CN116699531B - Radar signal noise reduction method, system and storage medium based on complex network - Google Patents
Radar signal noise reduction method, system and storage medium based on complex network Download PDFInfo
- Publication number
- CN116699531B CN116699531B CN202310959822.1A CN202310959822A CN116699531B CN 116699531 B CN116699531 B CN 116699531B CN 202310959822 A CN202310959822 A CN 202310959822A CN 116699531 B CN116699531 B CN 116699531B
- Authority
- CN
- China
- Prior art keywords
- complex
- radar signal
- noise reduction
- radar
- layer
- 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.)
- Active
Links
- 230000009467 reduction Effects 0.000 title claims abstract description 95
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 44
- 210000003743 erythrocyte Anatomy 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 19
- 238000010606 normalization Methods 0.000 claims description 18
- 238000004590 computer program Methods 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 18
- 238000012545 processing Methods 0.000 description 15
- 238000013528 artificial neural network Methods 0.000 description 9
- 238000000605 extraction Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 7
- 230000008901 benefit Effects 0.000 description 5
- 230000015556 catabolic process Effects 0.000 description 4
- 238000006731 degradation reaction Methods 0.000 description 4
- 239000013598 vector Substances 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 230000008602 contraction Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013524 data verification Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 230000000007 visual effect Effects 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
- G01S7/35—Details of non-pulse systems
- G01S7/352—Receivers
- G01S7/354—Extracting wanted echo-signals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Computer Networks & Wireless Communication (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The application discloses a radar signal noise reduction method, a system and a storage medium based on a complex network. Acquiring multiple groups of original data with specified length from any modulation type radar signal; constructing a training data set based on original radar signals of all modulation types of radar signals; based on a preset loss function, and using back propagation to update model parameters of a radar signal noise reduction model, training is performed; after training is completed, the radar signal to be processed is input into a trained radar signal noise reduction model so as to output the radar signal after noise reduction. The method solves the problems of high misjudgment degree of distinguishing signals from noise, poor noise reduction effect and the like of the traditional radar signal noise reduction method.
Description
Technical Field
The present application relates to the field of signal processing technologies, and in particular, to a method, a system, and a storage medium for noise reduction of radar signals based on a complex network.
Background
The task of radar signal processing is to suppress noise and interference to the greatest extent and extract information about the target properties. The increasingly complex battlefield electromagnetic environment and the unknown target characteristics bring more challenges to radar signal processing, particularly under the condition of severe environment or enemy interference, the noise content of the radar signal received by a receiver is high, the difficulty of direct processing and utilization is high, the misjudgment probability during recognition is high, the radar signal is successfully recognized on the premise of subsequent processing, so that denoising processing is often needed before subsequent use of the radar signal, and the method has important significance for denoising the received radar signal.
Taking radar radiation source identification which plays an important role in electronic warfare as an example, radar radiation source identification is an important component of electronic information reconnaissance, and the modulation mode of enemy radar signals can be more accurately identified by utilizing the radar signal change among pulses, so that more battlefield information and situations can be obtained, and a basis is provided for subsequent effective decisions.
The electromagnetic signal environment faced by modern electronic information reconnaissance is complex and changeable, the signal-to-noise ratio of the intercepted signal is low in many cases, the modulation patterns are diversified, and the signal identification needs to be developed to an automation and intelligent direction. Because most radar signals are nonstationary signals, the waveforms are complex and various, and noise interference exists, the radar signals cannot be accurately identified in a strong noise environment by the traditional identification method or the identification method based on machine learning, particularly deep learning, and therefore the method has important significance in the actual application of noise reduction treatment of the received radar signal data.
In the strong noise signal environment, the traditional signal noise reduction method such as filtering noise reduction, wavelet transformation noise reduction, spectral subtraction noise reduction and the like has poor effects, but the classical deep learning noise reduction method often uses only the amplitude information of the radar echo signal to carry out noise reduction treatment, so that the phase information is lost, and the deep residual error shrinkage network DRSN which integrates the threshold method noise reduction thought also has the defect of the traditional threshold noise reduction algorithm, namely the degree of distinction between the signal and the noise is poor in the strong noise environment.
