CN115345216A - FMCW radar interference elimination method fusing prior information - Google Patents
FMCW radar interference elimination method fusing prior information Download PDFInfo
- Publication number
- CN115345216A CN115345216A CN202210766130.0A CN202210766130A CN115345216A CN 115345216 A CN115345216 A CN 115345216A CN 202210766130 A CN202210766130 A CN 202210766130A CN 115345216 A CN115345216 A CN 115345216A
- Authority
- CN
- China
- Prior art keywords
- complex
- time
- fmcw radar
- interference
- 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
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
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
Abstract
The invention provides an FMCW radar interference elimination method fusing prior information. In the method, an FMCW radar reception signal is removed of mixed interference components and noise using a deep learning based method. Since FMCW radar return signals and interference signals have different distributions in the time-frequency domain, short-time Fourier transform is used to transform the signals from the time domain to the time-frequency domain for processing. In addition, the characteristic that the radar signals are in a complex form is utilized, the real parts and the imaginary parts of the complex radar signals are restrained by using a complex convolution neural network, the number of parameters of an interference elimination network is reduced, and the processing speed is accelerated. Specifically, a loss function combining prior information and average mean square error is used for training an interference elimination network, extra prior information is introduced, the characteristics extracted by the network are constrained, the problem of model overfitting is avoided as much as possible, the accuracy of recovery signals is improved, model convergence is accelerated, and the data volume required by network training is reduced.
Description
Technical Field
The invention belongs to the technical field of Frequency Modulated Continuous Wave (FMCW) radar signal interference elimination by using a deep convolutional neural network, relates to the technologies of neural networks, radar signal processing and the like, and particularly relates to an FMCW radar interference elimination method fusing prior information.
Background
With the coming of the automatic driving era and the gradual popularization of an auxiliary driving system, the FMCW radar is widely applied to the existing automobile system, can directly measure the distance and the speed of a target, monitors the surrounding environment of the automobile in real time, is not influenced by lamplight and weather conditions, and ensures the driving safety of the automatic driving automobile; in addition, the FMCW radar can also be used for vital sign detection of patients in medicine, intelligent building monitoring and weather monitoring in industry and the like; with the rapid development of the applications, the probability that the FMCW radar is interfered by other surrounding radars or radio devices is rapidly increased, so that an interfering signal and white gaussian noise are mixed in an echo signal received by the interfered FMCW radar, the echo signal from a target is shielded, the detection recognition rate of the target is reduced, even a false target is generated, the safe driving of a vehicle is influenced, and a new challenge is brought to the development of an anti-interference technology of the FMCW radar.
The FMCW radar interference elimination technology is a technology for automatically processing, analyzing and understanding echo signals received by an FMCW radar, identifying the echo signals from a target and removing mixed interference signals in the echo signals; to solve this problem, various conventional signal processing methods are used to suppress interference components in the signal, including a nulling and reconstruction method, a quantitative calculation method, and a spatial domain interference cancellation method; meanwhile, with the rapid development of deep learning and pattern recognition technologies and the great improvement of the parallel computing capability of a Graphics Processing Unit (GPU), the method based on deep learning is widely applied to various signal processing problems, and provides more inspirations for FMCW radar interference cancellation technologies.
Through the search of the prior art documents, the inventor of the present application discloses an "asynchronous interference cancellation method based on signal processing" in radar science and technology (2011,9 (06): 556-560+ 567), wherein the method comprises the steps of performing pulse delay cancellation on a received signal, modulo cancellation results, detecting interference components in the signal by using a Constant False Alarm Rate algorithm (CFAR), establishing a distance clutter map, and replacing the interference signal by using adjacent signal interpolation by using a self-defined threshold value, but the detection precision of the method when using CFAR to perform interference detection is influenced by super-parameter selection, and a specific singular value needs to be calculated when the interference is cancelled, so that the performance is unstable; guo Chunhui et al, published in meteorology hydrology and oceanographic instruments (2014,31 (02): 24-29+ 32), "identification and elimination of Doppler weather radar radial interference echo" respectively identify the interference component in radar echo signal by the number of points of echo intensity difference and effective reflectivity value, and eliminate the interference component by interpolation or filtering method based on the identification, but this method has high complexity of calculation and long processing time, and cannot meet the requirement of radar signal real-time processing, and due to the nonlinear characteristic of the filter, the echo signal from the target is lost to a certain extent, which affects the subsequent target detection process.
