938761A (en) * | 2023-09-15 | 2023-10-24 | 深圳市扬名伟创信息技术有限公司 | Internet of things terminal rapid testing system and method |
| CN117856941A (en) * | 2024-01-18 | 2024-04-09 | 阳光凯讯(北京)科技股份有限公司 | 5G wireless communication network interference signal identification method, system and electronic equipment |
| CN117953588A (en) * | 2024-03-26 | 2024-04-30 | 南昌航空大学 | Badminton player action intelligent recognition method integrating scene information |
| CN118118305A (en) * | 2024-04-29 | 2024-05-31 | 长沙先度科技有限公司 | Modulated signal identification method of signal receiver |
| CN118520335A (en) * | 2024-07-18 | 2024-08-20 | 长鹰恒容电磁科技(成都)有限公司 | Radar signal identification system, method, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109492822A (en) * | 2018-11-24 | 2019-03-19 | 上海师范大学 | Air pollutant concentration time-space domain interaction prediction method |
| CN110221256A (en) * | 2019-06-21 | 2019-09-10 | 西安电子科技大学 | SAR disturbance restraining method based on depth residual error network |
| CN111693944A (en) * | 2020-06-18 | 2020-09-22 | 上海志良电子科技有限公司 | Radar active interference signal parameter extraction and interference pattern identification method and device |
| CN111861027A (en) * | 2020-07-29 | 2020-10-30 | 北京工商大学 | Urban traffic flow prediction method based on deep learning fusion model |
-
2022
- 2022-01-06 CN CN202210009977.4A patent/CN116359851A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109492822A (en) * | 2018-11-24 | 2019-03-19 | 上海师范大学 | Air pollutant concentration time-space domain interaction prediction method |
| CN110221256A (en) * | 2019-06-21 | 2019-09-10 | 西安电子科技大学 | SAR disturbance restraining method based on depth residual error network |
| CN111693944A (en) * | 2020-06-18 | 2020-09-22 | 上海志良电子科技有限公司 | Radar active interference signal parameter extraction and interference pattern identification method and device |
| CN111861027A (en) * | 2020-07-29 | 2020-10-30 | 北京工商大学 | Urban traffic flow prediction method based on deep learning fusion model |
Non-Patent Citations (1)
| Title |
|---|
| 姜正云 等: "基于并联深度学习网络的雷达有源干扰智能识别方法", 现代雷达, vol. 43, no. 10, 31 October 2021 (2021-10-31), pages 9 - 14 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116938761A (en) * | 2023-09-15 | 2023-10-24 | 深圳市扬名伟创信息技术有限公司 | Internet of things terminal rapid testing system and method |
| CN116938761B (en) * | 2023-09-15 | 2024-01-12 | 深圳市扬名伟创信息技术有限公司 | Internet of things terminal rapid testing system and method |
| CN117856941A (en) * | 2024-01-18 | 2024-04-09 | 阳光凯讯(北京)科技股份有限公司 | 5G wireless communication network interference signal identification method, system and electronic equipment |
| CN117953588A (en) * | 2024-03-26 | 2024-04-30 | 南昌航空大学 | Badminton player action intelligent recognition method integrating scene information |
| CN118118305A (en) * | 2024-04-29 | 2024-05-31 | 长沙先度科技有限公司 | Modulated signal identification method of signal receiver |
| CN118520335A (en) * | 2024-07-18 | 2024-08-20 | 长鹰恒容电磁科技(成都)有限公司 | Radar signal identification system, method, equipment and storage medium |
| CN118520335B (en) * | 2024-07-18 | 2024-09-17 | 长鹰恒容电磁科技(成都)有限公司 | Radar signal identification system, method, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116359851A (en) | Radar active interference detection and identification method and device based on converged network | |
| CN111352102B (en) | Multi-target number detection method and device based on frequency modulation continuous wave radar | |
| CN108600135B (en) | Method for identifying signal modulation mode | |
| CN102279390B (en) | Intra-pulse modulation and recognition method of low signal-to-noise radar radiation source signal | |
| CN111693944B (en) | Radar active interference signal parameter extraction and interference pattern recognition method and device | |
| CN108919203B (en) | Radar active interference identification method and system | |
| CN108280395B (en) | Efficient identification method for flight control signals of low-small-slow unmanned aerial vehicle | |
| CN112560803A (en) | Radar signal modulation identification method based on time-frequency analysis and machine learning | |
| CN112087774B (en) | Communication radiation source individual identification method based on residual error neural network | |
| CN110133632B (en) | Composite modulation signal identification method based on CWD time-frequency analysis | |
| CN103941254A (en) | Soil physical property classification recognition method and device based on geological radar | |
| CN112859014A (en) | Radar interference suppression method, device and medium based on radar signal sorting | |
| Yang et al. | One-dimensional deep attention convolution network (ODACN) for signals classification | |
| CN111985349B (en) | Classification recognition method and system for radar received signal types | |
| Tian et al. | Modulation pattern recognition of communication signals based on fractional low-order Choi-Williams distribution and convolutional neural network in impulsive noise environment | |
| Chen et al. | GNSS interference type recognition with fingerprint spectrum DNN method | |
| CN114355298B (en) | Radar composite modulation pulse signal identification method | |
| CN112613443A (en) | Robustness communication radiation source intelligent identification method based on deep learning | |
| Tan et al. | Specific emitter identification based on software-defined radio and decision fusion | |
| Gongming et al. | Radar emitter sorting and recognition based on time-frequency image union feature | |
| Guven et al. | Classifying LPI radar waveforms with time-frequency transformations using multi-stage CNN system | |
| Liu et al. | RF fingerprint recognition based on spectrum waterfall diagram | |
| Jiang et al. | Frequency hopping signal parameter estimation algorithm based on time-frequency point correlation | |
| CN103605120A (en) | Range-extended target intrinsic mode local singular value orderly fusion detection method | |
| CN115905919A (en) | Radar signal sample data processing method based on AFGAN, target identification method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |