CN116540189A - Radar main lobe forwarding interference resisting method based on reversible residual error network - Google Patents
Radar main lobe forwarding interference resisting method based on reversible residual error network Download PDFInfo
- Publication number
- CN116540189A CN116540189A CN202310667580.9A CN202310667580A CN116540189A CN 116540189 A CN116540189 A CN 116540189A CN 202310667580 A CN202310667580 A CN 202310667580A CN 116540189 A CN116540189 A CN 116540189A
- Authority
- CN
- China
- Prior art keywords
- signal
- interference
- layer
- network
- target
- 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
- 238000000034 method Methods 0.000 title claims abstract description 27
- 230000002441 reversible effect Effects 0.000 title claims abstract description 24
- 239000011159 matrix material Substances 0.000 claims abstract description 80
- 230000001629 suppression Effects 0.000 claims abstract description 22
- 238000013527 convolutional neural network Methods 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 238000001228 spectrum Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 7
- 238000000354 decomposition reaction Methods 0.000 claims description 6
- 230000002452 interceptive effect Effects 0.000 claims description 6
- 238000004088 simulation Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 5
- 239000006185 dispersion Substances 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 2
- 230000005540 biological transmission Effects 0.000 abstract description 5
- 238000013528 artificial neural network Methods 0.000 abstract description 2
- 238000013135 deep learning Methods 0.000 abstract description 2
- 230000001066 destructive effect Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 7
- 238000001514 detection method Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 230000006872 improvement Effects 0.000 description 3
- 238000011084 recovery Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000000926 separation method Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 230000005764 inhibitory process Effects 0.000 description 2
- 230000035485 pulse pressure Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 244000126211 Hericium coralloides Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
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 belongs to the technical field of radar anti-interference, and particularly relates to an anti-radar main lobe forwarding interference method based on a reversible residual error network. The invention provides a set of radar main lobe forwarding interference suppression deep learning framework Inv-NMF net suitable for a multi-target multi-interference environment based on a non-Negative Matrix Factorization (NMF) algorithm and a reversible residual neural network. The method can decompose the interference signal by utilizing radar emission signal information, and characterize the target signal information and the interference signal information at different positions of a coefficient matrix. And then analyzing the coefficient matrix by using a reversible residual error network to generate a Mask matrix, and separating the interference signal component and the target signal component. The invention can ensure that the target signal strength is not influenced to the greatest extent because of the information transmission non-destructive property of the reversible network. The method is not only not affected by the number of radar targets, but also has better adaptability to different radar emission waveforms.
Description
Technical Field
The invention belongs to the technical field of radar anti-interference, and particularly relates to an anti-radar main lobe forwarding interference method based on a reversible residual error network.
Background
The radar system is one of the core detection means in modern communication, and can monitor a target area in real time all the weather by transmitting and receiving electromagnetic wave signals. At present, with the continuous development of digital radio frequency memory (Digital Radio Frequency Memory, DRFM) technology, radar main lobe forwarding interference is highly related to radar transmitting signals by virtue of the characteristics that the radar main lobe forwarding interference enters a radar receiver from an antenna main lobe, so that most traditional radar anti-interference strategies fail, and the target detection performance of the radar is drastically reduced. Therefore, the research of the inhibition method aiming at the radar main lobe forwarding interference is an important research direction for ensuring the radar detection capability, and has important research significance. Most of the traditional signal processing methods are limited to the forward interference suppression under a single target, and few algorithms capable of suppressing the forward interference under multiple targets have a very difficult interference suppression effect. Therefore, no suitable radar main lobe forwarding interference resisting algorithm can well solve the radar interference problem under multiple targets.
Disclosure of Invention
Aiming at the problems, the method starts from the time domain signal, reduces the complexity of signal separation by utilizing the information of the radar emission signal, refers to a non-negative matrix factorization method commonly used in the blind source separation theory, takes the radar emission signal as a dictionary matrix, and obtains the feature matrix of the radar echo. And replacing a time-frequency diagram matrix obtained by short-time Fourier transform with the feature matrix to perform subsequent interference signal separation. And for the problem of information deficiency in multi-target signal recovery, an attempt is made to extract signal characteristics by using a reversible residual error network, so that the signal is recovered approximately without damage.