The prior art literature retrieval results show that the traditional signal processing method for FMCW radar interference elimination needs to perform hyper-parameter selection and higher computational complexity, and most traditional methods make many assumptions and simplifications for simplifying computation and obtaining an analytic solution, and cannot be applied to more complex scenes; only a few documents solve the problem of interference of FMCW radar signals by using a deep learning theory, the methods process the radar signals in different transform domains, and the performance is not compared; because radar data in a real scene are difficult to collect, a large-scale radar signal data set is difficult to construct, and the performance of the FMCW radar interference elimination technology based on deep learning is limited; and the parameter quantity of most FMCW radar interference elimination networks exceeds the memory capacity of a radar sensor in a real environment, so that the FMCW radar interference elimination networks are difficult to be practically applied.
The key point for solving the problems is to design a loss function combining prior information and Mean Square Error (MSE), introduce additional prior information in the FMCW radar interference elimination network training process based on deep learning, and constrain the characteristics extracted by the network through a regularization term, so that the accuracy of an interference elimination model recovery signal is improved, the network training convergence is accelerated, and the required training data volume is reduced; in addition, because the radar signal is complex data, the better feature expression capability of the complex network to the complex tensor can be utilized by designing the complex convolution neural network, the parameter number of the network is reduced, and the memory capacity and the real-time processing requirement of the radar sensor are met.
Disclosure of Invention
The invention overcomes the defects of the existing FMCW radar interference elimination technology, provides an FMCW radar interference elimination method fusing prior experience information, aims to perform FMCW radar interference elimination model training by using a loss function combining prior information and average mean square error, and solves the problems that a large-scale Lei Daxin data set is difficult to obtain, the training time is long, and the model is easy to over-fit in a real scene; meanwhile, a complex convolution neural network is designed instead of the traditional real convolution neural network, and the problems of overlarge parameter quantity and long processing time of an FMCW radar interference elimination network are solved by utilizing the complex characteristics of radar signals and the better characteristic expression capability of the complex network on complex data.
In order to achieve the purpose, the invention adopts the following technical scheme:
an FMCW radar interference elimination method fusing prior information comprises the following steps:
step 200: generating an input signal y (t) containing an FMCW radar target echo signal, an interference signal and Gaussian white noise by using Matlab numerical simulation software; y (t) = s (t) + f (t) + n (t); wherein s (t) represents FMCW radar target echo signals, f (t) represents interference signals, and n (t) represents white Gaussian noise;
the simulated FMCW radar target echo signal and interference signal calculation formula is as follows:
wherein f is c Representing the center frequency of the disturbed FMCW radar, K representing the chirp rate of the FMCW radar emitting a frequency-modulated continuous wave, τ k Is the echo signal from the kth targetTime delay relative to the transmitted signal; p is a radical of formula * (t) denotes a reference signal for demodulation, f m (t) denotes the mth interference, H lp Represents a low pass filtering operation;
step 210: performing Short-Time Fourier Transform (STFT) on an input Signal Y (t) and a clean FMCW radar target echo Signal S (t) under different Signal-to-Interference plus Noise Ratio (SINR) to obtain corresponding Time-frequency matrixes Y (t, f) and S (t, f) under different SINR;
further, the calculation formula of the time-frequency matrix Y (t, f) is:
wherein f represents frequency, tau is an integral time variable, and h (t) is a Hamming window;
step 220: preprocessing the generated time-frequency matrix, dividing the matrix into a plurality of matrixes with the length and the width of 256 points along a time axis, and simultaneously overlapping eight sampling points among the plurality of matrixes to ensure the continuity of phases; taking a part of time-frequency matrixes under each SINR to construct a training data set, and forming a verification set and a test set by the rest time-frequency matrixes; wherein Y (t, f) is used as the input of the data set, S (t, f) is used as the corresponding label, and the two have the same size;
step 230: constructing an FMCW radar interference elimination network structure based on a complex convolutional neural network, wherein the interference elimination network based on the complex convolutional neural network is composed of N complex convolutional layers, and each complex convolutional layer is connected with a CReLU complex activation function except the last complex convolutional layer;
further, the specific calculation method of the plurality of convolutional layers is as follows:
W*h＝(A*x-B*y)+j(A*y+B*x) (4)
W＝A+jB (5)
h＝x+jy (6)
wherein W represents a complex convolution kernel, h represents an input complex tensor, A represents a real part of the complex convolution kernel, B represents an imaginary part of the complex convolution kernel, x represents a real part of the input complex tensor, and y represents an imaginary part of the input complex tensor;
the specific calculation method of the CReLU complex activation function is as follows:
wherein z represents the input complex tensor,
the real part of z is represented by,
denotes the imaginary part of z, reLU denotes the conventional linear rectifying activation function;
step 240: inputting an FMCW radar interference elimination training data set into the FMCW radar interference elimination network based on the complex convolution neural network constructed in the step 230 for training, wherein a loss function in the training process adopts a loss function combining prior information and an average mean square error, the loss function is continuously reduced by continuously carrying out cyclic iterative training on the network until the set iteration number Q is finished, and an optimal interference elimination model is stored;
further, the formula for calculating the loss function by combining the prior information and the average mean square error is as follows:
L＝L 1 +b*L 2 (8)
wherein BS denotes an FMCW radar interference cancellation networkThe number of training samples input in each iteration in the training process, L represents the loss average value of BS samples, and L 1 Representing the Mean Square Error (MSE) loss function, L 2 B represents a weight parameter which can be manually adjusted and is determined by actual conditions; m and N represent the length and width of a time-frequency matrix input by the interference elimination network, z represents the label value of a channel l when the coordinate of the ith sample in the time-frequency matrix is (j, k), and y represents the predicted value of the channel l when the coordinate of the ith sample in the time-frequency matrix is (j, k);
step 250: performing feature extraction on the FMCW radar signal by using the interference elimination test data set trained in the step 240 through the interference elimination model, distinguishing an echo signal and an interference signal of a target, and removing interference components and noise in the signal to obtain a time-frequency matrix of the signal predicted by the FMCW radar interference elimination test data set;
step 260: merging the predicted Time-frequency matrixes, removing the middle overlapped part, and performing Inverse Short-Time Fourier Transform (ISTFT) on the merged Time-frequency matrixes to obtain corresponding Time-domain signals; calculating the SINR of the recovered signal by using the recovered time domain signal and the corresponding label to obtain a test result of the FMCW radar interference elimination test data set;
further, the SINR is calculated as:
wherein N represents the number of sampling points of FMCW radar signal in time domain, s i Clean signal, y, representing the time domain i A recovery signal representing a time domain after being processed by the interference cancellation network;
advantageous effects
Compared with the prior FMCW radar interference elimination technology, the method has the beneficial effects that:
the method adopts a deep learning-based method to eliminate FMCW radar signal interference, adopts a complex convolution neural network to replace a traditional real convolution neural network to extract signal characteristics, uses the extracted characteristics to describe main information of a signal, and adopts a loss function combining prior information and average mean square error to train the interference elimination network based on the complex convolution neural network; the mean-square error loss function is only to calculate the difference between the recovery signal and the corresponding label, and can not completely eliminate the mixed interference component in the signal when a large amount of training data is lacked, while the loss function combining the prior information and the mean-square error introduces extra expert knowledge, and carries out regularization constraint on the extracted characteristic, thereby not only increasing the convergence speed of the training, but also distinguishing the interference and the target echo signal as much as possible when the training data set is small, and further improving the accuracy of the recovery signal of the interference elimination model; in addition, the complex convolutional neural network has better expression capability and better noise robustness by constraining the real part and the imaginary part of the radar signal in a complex form, can reduce the number of parameters of the interference elimination network, and meets the requirements of the memory capacity and the real-time processing of the radar sensor; the method is suitable for FMCW radar interference elimination, can reduce the dependence of the average mean square error loss function on the size of a data set required by training an interference elimination model based on a convolutional neural network, accelerates the model training convergence, improves the accuracy of signal recovery, reduces the number of model parameters and reduces the time required by signal processing.