The traditional radar signal noise reduction method mainly comprises the following defects:
the traditional one-dimensional signal noise reduction algorithm based on filtering and threshold values has poor generalization capability, poor non-stationary signal and nonlinear noise processing effect, greatly limited flexibility and can not meet the flexible processing requirement of radar signals in complex noise environments;
the essential characteristic, namely the modulation rule, of the radar signal is difficult to extract in a strong noise environment.
The deep learning algorithm which integrates the traditional one-dimensional signal noise reduction thought, such as a deep contraction network DRSN, and the like, has the problem that the radar signal and the noise are mixed to a high degree when the noise is large, so that the radar signal and the noise are difficult to distinguish.
Disclosure of Invention
The embodiment of the application provides a radar signal noise reduction method, a system and a storage medium based on a complex network, which are used for solving the problems of high misjudgment degree, poor noise reduction effect and the like of distinguishing signals and noise by the traditional radar signal noise reduction method under a strong noise environment and various complex modulation modes.
The embodiment of the application provides a radar signal noise reduction method based on a complex network, which is realized based on a radar signal noise reduction model, wherein the radar signal noise reduction model comprises an encoding part, a decoding part and a characteristic connecting part, the encoding part comprises two complex convolution layers CC, four RBCs and a complex full connection layer CFC, one complex convolution layer CC, four RBCs and one complex full connection layer CFC are sequentially arranged, the other complex convolution layer CC is connected between a third RBC and the complex full connection layer CFC, the decoding part comprises four complex convolution layers CC and five RBTs, wherein the three complex convolution layers CC and the five RBTs are sequentially arranged, the output of the complex full connection layer CFC is used as the input of a first complex convolution layer CC, the output of a final RBT is used as the output of the radar signal noise reduction model, the other complex convolution layer CC is arranged between the third complex convolution layer CC and the output of the first RBT, the RBT of the second and the third RBT is also respectively connected with the first RBT and the second RBT of the third complex convolution layer CC, and the characteristic connecting part is arranged between the first RBT and the second RBT of the third RBT:
training a radar signal noise reduction model in advance by adopting the following modes:
acquiring radar signals of various modulation types, and acquiring multiple groups of original data with specified lengths for the radar signals of any modulation type;
constructing a training data set based on original radar signals of all modulation types of radar signals;
based on a preset loss function, and using back propagation to update model parameters of a radar signal noise reduction model, training is performed;
after training is completed, the tested radar signals containing noise with different intensities are input into a trained radar signal noise reduction model so as to output the radar signals after noise reduction.
Optionally, any one of the RBCs is a one-dimensional complex split module OCRB formed based on convolution, any one of the RBC trunk parts is formed by cascading a complex convolution layer, a complex batch normalization layer CBN, a complex activation function ZReLU, a complex convolution layer and a complex batch normalization layer, and the branch part is formed by cascading a complex convolution layer and a complex batch normalization layer.
Optionally, any one of the RBTs is a one-dimensional complex split module OCRB formed by deconvolution, and any one of the RBT trunk parts is formed by cascading a complex transposed convolution layer, a complex batch normalization layer CBN, a complex activation function ZReLU, a complex transposed convolution layer, and a complex batch normalization layer, and the branch part is formed by cascading a complex transposed convolution layer and a complex batch normalization layer.
Optionally, constructing the training data set based on the original radar signal of each modulation type radar signal includes:
adding strong noise to a plurality of raw radar signals of each modulation type to obtain a noisy radar signal to construct a training data set.
Optionally, the method further comprises:
dividing the training data set into verification data sets according to a preset proportion; the method comprises the steps of,
noise of different intensities is added to the plurality of raw radar signals for each modulation type as a test data set.