The invention provides a set of radar main lobe forwarding interference suppression deep learning framework Inv-NMF net suitable for a multi-target multi-interference environment based on a non-Negative Matrix Factorization (NMF) algorithm and a reversible residual neural network. The method can decompose the interference signal by utilizing radar emission signal information, and characterize the target signal information and the interference signal information at different positions of a coefficient matrix. And then analyzing the coefficient matrix by using a reversible residual error network to generate a Mask matrix, and separating the interference signal component and the target signal component. The invention can ensure that the target signal strength is not influenced to the greatest extent because of the information transmission non-destructive property of the reversible network. The method is not only not affected by the number of radar targets, but also has better adaptability to different radar emission waveforms.
The technical scheme of the invention is as follows:
a radar main lobe forwarding interference resisting method based on a reversible residual error network comprises the following steps:
s1, taking a linear frequency modulation Signal as a radar transmitting Signal, obtaining an interference Signal through radar main lobe forwarding interference simulation, and further obtaining an echo Signal added with interference data Transmitting Signal, interfering Signal tag Jam target And Signal tag Signal without interference target ：
Signal data ＝{x 0 i |i＝1,2,…,L}∈C L
Signal＝{x 1 i |i＝1,2,…,L}∈C L
Jam target ＝{x 2 i |i＝1,2,…,L}∈C L
Signal target ＝{x 3 i |i＝1,2,…,L}∈C L
Wherein L is the signal length;
s2, to Signal data Signal and Signal target Respectively performing linear normalization processing, mapping Signal intensity to (0, 1), and using Signal data Maximum minimum value of (1) versus Jam target Normalizing to ensure that the normalization ratio of the two is consistent, and then utilizing a sliding window mode to carry out Signal data And in the Signal, each M long time slice Signal is segmented, and then spliced in the next dimension to form a corresponding M multiplied by N two-dimensional matrix X epsilon R M×N And S.epsilon.R M×N ，L＝M×N；
S3, setting the S as a dictionary matrix W, and carrying out NMF decomposition on the matrix X to obtain a coefficient matrix H corresponding to the signal sample matrix X, wherein the specific method comprises the following steps:
randomly initializing coefficient matrix H E R N×N Ensure H ij ≥0|i＝1,...,N；j＝1,...,N；
Updating the H matrix parameters and calculating an error e:
if the error e meets the set condition, outputting a coefficient matrix H, otherwise, continuing to update iteratively until the maximum iteration times are reached;
s4, constructing an interference suppression network based on a reversible residual network, wherein the interference suppression network comprises three reversible residual network layers with the same structure, the forward propagation input of the network is a normalized coefficient matrix H obtained by NMF decomposition, and the expected output is a separated interference signal Mask matrix Mask; each layer of network consists of a group of symmetrical Unet networks, and specifically comprises the following steps: defining two-terminal input of each layer network as X 1 And X 2 Taking coefficient matrix H as X of first-layer network 1 And X 2 Input, X 1 Extracting interference signal information in an original signal through a first Unet network in a first layer network, wherein the first Unet network comprises five CNN layers, input matrixes of a first layer and a fifth layer, input matrixes of a second layer and input matrixes of a fourth layer are the same in size, layer-jump connection is used for interaction, output of the upper layer and output of the upper layer are spliced in a channel dimension and are input to the next layer together, and particularly input of the first layer is X 1 The input of the second layer is the output of the first layer, the input of the third layer is the input of the second layer, the input of the fourth layer is the output of the second layer and the third layer, and the input of the fifth layer is the input of the first layer and the fourth layer; output of fifth layer and X 2 After combination, target echo information Y of preliminary interference filtering is obtained 2 ，Y 2 Extracting purer target echo information through a second Unet network in the first layer network,the second Unet network comprises five CNN layers, and the output of the fifth layer of the second Unet network is identical to that of the first Unet network 1 Combining to obtain an output Y 1 The method comprises the steps of carrying out a first treatment on the surface of the Output Y of layer one network 1 And Y 2 Respectively as input X of the next-layer network 2 And X 1 The output of the last layer of network is subjected to a Sigmoid function to obtain a Mask matrix Mask of the interference signal;
multiplying the output interference matrix Mask and the coefficient matrix H to obtain a coefficient matrix of an interference signal, and multiplying the coefficient matrix of the interference signal with a dictionary matrix W to obtain an interference signal J;
from Signal data Subtracting the interference signal J from the target signal, primarily recovering the target signal without interference, and obtaining a final target signal y without interference by the target signal without interference through a convolutional neural network, wherein the convolutional neural network is of a group of U-net model structures and is similar to the first Unet network structure in the S4. The downsampling times are increased to 5 times, and each downsampling layer is composed of a residual convolution layer, an activation function layer and a batch normalization layer. Downsampling is achieved by increasing the step length of the convolution kernel, the sampling multiple is 2 times, and the number of channels of each layer is increased layer by layer. The up-sampling process is the same, but the channel number is gradually decreased layer by layer, and finally the pure interference-free signal is recovered.