Drawings
FIG. 1 is a flow chart of an embodiment of the method of the present invention including data set preparation, network training, and network testing.
Fig. 2 is a signal processing flow diagram of the present invention.
Fig. 3 is a schematic diagram of the structure of the FMCW radar interference cancellation network based on the complex convolutional neural network in the present invention.
FIG. 4 is a graph of the comparison of test data performance after processing by a complex neural network and a corresponding real network in the present invention.
FIG. 5 is a graph comparing the performance of the test data after the model trained by adding the prior information is processed.
Fig. 6 is a time domain waveform diagram, a frequency spectrum diagram and a time-frequency diagram after test data is processed by the method for eliminating FMCW radar interference with fusion prior information.
Detailed Description
The invention provides an FMCW radar interference elimination method fusing prior information, aiming at the problem that the target detection recognition rate is reduced due to the fact that an FMCW radar is interfered by other radars and wireless equipment.
The method implementation flowchart of this case is shown in fig. 1, and its specific implementation steps are:
step 300: generating an input signal y (t) containing an FMCW radar target echo signal, an interference signal and Gaussian white noise by using Matlab numerical simulation software; y (t) = s (t) + f (t) + n (t); wherein s (t) represents FMCW radar target echo signals, f (t) represents interference signals, and n (t) represents white Gaussian noise;
step 310: performing Short Time Fourier Transform (STFT) on an input Signal Y (t) and a clean FMCW radar target echo Signal S (t) under different Signal to Interference plus Noise ratios (SINRs) to obtain corresponding Time-frequency matrixes Y (t, f) and S (t, f) under different SINRs;
step 320: preprocessing the generated time-frequency matrix, dividing the matrix into complex matrixes with the length and the width of 256 points along a time axis, and simultaneously overlapping eight sampling points between each complex matrix to ensure the continuity of phases; taking a part of time-frequency matrixes under each SINR to construct a training data set, and forming a verification set and a test set by the rest time-frequency matrixes; wherein Y (t, f) is used as the input of the data set, S (t, f) is used as the corresponding label, and the two have the same size;
step 330: constructing an FMCW radar interference elimination network structure based on a complex convolutional neural network, wherein the interference elimination network based on the complex convolutional neural network is composed of N complex convolutional layers, and each complex convolutional layer is connected with a CReLU complex activation function except the last complex convolutional layer;
step 340: inputting an FMCW radar interference elimination training data set into the FMCW radar interference elimination network based on the complex convolution neural network constructed in the step 330 for training, wherein a loss function in the training process adopts a loss function combining prior information and an average mean square error, the loss function is continuously reduced by continuously carrying out cyclic iterative training on the network until the set iteration number Q is finished, and an optimal interference elimination model is stored;
step 350: performing feature extraction on the FMCW radar signal by using the interference elimination model trained in the step 340 on the FMCW radar interference elimination test data set, distinguishing an echo signal and an interference signal of a target, and removing interference components and noise in the signal to obtain a time-frequency matrix of the signal predicted by the FMCW radar interference elimination test data set;
step 360: merging the predicted Time-frequency matrixes, removing the middle overlapped part, and performing Inverse Short-Time Fourier Transform (ISTFT) on the merged Time-frequency matrixes to obtain corresponding Time-domain signals; calculating the SINR of the recovered signal by using the recovered time domain signal and the corresponding label to obtain a test result of the FMCW radar interference elimination test data set;
the signal processing flow in this case is shown in fig. 2. After a signal received by an FMCW radar is subjected to low-pass filtering operation, converting a time domain to a time-frequency domain by using short-time Fourier transform, and slicing an obtained time-frequency matrix along a time axis to obtain a complex matrix with the length and the width of 256 sampling points; inputting the plural matrixes into a trained interference elimination network to obtain a time-frequency matrix of a recovered signal, and splicing the matrixes along a time axis to obtain a complete time-frequency matrix; and converting the time-frequency matrix of the recovery signal into a time domain by using an inverse short-time Fourier transform algorithm to obtain a time-domain waveform of the recovery signal.