Optionally, based on a preset loss function, and using back propagation to update model parameters of the radar signal noise reduction model, performing training includes:
training a model network using a radar signal training data set, updating model parameters using back propagation using a mean square loss function as a trained loss function, and selecting the model parameters with the minimum loss by verifying the data set loss.
The embodiment of the application also provides a radar signal noise reduction system based on a complex network, which comprises a processor and a memory, wherein the memory is stored with a computer program, and the computer program realizes the steps of the radar signal noise reduction method based on the complex network when being executed by the processor.
The embodiment of the application also provides a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and the computer program realizes the steps of the radar signal noise reduction method based on the complex network when being executed by a processor.
According to the embodiment of the application, the radar modulation signal one-dimensional sequence is divided into data blocks with the same fixed size according to the modulation rule, and the corresponding relation among the data blocks is utilized, so that the radar signal noise reduction method based on the complex network is provided, the influence of noise on the modulation rule is weakened, the purpose of noise reduction is realized, and the problems of high misjudgment degree of distinguishing signals and noise, poor noise reduction effect and the like of the traditional radar signal noise reduction method under a strong noise environment and various complex modulation modes are solved.
The foregoing description is only an overview of the present application, and is intended to be implemented in accordance with the teachings of the present application in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present application more readily apparent.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the application. Also, like reference numerals are used to designate like parts throughout the figures. In the drawings:
fig. 1 is a schematic diagram of a radar signal noise reduction model according to an embodiment of the present application;
FIG. 2 is a flow chart of training a radar signal noise reduction model according to an embodiment of the present application;
FIGS. 3a-3e are time-frequency diagrams of the modulated signals before and after processing under the condition of 5dB signal-to-noise ratio in the embodiment of the application, and FIGS. 3a-3e correspond to CW, LFM, NLFM, BFSK, BPSK respectively;
FIGS. 4a-4e are time-frequency diagrams of the modulated signals before and after processing under the signal-to-noise ratio condition of 15dB according to the embodiment of the application, and FIGS. 4a-4e correspond to CW, LFM, NLFM, BFSK, BPSK respectively;
fig. 5a-5e are time-frequency diagrams before and after processing each modulated signal under the condition of 20dB signal-to-noise ratio in the embodiment of the present application, and fig. 5a-5e correspond to CW, LFM, NLFM, BFSK, BPSK respectively.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
The embodiment of the application provides a radar signal noise reduction method based on a complex network, which is realized based on a radar signal noise reduction model, as shown in fig. 1, wherein the radar signal noise reduction model comprises an encoding part, a decoding part and a characteristic connecting part, the encoding part comprises two complex convolution layers CC, four RBCs and a complex full-connecting layer CFC, one complex convolution layer CC, four RBCs and one complex full-connecting layer CFC are sequentially arranged, and the other complex convolution layer CC is connected between a third-level RBC and the complex full-connecting layer CFC.
The decoding part comprises four complex convolution layers CC and five RBTs, wherein three complex convolution layers CC and five RBTs are sequentially arranged, the output of the complex full-connection layer CFC is used as the input of a first complex convolution layer CC, the output of a final RBT is used as the output of a radar signal noise reduction model, the other complex convolution layer CC is arranged between the third complex convolution layer CC and the output of the first RBT, and the RBT of the last-last stage and the third stage is also respectively connected with the RBC of the first stage and the second stage.
The feature connection is disposed between the first stage RBT and the second stage RBT. In some embodiments, the cross-sharing module SSB is a feature that connects some modules or layers that use the encoding portion in the decoding portion, which can effectively compensate for edge features lost during upsampling after previous downsampling while effectively utilizing the encoding features.
As shown in fig. 2, the radar signal noise reduction method of the present application includes the steps of:
training a radar signal noise reduction model in advance by adopting the following modes:
in step S101, radar signals of a plurality of modulation types are acquired, and a plurality of sets of original data of a specified length are acquired for the radar signals of any one modulation type. For example, in some specific applications, five modulation types of radar signals (CW, LFM, NLFM, BFSK, BPSK) may be selected, where each modulation type of one-dimensional radar signal obtains n sets of raw data, each set having a length L and a data type being complex.