S5, training the constructed interference suppression network, wherein the adopted loss function comprises the following steps:
l1 error loss of interfering signal:
wherein J represents the interference signal recovered by the mask matrix, and N is the length of the interference signal;
l1 error loss of target signal:
wherein y represents an interference-free target signal finally output by the model, and N is the length of the target signal;
hilbert transform loss:
wherein Hilbert () is a Hilbert transform function;
when the network loss basically converges, a final depth network model is obtained;
s6, processing radar received signals by using the obtained depth network model to realize radar main lobe forwarding interference resistance.
Further, the interference signals obtained through radar main lobe forwarding interference simulation in the S1 comprise intermittent sampling forwarding interference, smart noise interference, spectrum dispersion interference and comb spectrum interference. The intermittent sampling forwarding type interference works in such a way that an intercepted radar signal is sampled, one section of the signal is forwarded, and the next signal section is sampled and forwarded until the falling edge of the radar transmission signal is detected. According to different intermittent sampling modes, intermittent sampling forwarding interference can be subdivided into direct forwarding interference J ISDJ Repeating interference J ISRJ Cyclic forwarding interference J ISIJ Three kinds. The three interference signals are respectively expressed as:
wherein N is the number of slices, T s For the slice width, M is the number of times each interfering slice is forwarded, T u ＝(M+1)T s For two adjacent stemsThe acquisition time interval of the scrambling slice. a, a m T s For the capture time of the mth slice, b n T s The corresponding delay in the nth forwarding is performed for that slice. The method comprises the following steps:
the smart noise interference works by storing the intercepted radar transmit signal in a digital register using a Digital Radio Frequency Memory (DRFM). The noise unit is then controlled to generate a noise signal of matched length and type, depending on the form of the radar signal. And finally, the product or convolution operation of the noise signal and the radar signal is finished on a signal synthesizer, so that a smart noise interference signal with a good effect can be generated. The interference can be divided into noise convolution interference J according to different signal modulation modes SCN Sum-noise product interference J SPN Two, their mathematical expressions are:
J SCN (t)＝s(t-t)×n(t)
in the formula, the radar signal received by the DRFM is s (t), the delayed output is s (t-t), and the narrow-band Gaussian noise is n (t).
The spectrum dispersion interference (SMSP) works as follows: after the jammer intercepts the radar transmitting signal, the intercepted signal is firstly processed digitally and stored in the DRFM jammer. And then, using a shift register to obtain N sub-signals with modulation slope N times of the transmitting signal, serially inputting the sub-signals into a digital-to-analog converter, and obtaining the SMSP interference signal through frequency mixing.
The first sub-signal may be expressed as:
wherein k is the sub-signal skewRate, k j For sub-signal frequency modulation slope, k j =nk, will J SMSP1 After N-1 replicates, the SMSP interference is combined:
the comb spectrum interference works in the following way: the comb spectrum signal is multiplied by the intercepted radar emission signal and then modulated to generate an interference signal.