The structure of the interference cancellation network of this case is shown in fig. 3. The FMCW radar interference elimination network based on the complex convolutional neural network is completely composed of a plurality of convolutional layers, except for the last layer, each layer is connected with a complex activation function CReLU; the input of the network is a time-frequency matrix of an interfered signal with the length and width of 256 points, the output is a time-frequency matrix of a recovery signal with the same size, and the label is a time-frequency matrix of a clean signal (a radar echo signal without interference and noise) with the same size; the input of the network is a complex matrix, the real part and the imaginary part of each element of the matrix are respectively distributed on two channels, and two-dimensional complex convolution layers are used for processing the dual-channel data; the interference elimination network is composed of 9 layers of complex convolution layers, the number of convolution kernels of each layer is shown in FIG. 3, the size of the convolution kernels is set to be 3*3, the step size of the convolution kernels is set to be 1, and the output and the input of each layer keep the same size; the total parameter number of the interference elimination network is 55666, and the memory requirement of an actual radar sensor can be met.
The test results of the FMCW radar interference cancellation test data set are shown in fig. 4, 5 and 6. Fig. 4 shows the variation relationship between the signal to interference plus noise ratio of the recovered signal and the signal to interference plus noise ratio of the interfered signal after being processed by different interference cancellation networks. In fig. 4, the signal to interference plus noise ratios at equal distances are plotted on the abscissa, and it can be seen that the signal to interference plus noise ratio of the restored signal decreases with the decrease of the signal to interference plus noise ratio of the interfered signal, because the larger the interference component ratio in the signal decreases with the decrease of the signal to interference plus noise ratio of the interfered signal, the more the difficulty of signal restoration increases significantly; the two solid lines in fig. 4 compare the optimal complex convolutional neural network and the real convolutional neural network obtained through parameter search, and it can be seen that as the signal-to-interference-and-noise ratio of the interfered signal is reduced, the signal-to-interference-and-noise ratio of the complex convolutional neural network is gradually higher than that of the real convolutional neural network, because the complex convolutional layer takes into account the relationship between the real part and the imaginary part of the complex number, the complex convolutional layer has better feature expression capability and better noise robustness. In addition, the number of parameters of the optimal complex number network in fig. 4 is only 40% of the optimal real number network, because the real part and the imaginary part of the convolution kernel of the complex number convolution layer correspond to one feature map (feature map), respectively, and the number of the actual convolution kernels is twice as large as that of the real number convolution layer. The four dotted lines in fig. 4 compare the interference cancellation performance of the complex network and the real network when the number of convolutional layers of the convolutional neural network is 5 and 6, respectively, so that it can be seen that the complex network has better performance when the interference strength of the interfered signal is greater, and meanwhile, the complex network can reduce the number of parameters of the network and accelerate the signal processing speed.
Fig. 5 shows the variation relationship between the signal-to-interference-and-noise ratio of the recovered signal and the signal-to-interference-and-noise ratio of the interfered signal after the interference cancellation network trained by using a small amount of sample data processes. The FMCW radar interference cancellation training set only includes 540 samples, equidistant signal-to-interference-and-noise ratios are used as abscissa in fig. 5, and the 4 solid lines in the diagram are used as the signal-to-interference-and-noise ratios of the restored signals when the super-parameters b in the loss function are respectively 0, 128, 256 and 400, and it can be seen that the signal-to-interference-and-noise ratios of the restored signals are significantly improved as the value of the super-parameter b is gradually increased, because when the training data volume is small, the interference cancellation network trained by using the mean square error MSE can only extract features from a small amount of data, and the trained network performance is poor; when MSE is combined with prior information, the prior information item can introduce extra expert knowledge as a regularization item to restrict the characteristics extracted by the network, so that the interference components in the signal can be more fully removed.