In step S102, a training data set is constructed based on the original radar signals of the respective modulation type radar signals. In some embodiments, constructing the training data set based on the original radar signal of each modulation type radar signal comprises: adding strong noise to a plurality of raw radar signals of each modulation type to obtain a noisy radar signal to construct a training data set. Adding strong noise to n original radar signals of each modulation type to obtain noise-added signal data, and forming a training data setWherein->Representing the original signal data,/->Representing corresponding strong noise signal data.
In some embodiments, further comprising: dividing the training data set into verification data sets according to a preset proportion; and adding noise of different intensities to the plurality of raw radar signals of each modulation type as a test data set.
The verification set can be cut according to a certain proportion according to the training data set, and noise with different intensities is added to radar signals of each modulation type based on the original radar signals to serve as a radar signal test data set.
In step S103, training is performed based on a preset loss function and using back propagation to update model parameters of the radar signal noise reduction model. In some embodiments, performing training based on a preset loss function and using back propagation to update model parameters of a radar signal noise reduction model includes:
the radar signal training data set is used for training a model network, a mean square loss function is used as a training loss function, model parameters are updated through back propagation, and model parameters with minimum loss are selected through verification of data set loss, so that after training, a radar signal noise reduction model with a good training effect and based on a complex network is obtained.
After training is completed, the radar signal to be processed is input into a trained radar signal noise reduction model so as to output the radar signal after noise reduction.
According to the embodiment of the application, the radar modulation signal one-dimensional sequence is divided into data blocks with the same fixed size according to the modulation rule, and the corresponding relation among the data blocks is utilized, so that the radar signal noise reduction method based on the complex network is provided, the influence of noise on the modulation rule is weakened, the purpose of noise reduction is realized, and the problems of high misjudgment degree of distinguishing signals and noise, poor noise reduction effect and the like of the traditional radar signal noise reduction method under a strong noise environment and various complex modulation modes are solved.
In some embodiments, any one of the RBCs is a one-dimensional complex split module OCRB configured based on convolution, as shown in fig. 1, and any one of the RBC trunk parts is formed by cascading a complex convolution layer, a complex batch normalization layer CBN, a complex activation function ZReLU, a complex convolution layer, and a complex batch normalization layer, and the branch part is formed by cascading a complex convolution layer and a complex batch normalization layer.
In some embodiments, any of the RBTs is a one-dimensional complex split module OCRB formed by deconvolution, and as shown in fig. 1, any of the RBT trunk sections is formed by concatenating a complex transposed convolution layer, a complex batch normalization layer CBN, a complex activation function ZReLU, a complex transposed convolution layer, and a complex batch normalization layer, and the branch sections are formed by concatenating a complex transposed convolution layer and a complex batch normalization layer.
The existing neural network one-dimensional signal noise reduction model mostly only adopts amplitude information and ignores phase information, so that in order to fully utilize one-dimensional signal information, the model adopts a one-dimensional complex network, so that the essential characteristics of signals can be better extracted, and the probability that useful information is misjudged as noise is reduced. The radar signal noise reduction model provided by the embodiment of the application enables the network main body to select the coding and decoding structure to enable the output and the input to keep the same type for realizing the noise reduction function of the radar signal.
In this embodiment, the input enters the neural network to obtain a more complete characteristic through a complex convolution layer, retains more information, and outputs o1.
The radar signal noise reduction model abandons the idea that the neural network noise reduction model introduces traditional signal noise reduction in the prior art, for example, the depth residual error contraction network DRSN adopts a attention mechanism to generate a signal threshold value in a residual error branch to realize noise reduction, and the method has better performance under the low noise condition, but the noise is far greater than a useful signal under the strong noise condition, so that the traditional signal noise reduction method has larger misjudgment probability and poorer noise reduction effect. The radar signal noise reduction model provided by the application adopts a noise-free signal sample as a target sample, and takes a strong noise signal as input to extract the essential characteristics of the signal, namely the modulation rule, more quickly and accurately.