The expression of the comb spectrum signal is:
wherein f i Corresponding to the frequency point of each comb tooth, a i Is the amplitude at the i-th frequency point.
The mathematical model of radar comb spectrum interference is:
the beneficial effects of the invention are as follows: aiming at complex detection scenes of multiple targets and multiple interference sources, the invention designs a multi-target interference suppression network model based on an NMF and a reversible residual error network. Firstly, the model uses radar emission signals as priori information to generate a dictionary matrix, and the NMF is used for decomposing and calculating a coefficient matrix of the received signals, so that complex one-dimensional signals are mapped into a two-dimensional sparse matrix, and the complexity of network learning is greatly reduced. And then the model separates the decomposed coefficient matrix through a reversible residual error network, so that the interference signal can be effectively restrained, and the approximate lossless transmission of the target information can be ensured. The invention can realize good inhibition effect under the condition of multiple targets and multiple interference sources, greatly improves the target detection capability of the subsequent radar, and ensures the accurate detection of the radar on multiple targets.
Drawings
FIG. 1 is a schematic diagram of the overall process flow of the present invention.
Fig. 2 is a schematic diagram of an interference suppression network structure based on a reversible residual network in the present invention.
Fig. 3 is a schematic diagram of a reversible residual network structure in the present invention.
Fig. 4 is a schematic diagram showing the effect of the present invention on three ISRJ interference suppression.
Fig. 5 is a schematic diagram showing the effect of the present invention on SCN, SJN, combo interference suppression.
Fig. 6 is a schematic diagram showing the effect of the model of the present invention on SMSP interference suppression.
Detailed Description
The technical scheme of the invention is further described in detail below with reference to the accompanying drawings and examples:
according to the method, firstly, according to the working mechanism of the main lobe forwarding interference signals, radar interference signals are simulated, and an interference target signal, an interference signal and an undisturbed target signal are generated and used as signal samples and label samples. Then, the signals are subjected to normalization preprocessing, the signals are mapped to between 0 and 1, and then a two-dimensional matrix of the signals is obtained through sliding window processing. Next, the Inv-NMF net model was designed. The internal structure of the device mainly comprises three parts, as shown in figure 1: the first part is NMF decomposition. Using radar-transmitted signals s (t) ∈R 1×L As dictionary matrix W, we can calculate radar received signal x (t) ∈R by multiplication update method 1×L Coefficient matrix H of (a). The second part is the forward propagating part of the reversible network. The signal features in the coefficient matrix H are extracted by using a reversible residual network, and a Mask matrix Mask is generated according to the signal features. The Mask matrix is multiplied by the coefficient matrix H to obtain a coefficient matrix of the interference signal, and the coefficient matrix is multiplied by the dictionary matrix to generate an interference signal y in the radar receiving signal. Finally, subtracting the interference signal part from the received signal, we can obtain the information of the target signal. The third part is the recovery part of the target signal. A simple CNN module is used to accurately recover the target signal. Finally training the built network model, storing the model after the model converges, and forwarding the interference signal by utilizing the randomly generated radar main lobeAnd testing, and counting the interference suppression effect of the model.
Examples
The specific steps of the method are as follows:
s1: establishing an interference signal model: firstly, according to the parameters of table 1, the radar interference signal is simulated
Table 1 simulation parameters
Finally, the output of the simulation radar system is the echo Signal with the added interference data Transmitting Signal, interference Signal Jam target And interference-free target Signal target 。
Signal data ＝{x 0 i |i＝1,2,…,20000}∈C 20000
Signal＝{x 1 i |i＝1,2,…,20000}∈C 20000
Jam target ＝{x 2 i |i＝1,2,…,20000}∈C 20000
Signal target ＝{x 3 i |i＝1,2,…,20000}∈C 20000
S2: signal to Signal sample data Signal of transmission and Signal of sample tag target Respectively performing linear normalization processing, mapping Signal intensity to (0, 1), and using Signal data Maximum minimum value of (1) versus Jam target Normalizing to ensure that the normalization proportion of the two is consistent. Then, by using a sliding window method, to Signal data And in the Signal, each 200 long time slice Signal is segmented and then spliced in the next dimension to form a corresponding 200X 100 two-dimensional matrix X epsilon R 200 ×100 And S.epsilon.R 200×100 。
S3: setting S as a dictionary matrix W, and carrying out NMF decomposition on the matrix X to obtain a coefficient matrix H corresponding to the signal sample matrix X. The specific process is as follows:
first, a coefficient matrix H.epsilon.R is randomly initialized 100×100 Ensure H ij ≥0|i＝1,...,100；j＝1,...,100
Then, the H matrix parameters are updated and the error e is calculated:
finally, if the error e <1e-5, outputting the coefficient matrix H, otherwise, continuing to update iteratively until the maximum iteration number of 500 times is reached.