Fig. 6 shows a time domain waveform diagram, a distance amplitude spectrum and a time-frequency diagram corresponding to an interfered signal, a corresponding clean signal, and a recovery signal processed by an interference elimination network of the FMCW radar. FIG. (a) is a time domain waveform of a disturbed radar signal, and a pulse caused by disturbance can be seen, wherein the amplitude of the pulse is far larger than that of a target echo signal; the graph (b) is a distance amplitude spectrum of the interfered radar and the corresponding clean signal, two targets detected by the radar can be seen, the positions of two peak values can be clearly seen in the distance amplitude spectrum of the clean signal, and due to the influence of interference, the second target with weaker strength is shielded and cannot be detected; graphs (c) and (d) are time-frequency graphs of the interfered signal and the clean signal generated by using a short-time Fourier transform algorithm, the interference is represented by a plurality of wider oblique lines, and the echo signal of the target is represented by a thinner transverse line; the time-frequency diagram of the recovery signal processed by the interference elimination network is given in the diagram (e), the loss function combining the prior information and the average mean square error is used during the network training, the value of the super parameter is set to 400, and it can be seen that after the processing of the interference elimination network, the noise and the interference components are almost completely removed; the graph (f) is a spectrogram corresponding to the recovered signal after being processed by the interference cancellation network trained by different hyper-parameter values, and as can be seen from the graph, the second weaker signal can be detected again after being processed by the interference cancellation network, and meanwhile, as the value of the hyper-parameter b is gradually increased, the noise intensity is gradually reduced, and the target detection accuracy is gradually increased.
Claims (4)
1. An FMCW radar interference elimination method fused with prior information is characterized by comprising the following steps: simulating an interfered signal and a corresponding clean signal of an FMCW radar by using Matlab software, performing short-time Fourier transform to generate a time-frequency matrix of the two signals, and preparing an FMCW radar interference elimination training data set and a test data set; constructing an FMCW radar interference elimination network structure based on a complex convolutional neural network, wherein the interference elimination network based on the complex convolutional neural network is composed of N two-dimensional complex convolutional layers connected in series, and each complex convolutional layer except the last complex convolutional layer is connected with a CReLU complex activation function; inputting an FMCW radar interference elimination training data set into a constructed interference elimination network based on a complex convolution neural network for training, wherein a loss function in the training process adopts a loss function combining prior information and Mean Square Error (MSE), the loss function is continuously reduced by continuously carrying out cyclic iterative training on the network until the set iteration Q is finished, and an interference elimination model is stored; performing characteristic extraction on FMCW radar signals by using the obtained interference elimination model on an FMCW radar interference elimination test data set, distinguishing target echo signals and interference, removing interference components in the signals, and obtaining a time-frequency matrix of recovery signals; and transforming the recovered signal from a time-frequency domain to a time domain by using an inverse short-time Fourier transform algorithm, and calculating the signal-to-interference-and-noise ratio of the recovered signal to obtain a test result of the FMCW radar interference elimination test data set.
2. The method according to claim 1, wherein in the process of generating the data set, after performing short-time fourier transform, the generated time-frequency matrix needs to be preprocessed, the matrix is divided into complex matrices with length and width of 256 points along a time axis, and simultaneously, eight sampling points are overlapped between each complex matrix to ensure phase continuity.
3. The method of claim 1, wherein the complex convolutional neural network is composed of complex convolutional layers and complex activation functions CReLU, and the specific calculation between the complex convolutional layers and the input complex tensor is as follows:
W*h＝(A*x-B*y)+j(A*y+B*x)
W＝A+jB
h＝x+jy
wherein, W represents a complex convolution kernel, h represents an input complex tensor, A represents a real part of the complex convolution kernel, B represents an imaginary part of the complex convolution kernel, x represents a real part of the input complex tensor, and y represents an imaginary part of the input complex tensor;
the specific calculation method of the CReLU complex activation function is as follows:
wherein z represents the input complex tensor,
the real part of z is represented by,
denotes the imaginary part of z and ReLU denotes the conventional linear rectifying activation function.
4. The method of claim 1, wherein the combined prior information and Mean Square Error (MSE) loss function for training the interference cancellation network is calculated as:
wherein BS represents the number of training samples input in each iteration in the training process of the FMCW radar interference elimination network, L represents the loss average value of BS samples, and L represents the loss average value of BS samples 1 Representing the Mean Square Error (MSE) loss function, L 2 B represents a weight parameter which can be manually adjusted and is determined by actual conditions; m and N represent the length and width of a time-frequency matrix input by the interference elimination network, z represents the label value of the channel l when the coordinate of the ith sample in the time-frequency matrix is (j, k), and y represents the predicted value of the channel l when the coordinate of the ith sample in the time-frequency matrix is (j, k).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210766130.0A CN115345216A (en) | 2022-06-30 | 2022-06-30 | FMCW radar interference elimination method fusing prior information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210766130.0A CN115345216A (en) | 2022-06-30 | 2022-06-30 | FMCW radar interference elimination method fusing prior information |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115345216A true CN115345216A (en) | 2022-11-15 |
Family
ID=83948577
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210766130.0A Pending CN115345216A (en) | 2022-06-30 | 2022-06-30 | FMCW radar interference elimination method fusing prior information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115345216A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116699531A (en) * | 2023-08-02 | 2023-09-05 | 中国人民解放军战略支援部队航天工程大学 | Radar signal noise reduction method, system and storage medium based on complex network |
-
2022
- 2022-06-30 CN CN202210766130.0A patent/CN115345216A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116699531A (en) * | 2023-08-02 | 2023-09-05 | 中国人民解放军战略支援部队航天工程大学 | Radar signal noise reduction method, system and storage medium based on complex network |
| CN116699531B (en) * | 2023-08-02 | 2023-11-17 | 中国人民解放军战略支援部队航天工程大学 | Radar signal noise reduction method, system and storage medium based on complex network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107220606B (en) | Radar radiation source signal identification method based on one-dimensional convolutional neural network | |
| CN111352102B (en) | Multi-target number detection method and device based on frequency modulation continuous wave radar | |
| CN106468770B (en) | Nearly optimal radar target detection method under K Distribution Clutter plus noise | |
| Liu et al. | Deep learning and recognition of radar jamming based on CNN | |
| Huynh-The et al. | Accurate LPI radar waveform recognition with CWD-TFA for deep convolutional network | |
| CN111722188B (en) | PRI conversion radar signal sorting method based on STFT pre-sorting | |
| CN103941254A (en) | Soil physical property classification recognition method and device based on geological radar | |
| CN110133632B (en) | Composite modulation signal identification method based on CWD time-frequency analysis | |
| Cain et al. | Convolutional neural networks for radar emitter classification | |
| KR100780470B1 (en) | Adaptive target signal discrimination of surveillance radar | |
| CN108594177A (en) | Based on radar signal modulation system analysis method, the signal processing system for improving HHT | |
| CN112859014A (en) | Radar interference suppression method, device and medium based on radar signal sorting | |
| CN105158749A (en) | High-frequency radar sea-clutter amplitude statistical distribution test method | |
| CN111427018A (en) | Radar interference equipment interference effect evaluation method | |
| Ni et al. | LPI radar waveform recognition based on multi-resolution deep feature fusion | |
| CN111474524A (en) | Radar interference equipment interference effect monitoring and decision support system | |
| CN114895263A (en) | Radar active interference signal identification method based on deep migration learning | |
| CN114114166A (en) | Radar pulse de-interlacing method based on DTM algorithm | |
| CN115345216A (en) | FMCW radar interference elimination method fusing prior information | |
| Ristea et al. | Estimating the magnitude and phase of automotive radar signals under multiple interference sources with fully convolutional networks | |
| CN107329115A (en) | LFM modulated parameter estimating methods based on GCRBF networks | |
| CN114117912A (en) | Sea clutter modeling and inhibiting method under data model dual drive | |
| CN112327286B (en) | Method, device, equipment and storage medium for classifying daily activities under low complexity | |
| Limin et al. | Low probability of intercept radar signal recognition based on the improved AlexNet model | |
| CN116797796A (en) | Signal identification method based on time-frequency analysis and deep learning under DRFM intermittent sampling |
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 |