The coding part inputs o1 into a RBC to obtain output o2; o2 is input into a RBC to obtain o3, o3 is input into the RBC to obtain o4, o4 is input into the RBC to obtain o5a, o4 is input into a CC to obtain o5b, the weighted average of o5a and o5b is obtained to obtain o5, and o5 is input into the CFC to obtain o6.
The traditional coding and decoding structure has the problems of gradient oscillation, difficult convergence and the like, so that the characteristic extraction effect and the nonlinear expression capability are poor when the network is shallow, and the effect cannot meet the ideal requirement; when the network is deeper, the random gradient descent is used for training the network, the globally optimal solution is often not found, the solution space of the deep network is more complex, the calculation complexity is rapidly increased, and the model effect can even appear degradation phenomenon. The radar signal noise reduction model of the application provides a one-dimensional complex split module OCRB to adapt to requirements, and uses two paths of feature extraction of different network structures with different network layers, so that the model combines features of different layers during each feature extraction, simultaneously avoids the whole neural network from being artificially set too deep and too shallow, and the network adaptively selects an effective network path in one interval to select proper features. The one-dimensional complex dividing and combining module can not only solve the degradation problem of the neural network, but also extract and retain the modulation rule of the radar signal more effectively.
In the decoding part and the crossing sharing module, o6 is firstly input into RBT through three CCs to obtain o7, o7 is input into RBT to obtain o8a, o7 is input into CCs to obtain o8b, o8a and o8b are weighted and averaged to obtain o8, o8 is input into RBT to obtain o9, o8 and o9 and o3 are weighted and averaged to obtain o10, o10 and o2 are weighted and averaged to obtain o11, and o11 is input into the last RBT to obtain output.
Compared with the traditional code-decoding type neural network, the partial similarity of some radar modulation signals in the feature extraction part is higher, and if the subsequent decoding part continues to extract the partial features, the radar signal modulation rule distinction degree is insufficient, and the noise reduction effect is affected. The applicant has innovatively added a full-connection layer in the encoding and decoding process for combining features, converting information between the encoding part and the decoding part, so that the network can better learn the relation between the input sequence and the output sequence. In particular, the addition of the fully-connected layer may convert high-dimensional feature vectors in the encoder to low-dimensional vectors while converting low-dimensional vectors in the decoder to high-dimensional vectors, thereby allowing the network to better process the sequence data. The influence of the feature positions on feature discrimination is reduced, and the robustness is enhanced. In addition, the full connection layer can also increase the depth and complexity of the network, and improve the fitting capacity and generalization capacity of the network.
In the neural network of the traditional coding and decoding structure, the coding and decoding parts are only formed by simple cascading, the characteristics of the decoding part are only the re-extraction and expansion of the characteristics of the coding part, and after the data is subjected to downsampling and upsampling, some edge information is lost and cannot be retrieved, and the characteristics are single in use. The radar signal noise reduction model adds the feature connection parts in the front and back of the network, especially in the key layers of some feature extraction, so that the network can make up for the edge features lost in the downsampling process of the coding part when decoding and restoring the signal structure, and can grasp the features of different layers and integrate the features in a feature superposition mode. The characteristics of different layers or receptive fields with different sizes are different in sensitivity to signal modulation rules, radar signal fluctuation of some modulation modes is small, radar signal characteristics of some modulation modes are obvious, edge information of some obvious characteristics and modulation information with small changes in the primary downsampling and upsampling of the neural network are easy to lose, the network has receptive fields with different sizes after the characteristic connecting part is added, the characteristics are reused, and the noise reduction effect is better.