S4: an Inv-U net interference suppression network is constructed, as in fig. 2, whose forward propagation input is a normalized coefficient matrix H decomposed by NMF, and whose desired output is a separate interference signal Mask matrix Mask. First, I input coefficient matrix H to both ends X of Inv-U net respectively 1 And X 2 . The network comprises three layers in total, and the F and G functions of each layer are composed of a group of symmetrical Unet networks. Wherein the former Unet module F (X 1 ) For extracting the interference signal information from the original signal by X 2 Binding F (X) 1 ) Can obtain the target echo information Y of preliminary filtering interference 2 . The latter Unet module G (Y 2 ) For at Y 2 Extracting purer target echo information based on the above, and comparing it with X 1 Combining to obtain the output Y 1 And obtaining a Mask matrix Mask of the interference signal through a Sigmoid function.
S41: each of the Unet modules is composed of five CNN layers, as shown in fig. 3, wherein the input matrixes of the first layer and the fifth layer, the input matrixes of the second layer and the input matrixes of the fourth layer are the same in size, and are interacted by using layer-jump connection, and the output of the upper layer are spliced in the channel dimension and are input to the lower layer together. Specific network parameters are shown in table 2:
table 2 uiet module parameters
And finally, multiplying the output interference matrix Mask and the coefficient matrix H to obtain the coefficient matrix of the interference signal. The interference signal J can be obtained by multiplying it by the dictionary matrix W.
S5: the interference signal J is subtracted from the signal sample, and the target signal without interference can be initially recovered. And then the target signal is subjected to a set of CNN model structures, so that finer recovery can be realized, and a final interference-free target signal y is obtained. The model structure is similar to the F-function structure described in S4. The downsampling times are 4 times, each downsampling layer is composed of a residual convolution layer, an activation function layer and a batch normalization layer, and the number of channels is increased layer by layer. The up-sampling process is the same, but the number of channels decreases layer by layer. Specific convolutional layer parameters are shown in table 3:
table 3 CNN module parameters
S6: model output signal, and interference signal Jac target And interference-free target Signal target A loss value is calculated. First Loss function Loss 1 Loss of error for L1 (absolute error of both) of the interfering signal:
wherein J represents an interference signal recovered by using a mask matrix, and Jac target Representing the actual jamming signal labels, N being the jamming signal length.
Second Loss function Loss 2 L1 error loss for the target signal:
wherein y represents the interference-free target Signal finally output by the model, signal target And (3) representing a real target signal label, wherein N is the target signal length. The L1 loss is used to measure how good the signal is recovered because it has a stable gradient and does not lead to gradient explosion problems. Thus, not only some outliers can be well ignored, but also a larger penalty can be made for small differences.
Third Loss function Loss 3 Loss for hilbert transform:
where Hilbert () is a Hilbert transform function. On the training set, the batch_size is selected to be 16, the learning rate is adjusted within the range of 0.0005, and the optimizer is selected to be Adam. And (5) after training is completed, storing the model.
Finally, under various different main lobe forwarding interference modes, randomly generated interference signals with different sampling duty ratios are adopted, main lobe interference suppression is carried out through the dual-stage deep network, and target interference-signal ratio improvement before and after interference suppression is counted, as shown in fig. 4-6.
The interference signal to noise ratio improvement factor JSR-IF measures the interference suppression effect in a multi-target environment:
wherein a is star Is the minimum target amplitude in the plurality of targets after echo pulse compression; and a is jam Is the maximum amplitude of the interference target after pulse compression.
The interference-to-signal ratio improvement factor (JSR-IF) can be expressed as:
JSR-IF＝JSR unfiltered -JSR filtered
wherein JSR is filtered JSR is the result of pulse pressure after interference suppression; and JSR (JSR) unfiltered Is JSR of the original signal pulse pressure result without interference suppression processing.
Claims (2)
1. The radar main lobe forwarding interference resisting method based on the reversible residual error network is characterized by comprising the following steps of:
s1, taking a linear frequency modulation Signal as a radar transmitting Signal, obtaining an interference Signal through radar main lobe forwarding interference simulation, and further obtaining an echo Signal added with interference data Transmitting Signal, interfering Signal tag Jam target And Signal tag Signal without interference target ：
Signal data ＝{x 0 i |i＝1,2,…,L}∈C L
Signal＝{x 1 i |i＝1,2,…,L}∈C L
Jam target ＝{x 2 i |i＝1,2,…,L}∈C L
Signal target ＝{x 3 i |i＝1,2,…,L}∈C L
Wherein L is the signal length;
s2, to Signal data Signal and Signal target Respectively performing linear normalization processing, mapping Signal intensity to (0, 1), and using Signal data Maximum minimum value of (1) versus Jam target Normalizing to ensure that the normalization ratio of the two is consistent, and then utilizing a sliding window mode to carry out Signal data And in the Signal, each M long time slice Signal is segmented, and then spliced in the next dimension to form a corresponding M multiplied by N two-dimensional matrix X epsilon R M×N And S.epsilon.R M×N ，L＝M×N；
S3, setting the S as a dictionary matrix W, and carrying out NMF decomposition on the matrix X to obtain a coefficient matrix H corresponding to the signal sample matrix X, wherein the specific method comprises the following steps:
randomly initializing coefficient matrix H E R N×N Ensure H ij ≥0|i＝1,...,N；j＝1,...,N；
Updating the H matrix parameters and calculating an error e:
if the error e meets the set condition, outputting a coefficient matrix H, otherwise, continuing to update iteratively until the maximum iteration times are reached;
s4, constructing an interference suppression network based on a reversible residual error network, wherein the interference suppression network comprises three reversible residual error network layers with the same structure, the forward propagation input of the network is a normalized coefficient matrix H obtained by NMF decomposition, and the expected output is a separated interference signal Mask matrix Mask; each layer of network consists of a group of symmetrical Unet networks, and specifically comprises the following steps: defining two-terminal input of each layer network as X 1 And X 2 Taking coefficient matrix H as X of first-layer network 1 And X 2 Input, X 1 Extracting interference signal information in an original signal through a first Unet network in a first layer network, wherein the first Unet network comprises five CNN layers, input matrixes of a first layer and a fifth layer, input matrixes of a second layer and input matrixes of a fourth layer are the same in size, layer-jump connection is used for interaction, output of the upper layer and output of the upper layer are spliced in a channel dimension and are input to the next layer together, and particularly input of the first layer is X 1 The input of the second layer is the output of the first layer, the input of the third layer is the input of the second layer, the input of the fourth layer is the output of the second layer and the third layer, and the input of the fifth layer is the input of the first layer and the fourth layer; output of fifth layer and X 2 After combination, target echo information Y of preliminary interference filtering is obtained 2 ，Y 2 Extracting purer target echo information through a second Unet network in the first layer network, wherein the second Unet network comprises five CNN layers, and the output of a fifth layer of the second Unet network is identical to that of the first Unet network, and the output of a fifth layer of the second Unet network is identical to that of the X layer of the first Unet network 1 Combining to obtain an output Y 1 The method comprises the steps of carrying out a first treatment on the surface of the Output Y of layer one network 1 And Y 2 Respectively as input X of the next-layer network 2 And X 1 The output of the last layer of network is subjected to a Sigmoid function to obtain a Mask matrix Mask of the interference signal;
multiplying the output interference matrix Mask and the coefficient matrix H to obtain a coefficient matrix of an interference signal, and multiplying the coefficient matrix of the interference signal with a dictionary matrix W to obtain an interference signal J;
from Signal data Signal data Subtracting the interference signal J, primarily recovering an interference-free target signal, and then obtaining a final interference-free target signal y by the primarily recovered interference-free target signal through a convolutional neural network;
s5, training the constructed interference suppression network, wherein the adopted loss function comprises the following steps:
l1 error loss of interfering signal:
wherein J represents the interference signal recovered by the mask matrix, and N is the length of the interference signal;
l1 error loss of target signal:
wherein y represents an interference-free target signal finally output by the model, and N is the length of the target signal;
hilbert transform loss:
wherein Hilbert () is a Hilbert transform function;
when the network loss basically converges, a final depth network model is obtained;
s6, processing radar received signals by using the obtained depth network model to realize radar main lobe forwarding interference resistance.
2. The method for resisting radar main lobe forwarding interference based on the reversible residual error network according to claim 1, wherein the interference signals obtained by radar main lobe forwarding interference simulation in S1 include intermittent sampling forwarding interference, smart noise interference, spectrum dispersion interference and comb spectrum interference.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310667580.9A CN116540189A (en) | 2023-06-07 | 2023-06-07 | Radar main lobe forwarding interference resisting method based on reversible residual error network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310667580.9A CN116540189A (en) | 2023-06-07 | 2023-06-07 | Radar main lobe forwarding interference resisting method based on reversible residual error network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116540189A true CN116540189A (en) | 2023-08-04 |
Family
ID=87443647
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310667580.9A Pending CN116540189A (en) | 2023-06-07 | 2023-06-07 | Radar main lobe forwarding interference resisting method based on reversible residual error network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116540189A (en) |
-
2023
- 2023-06-07 CN CN202310667580.9A patent/CN116540189A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107220606B (en) | Radar radiation source signal identification method based on one-dimensional convolutional neural network | |
| CN109379311B (en) | Ultra-short wave specific signal identification method based on convolutional neural network | |
| CN112098957A (en) | Complex radar radiation source identification method based on one-dimensional self-walking convolution neural network | |
| CN109471074B (en) | Radar radiation source identification method based on singular value decomposition and one-dimensional CNN network | |
| CN112183280B (en) | Underwater sound target radiation noise classification method and system based on EMD and compressed sensing | |
| CN113221631B (en) | Sequence pulse anti-interference target detection method based on convolutional neural network | |
| CN114636975A (en) | LPI radar signal identification method based on spectrogram fusion and attention mechanism | |
| CN109711314B (en) | Radar radiation source signal classification method based on feature fusion and SAE | |
| CN108231087A (en) | A kind of single channel blind source separating method | |
| CN111507047B (en) | Inverse scattering imaging method based on SP-CUnet | |
| CN114881093B (en) | Signal classification and identification method | |
| CN110610717A (en) | Separation method of mixed signals in complex frequency spectrum environment | |
| CN111562597A (en) | Beidou satellite navigation interference source identification method based on BP neural network | |
| CN114239657A (en) | Time sequence signal identification method based on complex value interference neural network | |
| Tang et al. | Jamming recognition based on AC-VAEGAN | |
| Mao et al. | An radio frequency interference mitigation approach for spaceborne SAR system in low SINR condition | |
| CN115616503A (en) | Radar interference signal type identification method based on convolutional neural network model | |
| CN114296041A (en) | Radar radiation source identification method based on DCNN and Transformer | |
| CN114936570A (en) | Interference signal intelligent identification method based on lightweight CNN network | |
| CN116540189A (en) | Radar main lobe forwarding interference resisting method based on reversible residual error network | |
| CN115267713B (en) | Intermittent sampling interference identification and suppression method based on semantic segmentation | |
| CN115951315B (en) | Radar spoofing interference identification method and system based on improved wavelet packet energy spectrum | |
| CN104463245A (en) | Target recognition method | |
| He et al. | Fault diagnosis of analog circuits based on wavelet packets | |
| CN113420668B (en) | Underwater target identification method based on two-dimensional multi-scale permutation entropy |
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 |