The application builds a one-dimensional complex combining network OCRN based on a complex network radar signal noise reduction model. The neural network takes a coding and decoding structure built by a one-dimensional complex split module as a main body, a full-connection layer is added in the middle part, and the superposition of features is performed before and after the network across a sharing module SSB. The radar signal noise reduction model has the following advantages: A. resistance to training degradation: the one-dimensional complex dividing and combining module can effectively avoid the problems of gradient disappearance and gradient explosion, thereby improving the training stability and the robustness of the network. B. High resolution feature extraction: the coding part uses one-dimensional complex division and combination module to extract the characteristics, so that the important information of the input data can be better reserved, and the characteristic representation with higher resolution is generated. C. Decoder alignment: the characteristic connection before and after the coding and decoding part can lead the network to be more aligned when generating the output sequence, thereby improving the quality of the generated sequence. D. End-to-end learning: the whole network can be regarded as an end-to-end learning process, and intermediate steps such as manually designing a feature extractor and an aligner are not needed, so that the design and implementation of the model are simplified. E. Scalability: the coding part and the decoding part can increase the expression capacity of the network by increasing the layer number, thereby adapting to the data sets and task demands of different scales, being not limited to the denoising of radar signals and being capable of expanding to other communication signals.
In some applications, after training is completed, a radar signal sample which is in a test data set and contains noise with different intensities is input into a trained complex network radar signal noise reduction model for testing, a noise-reduced signal is output, and initial observation effect is lost from output, if the loss is smaller, a Cohen time-frequency distribution which is smooth pseudo Wigner-Ville distribution can be utilized to perform time-frequency transformation on five radar modulation signals (CW, LFM, NLFM, BPSK, BFSK) before and after testing under radar signal conditions with different signal to noise ratios (the signal to noise ratios are respectively-5 dB, -15dB and-20 dB), so that a time-frequency distribution diagram which is compared before and after under different noise conditions is obtained.
As shown in fig. 3a-3e, 4a-4e, 5a-5e, the time-frequency diagrams before and after processing each modulated signal under-5 dB signal-to-noise ratio condition, -the time-frequency diagrams before and after processing each modulated signal under-15 dB signal-to-noise ratio condition, and the time-frequency diagrams before and after processing each modulated signal under-20 dB signal-to-noise ratio condition are shown, respectively.
As can be seen from the visual effects of fig. 3a-3e, 4a-4e and 5a-5e, the radar signal noise reduction model has better extraction effect on the modulation rule of the radar signal under the condition of strong noise, and the noise reduction effect of the radar signal noise reduction model is far stronger than that of the traditional method. The radar signal noise reduction model has good radar signal noise reduction effects under different signal to noise ratio conditions, and has good generalization capability and robustness, thus having important significance for subsequent processing of radar signals under different noise conditions.
The radar signal noise reduction model takes a coding and decoding structure built by a one-dimensional complex split module as a main body, a full-connection layer is added in the middle part, and a crossing sharing module is added at the same time, so that the feature sharing utilization before and after a network is realized. Therefore, the model has good training degradation resistance, high-resolution feature extraction capability, higher reliability and expansibility and noise reduction effect far exceeding the traditional noise reduction method.
The embodiment of the application also provides a radar signal noise reduction system based on a complex network, which comprises a processor and a memory, wherein the memory is stored with a computer program, and the computer program realizes the steps of the radar signal noise reduction method based on the complex network when being executed by the processor.
The embodiment of the application also provides a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and the computer program realizes the steps of the radar signal noise reduction method based on the complex network when being executed by a processor.
It should be noted that, in the embodiments of the present disclosure, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The foregoing embodiment numbers of the present application are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present application.
The embodiments of the present application have been described above with reference to the accompanying drawings, but the present application is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those having ordinary skill in the art without departing from the spirit of the present application and the scope of the claims, which are to be protected by the present application.
Claims (6)
1. The radar signal noise reduction method based on the complex network is characterized by being realized based on a radar signal noise reduction model, the radar signal noise reduction model comprises an encoding part, a decoding part and a characteristic connecting part, the encoding part comprises two complex convolution layers CC, four RBCs and a complex full connection layer CFC, one complex convolution layer CC, four RBCs and one complex full connection layer CFC are sequentially arranged, the other complex convolution layer CC is connected between a third RBC and the complex full connection layer CFC, the decoding part comprises four complex convolution layers CC and five RBTs, three complex convolution layers CC and five RBTs are sequentially arranged, the output of the complex full connection layer CFC serves as the input of a first complex convolution layer CC, the output of a final RBT serves as the output of the radar signal noise reduction model, the other complex convolution layer CC is arranged between the third complex convolution layer CC and the output of the first RBT, the RBT of the second stage and the third stage are also respectively connected to the first stage and the second stage, and the characteristic connecting part is arranged between the first stage and the second RBT and comprises:
training a radar signal noise reduction model in advance by adopting the following modes:
acquiring radar signals of various modulation types, and acquiring multiple groups of original data with specified lengths for the radar signals of any modulation type;
constructing a training data set based on original radar signals of all modulation types of radar signals;
based on a preset loss function, and using back propagation to update model parameters of a radar signal noise reduction model, training is performed;
after training is completed, inputting the radar signal to be processed into a trained radar signal noise reduction model to output a noise-reduced radar signal;
any RBC is a one-dimensional complex split module OCRB formed based on convolution, any RBC trunk part is formed by cascading a complex convolution layer, a complex batch normalization layer CBN, a complex activation function ZReLU, a complex convolution layer and a complex batch normalization layer, and branch parts are formed by cascading a complex convolution layer and a complex batch normalization layer;
any RBT is a one-dimensional complex split module OCRB formed by deconvolution, any RBT trunk part is formed by cascading a complex transpose convolution layer, a complex batch normalization layer CBN, a complex activation function ZReLU, a complex transpose convolution layer and a complex batch normalization layer, and branch parts are formed by cascading a complex transpose convolution layer and a complex batch normalization layer.
2. The complex network-based radar signal denoising method of claim 1, wherein constructing a training data set based on original radar signals of each modulation type radar signal comprises:
adding strong noise to a plurality of raw radar signals of each modulation type to obtain a noisy radar signal to construct a training data set.
3. The complex network-based radar signal denoising method of claim 2, further comprising:
dividing the training data set into verification data sets according to a preset proportion; the method comprises the steps of,
noise of different intensities is added to the plurality of raw radar signals for each modulation type as a test data set.
4. A complex network-based radar signal denoising method as claimed in claim 3, wherein performing training based on a predetermined loss function and using back propagation to update model parameters of a radar signal denoising model comprises:
training a model network using a radar signal training data set, updating model parameters using back propagation using a mean square loss function as a trained loss function, and selecting the model parameters with the minimum loss by verifying the data set loss.
5. A complex network based radar signal noise reduction system comprising a processor and a memory, the memory having stored thereon a computer program which, when executed by the processor, implements the steps of the complex network based radar signal noise reduction method of any of claims 1 to 4.
6. A computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the steps of the complex network-based radar signal noise reduction method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310959822.1A CN116699531B (en) | 2023-08-02 | 2023-08-02 | Radar signal noise reduction method, system and storage medium based on complex network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310959822.1A CN116699531B (en) | 2023-08-02 | 2023-08-02 | Radar signal noise reduction method, system and storage medium based on complex network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116699531A CN116699531A (en) | 2023-09-05 |
| CN116699531B true CN116699531B (en) | 2023-11-17 |
Family
ID=87824220
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310959822.1A Active CN116699531B (en) | 2023-08-02 | 2023-08-02 | Radar signal noise reduction method, system and storage medium based on complex network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116699531B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102015A (en) * | 2018-08-06 | 2018-12-28 | 西安电子科技大学 | A kind of SAR image change detection based on complex-valued neural networks |
| CN113191185A (en) * | 2021-03-10 | 2021-07-30 | 中国民航大学 | Method for classifying targets of unmanned aerial vehicle by radar detection through Dense2Net |
| CN113376600A (en) * | 2021-05-10 | 2021-09-10 | 西安电子科技大学 | Pedestrian radar echo denoising method based on RSDNet |
| CN115345216A (en) * | 2022-06-30 | 2022-11-15 | 北京邮电大学 | FMCW radar interference elimination method fusing prior information |
| CN115754954A (en) * | 2022-10-21 | 2023-03-07 | 江苏科技大学 | Feature fusion method applied to radar and AIS track association |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220076111A1 (en) * | 2020-09-08 | 2022-03-10 | United States Of America As Represented By The Secretary Of The Navy | Neural Network Approach for Identifying a Radar Signal in the Presence of Noise |
-
2023
- 2023-08-02 CN CN202310959822.1A patent/CN116699531B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102015A (en) * | 2018-08-06 | 2018-12-28 | 西安电子科技大学 | A kind of SAR image change detection based on complex-valued neural networks |
| CN113191185A (en) * | 2021-03-10 | 2021-07-30 | 中国民航大学 | Method for classifying targets of unmanned aerial vehicle by radar detection through Dense2Net |
| CN113376600A (en) * | 2021-05-10 | 2021-09-10 | 西安电子科技大学 | Pedestrian radar echo denoising method based on RSDNet |
| CN115345216A (en) * | 2022-06-30 | 2022-11-15 | 北京邮电大学 | FMCW radar interference elimination method fusing prior information |
| CN115754954A (en) * | 2022-10-21 | 2023-03-07 | 江苏科技大学 | Feature fusion method applied to radar and AIS track association |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116699531A (en) | 2023-09-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110824450B (en) | Radar target HRRP robust identification method in noise environment | |
| CN111242862B (en) | Multi-scale fusion parallel dense residual convolution neural network image denoising method | |
| CN108509911B (en) | Interference signal identification method based on convolutional neural network | |
| CN112087774B (en) | Communication radiation source individual identification method based on residual error neural network | |
| CN110941794A (en) | Anti-attack defense method based on universal inverse disturbance defense matrix | |
| CN102279390A (en) | Intra-pulse modulation and recognition method of low signal-to-noise radar radiation source signal | |
| Xiao et al. | Radar signal recognition based on transfer learning and feature fusion | |
| Ni et al. | LPI radar waveform recognition based on multi-resolution deep feature fusion | |
| CN113673312B (en) | Deep learning-based radar signal intra-pulse modulation identification method | |
| CN115359771B (en) | Underwater sound signal noise reduction method, system, equipment and storage medium | |
| CN115982613A (en) | Signal modulation identification system and method based on improved convolutional neural network | |
| An et al. | Robust automatic modulation classification in low signal to noise ratio | |
| CN116699531B (en) | Radar signal noise reduction method, system and storage medium based on complex network | |
| CN114399440A (en) | Image processing method, image processing network training method and device and electronic equipment | |
| Xia et al. | Radar waveform recognition based on a two‐stream convolutional network and software defined radio | |
| Zha et al. | Intelligent identification technology for high‐order digital modulation signals under low signal‐to‐noise ratio conditions | |
| CN112054979B (en) | Radio automatic modulation identification method based on fuzzy dense convolution network | |
| CN115685133A (en) | Positioning method for autonomous vehicle, control device, storage medium, and vehicle | |
| CN115514597A (en) | Joint modulation identification method based on attention mechanism and residual error structure | |
| CN115563485A (en) | Low-complexity interference identification method based on deep learning | |
| CN114358050A (en) | Radar radiation source intelligent identification method based on bicubic interpolation and WVD characteristic square matrix | |
| Zhong et al. | Multimodal deep learning model for specific emitter identification | |
| CN115062658B (en) | Overlapping radar signal modulation type identification method based on self-adaptive threshold network | |
| CN115409058B (en) | Anti-disturbance generation method and system for automatic modulation recognition depth network | |
| Jiang et al. | Individual recognition for satellite communication based on feature fusion method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |