Spectrum-sensing based ultra-wideband radar signal jamming detection method

Abstract

The application discloses a kind of based on spectrum sensing's ultra-wideband radar signal interference detection method, first, the echo signal of ultra-wideband radar and interference signal are modeled to generate a large number of interference data sets for network learning, then the echo matrix after interference is transformed to frequency domain, along the direction of orientation is divided into pulse vector, and the spectrum sensing model of pulse is constructed, then the interference perception network model including offline learning network and online test network is established, the offline training of offline learning network is carried out, the detector with the length L is developed, and the online test network uses detector to perform sliding detection to the echo signal after processing, finally, the interference part in the echo detected is set to 0, complete ultra-wideband radar signal interference detection.The method of the application proposes a new spectrum sensing network, which can better adapt to the complex electromagnetic environment of ultra-wideband radar system, has the advantages of high interference detection probability, fast network learning speed and easy deployment on hardware platform.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of ultra-wideband radar signal processing, and particularly relates to an ultra-wideband radar signal interference detection method based on spectrum sensing. BACKGROUND

[0002] Ultra-wideband radar is highly concerned in the civil field due to its high range resolution, low probability of intercept, strong anti-interference, anti-stealth target and strong penetration to the ground and wall.

[0003] However, due to the progress of interference technology and the complexity of the ultra-wideband radar signal spectrum space, the ultra-wideband radar signal is easily interfered in the working frequency band. The interference signal forms include various radio frequency interference (RFI) and repeater interference caused by special jammer. These interferences will cause the radar system to misjudge or miss, seriously affecting the detection efficiency of the ultra-wideband radar and increasing the complexity of subsequent image interpretation. Therefore, the effective detection of interference in the ultra-wideband radar channel has important significance in ensuring the normal operation of the ultra-wideband radar system and the development of anti-interference equipment.

[0004] Spectrum sensing is a key technology of cognitive radio, which refers to obtaining the spectrum usage information in the wireless network through various signal detection and processing methods. In the field of radar, spectrum sensing technology can monitor and analyze the use of radar spectrum to achieve the best utilization and management of the spectrum, and can be used in the field of radar anti-interference. Common spectrum sensing methods include energy detection, matched filter detection and cyclostationary detection. Among them, energy detection has a simple structure, provides an effective trade-off between computational complexity and performance, and does not require prior information about the transmitted signal, but this method is based on signal power implementation, which is easily affected by random noise power, and performs poorly under low signal-to-noise ratio conditions. In order to obtain better detection performance, the matched filter detection method is proposed, but this method requires prior information about the signal and noise, which is usually not feasible in practical applications. Subsequently, researchers proposed feature-based methods, such as cyclostationary detection, although these methods can improve performance, but they have higher computational complexity, which reduces the detection efficiency in real-time spectrum sensing scenarios.

[0005] In recent years, the development of deep learning has introduced new ideas and methods for spectrum sensing research. Deep learning is good at nonlinear modeling and adaptive learning, significantly improving the detection performance and speed of spectrum sensing. Since traditional deep learning methods all need to construct a feature vector, this will cause the loss of the original structure information of the received signal to some extent, affecting the performance of the algorithm. In order to solve this problem, researchers have proposed long short-term memory networks (LSTM), which can effectively transmit and express information in long time series and will not cause useful information to be forgotten long time ago. At the same time, LSTM can also solve the gradient disappearance and explosion problem in RNN. Currently, several famous network models have been applied to spectrum sensing technology, including convolutional neural networks (CNN) and long short-term memory networks (LSTM), and it has been proven that spectrum sensing algorithms based on deep learning are more superior than existing methods.

[0006] Despite the above progress, due to the complex electromagnetic environment of ultra-wideband radar signals, there are various types of interference and changes over time, and the existing network cannot well adapt to the complex electromagnetic environment. Therefore, it is still a formidable challenge for ultra-wideband radar to achieve robust anti-interference performance in a dynamic environment. SUMMARY

[0007] To solve the above technical problems, the present application provides an ultra-wideband radar signal interference detection method based on spectrum sensing, which improves the LSTM network and proposes a new spectrum sensing network, which can better adapt to the complex electromagnetic environment possessed by the ultra-wideband radar system and achieve a 99% interference detection probability in various interference scenarios, even under strong electromagnetic interference conditions. It shows good robustness.

[0008] The technical scheme adopted by the present application is: an ultra-wideband radar signal interference detection method based on spectrum sensing, the specific steps are as follows:

[0009] S1, model the ultra-wideband radar echo signal and the interference signal;

[0010] S2, based on step S1, generate a large number of interference data sets for network learning;

[0011] S3, based on step S2, transform the interference echo matrix to the frequency domain, divide it into pulse vectors along the azimuth direction, and construct a pulse spectrum sensing model;

[0012] S4, establish an interference sensing network model including an offline learning network and an online test network;

[0013] S5, offline train the offline learning network of step S4 to develop a detector with a length of L;

[0014] S6, performing sliding detection on the echo signal processed in step S3 using the detector developed in step S5 in the online test network in step S4;

[0015] S7, setting the interference part in the echo detected in step S6 to 0, and completing the interference detection of the ultra-wideband radar signal.

[0016] Further, the step S1 is specifically as follows:

[0017] S11, modeling the ultra-wideband radar echo signal model based on the ASC model;

[0018] The echo signal model expression is as follows:

[0019]

[0020] Wherein, η and f respectively represent the azimuth time and the distance frequency, I represents that the echo includes I scattering points, A i represents the scattering coefficient of the i-th scattering point, α i represents the reflection coefficient of the i-th reflection point, represents the additional scattering phase of the i-th scattering point, ω r (·) and ω a (·) represent the rectangular window of the echo signal in the distance direction and the azimuth direction, represents the azimuth imaging center, R i (η) represents the distance history of the i-th scattering center, k r represents the frequency modulation of the transmitted signal, c and f c respectively represent the light speed and the center frequency of the transmitted signal.

[0021] According to the principle of stationary phase POSP, the echo expression in two-dimensional time domain is as follows:

[0022]

[0023] Wherein, t represents the distance sampling time, and λ represents the radar wavelength.

[0024] Then, the echo signal expression along the azimuth direction in the ultra-wideband radar signal echo is as follows:

[0025]

[0026] S12, modeling the interference signal;

[0027] The expression of the frequency modulation wideband interference is as follows:

[0028]

[0029] Wherein, Q represents the total number of widebands, B n(η) represents the complex amplitude of the qth wideband jammer in η pulses, K q and f cq respectively represent the frequency modulation and center frequency of the qth wideband jammer.

[0030] The sinusoidal modulation SM jammer expression is as follows:

[0031]

[0032] where β q and respectively represent the modulation index and initial phase of the qth wideband jammer at azimuthal sample η, U q represents the amplitude of the qth sinusoidal jammer, f cq represents the center frequency of the qth wideband jammer.

[0033] The noise blocking jammer is obtained by adding a rectangular window to the Gaussian white noise in the frequency domain, and its power spectral density expression is as follows:

[0034]

[0035] where B represents the jammer bandwidth, a represents the energy intensity of the noise blocking jammer, f v represents the frequency band of the jammer signal.

[0036] Further, the step S2 is specifically as follows:

[0037] S21, simulate three forms of jamming;

[0038] (1) wideband suppression jamming from a single jammer source, the set jammer bandwidth accounts for p% of the signal frequency band;

[0039] (2) joint jamming from multiple sources within the same bandwidth, the jamming frequency band accounts for q% of the signal;

[0040] (3) jamming caused by multiple sources distributed in different frequency bands, the jammed signal accounts for r% of the signal.

[0041] Wherein, the values of p, q, and r cannot exceed 50, that is, the proportion of the jamming frequency band to the total signal frequency band cannot exceed 50%. For each signal, the center frequency and frequency modulation slope of the jamming are randomly determined when generated. U samples are generated for each of the three forms of jamming, and 70% of them are used as the training set and 30% are used as the test set.

[0042] S22, the jammed echo signal expression is as follows:

[0043] S r (t) = S t (t) + WBI CM (t) + WBISM (t) + Ψ(G(f))

[0044] where Ψ(·) denotes the transform that maps the power spectrum to the time domain signal, S t (t) denotes the echo signal expression along the azimuth direction in the super wideband radar signal echo given in step S11, WBI CM (t), WBI SM (t) and G(f) denote the frequency modulation wideband interference signal expression, the sinusoidal modulation interference signal expression and the noise blocking interference signal power spectrum density expression given in step S12, respectively.

[0045] Further, the step S3 is specifically as follows:

[0046] S31, transform the interference echo matrix to the frequency domain, and divide into pulse vectors along the azimuth direction;

[0047] The Fourier transform of the interference echo in the distance direction is performed to obtain the frequency spectrum expression of the nth pulse signal of the echo as follows:

[0048]

[0049] where N denotes the number of pulses in the echo that need to be interfered and detected, S r_n (t) denotes the nth interfered pulse echo, obtained from step S22, denotes the Fourier transform of the nth pulse echo, f r denotes the distance sampling frequency.

[0050] Suppose that the interference echo collected by the super wideband radar once is an N*M matrix, and the matrix is discretized to have:

[0051]

[0052] where (·) T denotes the transpose of the matrix, and M denotes the length of each pulse sequence.

[0053] S32, construct a spectrum sensing model of the pulse;

[0054] Combined with the signal processing model of the spectrum sensing algorithm, the spectrum sensing model of the N pulses is expressed as follows:

[0055]

[0056] where x n (m) denotes the echo signal transmitted by the MWP-SAR in the nth pulse, ξ n (m) denotes other sources or interference existing in the same pulse, and v n(m) represents a noise vector specific to the n-th pulse.

[0057] Further, the step S4 is specifically as follows:

[0058] First, the interference-aware network architecture is constructed, which includes two convolutional Conv layers and a dense layer, two long short-term memory LSTM layers, and finally two dense layers.

[0059] The softmax activation function is used in the last two dense layers, and the ReLu activation function is applied in the previous layers to process nonlinear data.

[0060] Then, Dropout is added after each layer of the network to prevent overfitting and enhance the generalization ability of the model, and the network hyperparameters are optimized through extensive cross-validation to ensure the best performance of the interference detection task.

[0061] The network hyperparameters include the kernel size, filter size, stride, padding size of each Conv layer, the number of units of each LSTM layer, the number of neurons of each Dense, batch size, and dropout rate.

[0062] Further, the step S5 is specifically as follows:

[0063] S51, input the training set samples generated in step S21 into the offline learning network in step S4;

[0064] Let each sample be a complex number with length L, and the sample set expression is as follows:

[0065] {X, Y, D} = {(x (1) , y (1) , d (1) ),..., (x (k) , y (k) , d (k) )}

[0066] Where k = 1...K, K represents the number of samples, {X, Y, D} represents a set containing K samples, (x (k) , y (k) , d (k) ) represents the k-th sample set, x (k) represents the real part of the echo signal with length L, y (k) represents the imaginary part of the echo signal with length L, both are 1xL vectors. x (k) , y (k) together form the input to the training network, d (k) represents the label of the k-th sample, d (k) = 0 and d (k)= 1 respectively represent the two criteria H0 and H1 described in step S32.

[0067] S52, based on step S51, offline training is performed to optimize the probability of correctly detecting interference;

[0068] The goal of offline training is to optimize the probability of correctly detecting interference, and the optimization objective expression is as follows:

[0069]

[0070] wherein H i is set to be H i , represents the detection probability under H i ; the decision threshold of the statistical quantity T is represented by γ; the detection probability P d and the false alarm probability P f are respectively defined as P d = P{T > γ | H1} and P f = P{T > γ | H0}, and P represents probability; represents the minimum acceptable probability value in interference detection.

[0071] S53, a detector with a length of L is developed;

[0072] The output obtained after the training sample passes through the offline learning network is written as a 2x1 score vector, and the expression is as follows:

[0073]

[0074] wherein h θ (·) and respectively represent the expected distribution of the offline learning network output under θ parameters and the expected output under decision H i .

[0075] Then the probability expressions of the two hypotheses are as follows:

[0076]

[0077] wherein P(d (k) = 1 | x (k) , y (k) ; θ) represents the probability that the kth sample output of the offline learning network output under θ parameters satisfies the H1 criterion, and P(d (k) = 0 | x (k) , y (k) ; θ) represents the probability that the kth sample output of the offline learning network output under θ parameters satisfies the H0 criterion.

[0078] The objective of DNN training is to maximize the likelihood, expressed as follows:

[0079]

[0080] where J(θ) represents the objective function.

[0081] The solution θ of the optimal parameter under the maximum posteriori (MAP) criterion * The expression is as follows:

[0082]

[0083] Further, the step S6 is specifically as follows:

[0084] S61, input the interference echo to be detected to the online test network;

[0085] The echo data in the test set generated in step S21 is subjected to the operation in step S1, and the echo matrix is expressed as follows:

[0086]

[0087] wherein, represents the echo matrix of the kth sample, represents the target echo signal of the Lth point in the kth sample signal.

[0088] S62, using the detector to perform sliding detection on the echo signal processed in step S61;

[0089] After obtaining the detector of length L trained through step S5, the MWP-SAR data is quickly processed in a simplified manner, and the criterion expression is as follows when detecting:

[0090]

[0091] wherein, represents the criterion formed by the output of the kth sample after the online test network. When , it indicates that the interference is detected, and when , it indicates that the interference is not detected. γ represents the discrimination threshold required in the detection process, which is derived from the false alarm constraint.

[0092] After being determined by the online test network, the disturbed components in the spectrum are accurately labeled.

[0093] Further, the step S7 is specifically as follows:

[0094] The interference part in the interference echo frequency domain model is set to 0, and the interference signal is eliminated, and the expression is as follows:

[0095]

[0096] in, This represents the N*M dimensional echo matrix obtained from a single acquisition by an ultra-wideband radar after interference removal.

[0097] The beneficial effects of this invention are as follows: First, the method of this invention models the ultra-wideband radar echo signal and interference signal to generate a large dataset of interfered signals for network learning. Then, it transforms the interfered echo matrix to the frequency domain, divides it into pulse vectors along the azimuth direction, and constructs a pulse spectrum sensing model. Next, it establishes an interference sensing network model including an offline learning network and an online testing network. The offline learning network is trained offline, and a detector of length L is developed. The online testing network uses the detector to perform sliding detection on the processed echo signal. Finally, the interference portion in the detected echo is set to 0, completing the ultra-wideband radar signal interference detection. The novel spectrum sensing network proposed by this invention can better adapt to the complex electromagnetic environment of ultra-wideband radar systems, and has the advantages of high interference detection probability, fast network learning speed, and easy deployment on hardware platforms. Attached Figure Description

[0098] Figure 1 This is a flowchart of an ultra-wideband radar signal interference detection method based on spectrum sensing according to the present invention.

[0099] Figure 2 The above are time-frequency diagrams of three types of interference signals in an embodiment of the present invention.

[0100] Figure 3 This is a network architecture diagram of the interference sensing network in an embodiment of the present invention.

[0101] Figure 4 This is a schematic diagram illustrating the detection effects of the method of the present invention and existing different detection networks on different interference data in an embodiment of the present invention.

[0102] Figure 5 This is a schematic diagram illustrating the change in detection probability relative to false alarm probability between the method of the present invention and existing different networks in an embodiment of the present invention.

[0103] Figure 6 This is a schematic diagram illustrating the different learning speeds of the method of the present invention and existing networks in this embodiment. Detailed Implementation

[0104] The method of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0105] like Figure 1 The flowchart shown is a method for detecting interference in ultra-wideband radar signals based on spectrum sensing according to the present invention. The specific steps are as follows:

[0106] S1, modeling the ultra-wideband radar echo signal and the interference signal;

[0107] S2, generating a large number of interference data sets for network learning based on step S1;

[0108] S3, transforming the interference echo matrix to the frequency domain based on step S2, dividing it into pulse vectors along the azimuth direction, and constructing a pulse spectrum sensing model;

[0109] S4, establishing an interference sensing network model including an offline learning network and an online test network;

[0110] S5, offline training the offline learning network of step S4 to develop a detector with a length of L;

[0111] S6, using the detector developed in step S5 in the online test network of step S4 to perform sliding detection on the echo signal processed in step S3;

[0112] S7, setting the interference part in the echo detected in step S6 to 0 to complete the ultra-wideband radar signal interference detection.

[0113] In this embodiment, step S1 is specifically as follows:

[0114] S11, modeling the ultra-wideband radar echo signal model based on the ASC model;

[0115] The echo signal model expression is as follows:

[0116]

[0117] wherein η and f represent the azimuth time and the range frequency, respectively, I represents that the echo includes I scattering points, A i represents the scattering coefficient of the i-th scattering point, α i represents the reflection coefficient of the i-th reflection point, represents the additional scattering phase of the i-th scattering point, ω r (·) and ω a (·) represent the rectangular window of the echo signal in the range direction and the azimuth direction, represents the azimuth imaging center, R i (η) represents the distance history of the i-th scattering center, k r represents the frequency modulation of the transmitted signal, c and f c represent the speed of light and the center frequency of the transmitted signal, respectively.

[0118] According to the principle of stationary phase (POSP), the echo expression in two-dimensional time domain is as follows:

[0119]

[0120] where t represents the distance sampling time, and λ represents the radar wavelength.

[0121] The echo signal expression of the ultra-wideband radar signal echo sampled along the azimuth direction is as follows:

[0122]

[0123] S12, model the interference signal;

[0124] The expression of the frequency-modulated wideband interference is as follows:

[0125]

[0126] where Q represents the total number of widebands, B n represents the complex amplitude of the qth wideband interference under η pulses, K q and f cq represent the frequency modulation rate and the center frequency of the qth wideband interference respectively.

[0127] The expression of the sinusoidal modulation (SM) interference is as follows:

[0128]

[0129] where β q and represent the modulation index and the initial phase of the qth wideband interference at the azimuth sampling η, U q represents the amplitude of the qth sinusoidal interference signal, f cq represents the center frequency of the qth wideband interference.

[0130] The noise blocking interference is obtained by adding a rectangular window to the Gaussian white noise in the frequency domain, and the power spectral density expression is as follows:

[0131]

[0132] where B represents the interference bandwidth, a represents the energy intensity of the noise blocking interference, f v represents the frequency band of the interference signal.

[0133] In the embodiment, the step S2 is specifically as follows:

[0134] S21, considering the complex spectrum environment in the operation of the ultra-wideband radar, simulate three forms of interference;

[0135] (1) wideband suppression interference originating from a single interference source, and the set interference bandwidth accounts for p% of the signal frequency band;

[0136] (2) joint interference from multiple sources within the same bandwidth, interfering band occupies q% of the signal;

[0137] (3) interference from multiple sources distributed in different frequency bands, interfering signal occupies r% of the signal.

[0138] Wherein, in this embodiment, the value of p, q, r cannot exceed 50, that is, the proportion of the interfering frequency band to the total signal frequency band cannot exceed 50%; p is set to 40, q is set to 25, and r is set to 50; for each signal, the center frequency and frequency modulation slope of the interference are randomly determined when generated. U=10000 samples are generated for each of the three forms of interference, 70% of which are used as the training set and 30% of which are used as the test set. The time-frequency diagram of the three interference signals is shown in Figure 2 . Figure 2 (a) is the time-frequency diagram of the first interference; Figure 2 (b) is the time-frequency diagram of the second interference; Figure 2 (c) is the time-frequency diagram of the third interference.

[0139] The expression of the echo signal interfered is as follows:

[0140] S r (t) = S t (t) + WBI CM (t) + WBI SM (t) + Ψ(G(f))

[0141] Wherein, Ψ(·) represents the transformation of power spectrum to time domain signal, S t (t) represents the expression of the echo signal sampled along the azimuth direction in the ultra-wideband radar signal echo given in step S11, WBI CM (t), WBI SM (t) and G(f) represent the expression of the frequency modulation wideband interference signal, the expression of the sinusoidal modulation interference signal and the expression of the noise blocking interference signal power spectrum density given in step S12, respectively.

[0142] In this embodiment, the step S3 is specifically as follows:

[0143] S31, transform the echo matrix after interference to the frequency domain, divide into pulse vectors along the azimuth direction, which is convenient for subsequent interference detection of each pulse separately;

[0144] The distance Fourier transform of the echo after interference is performed, and the spectral expression of the nth pulse signal of the echo is as follows:

[0145]

[0146] Wherein, N represents the number of pulses in the echo that need to be interfered and detected, Sr_n (t) represents the nth interfered pulse echo, obtained by step S22, represents the Fourier transform of the nth pulse echo, f r represents the distance sampling frequency.

[0147] Suppose that the interference echo collected by the ultra-wideband radar once is an N*M matrix, and the matrix is discretized, that is:

[0148]

[0149] Where (·) T represents the transpose of the matrix, and M represents the length of each pulse sequence.

[0150] S32, construct a spectrum sensing model of the pulse;

[0151] Combined with the signal processing model of the spectrum sensing algorithm, the spectrum sensing model of the N pulses is expressed as follows:

[0152]

[0153] Where x n (m) represents the echo signal transmitted by the MWP-SAR in the nth pulse, ξ n (m) represents other sources or interference existing in the same pulse, v n (m) represents the noise vector specific to the nth pulse.

[0154] In this embodiment, the step S4 is specifically as follows:

[0155] First, the network architecture of interference sensing is constructed, which includes two convolution (Conv) layers and one dense layer, two long short-term memory (LSTM) layers, and finally two dense layers. The network architecture of interference sensing is shown in Figure 3 .

[0156] Where the Conv layer is good at capturing spatial features important for identifying interference patterns, and the LSTM layer is good at processing time series data, which is crucial for tracking interference that changes dynamically over time. The integration of these layers allows the network to identify complex interference patterns that change in space and time.

[0157] In order to ensure robust classification, the softmax activation function is used in the last two dense layers, and the ReLu activation function is applied in the previous layers (i.e. the two convolution (Conv) layers and one dense layer, two long short-term memory (LSTM) layers) to process nonlinear data.

[0158] Then Dropout is added after each layer of the network to prevent overfitting and enhance the generalization ability of the model, and the network hyperparameters are optimized through extensive cross-validation to ensure the best performance of the interference detection task.

[0159] The network hyperparameters include: kernel size, filter size, stride, padding size of each Conv layer, units of each LSTM layer, neurons of each Dense, batch size, dropout rate. The network hyperparameters in the embodiment are shown in Table 1.

[0160] Table 1

[0161] Hyperparameters Values Kernel size for each Conv layer 30&60 Filter size 9 Stride 1 Padding size 4 Units for each LSTM layer 128 Neurons for each Dense 128&64&2 Batch size 2048 Dropout rate 0.2

[0162] In the embodiment, the step S5 is specifically as follows:

[0163] S51, input the training set sample generated in step S21 into the offline learning network in step S4;

[0164] Suppose each sample is a complex number with length L, and the sample set expression is as follows:

[0165] {X,Y,D}={(x (k) ,y (k) ,d (k) ),...,(x (k) ,y (k) ,d (k) )}

[0166] Wherein, k = 1...K, K represents the number of samples, {X,Y,D} represents a set containing K samples, (x (k) ,y (k) ,d (k) ) represents the kth sample set, x (k) represents the real part of the echo signal with length L, y i represents the imaginary part of the echo signal with length L, both are 1xL vectors. x i ,y i together form the input of the training network, d d represents the label of the kth sample, d f = 0 and d d = 1 respectively represent H0 and H1 two criteria in step S32.

[0167] S52, based on step S51, offline training is performed to optimize the probability of correctly detecting interference;

[0168] The goal of offline training is to optimize the probability of correctly detecting interference, and the optimization goal expression is as follows:

[0169]

[0170] where T|H i denotes the detection state under H i , denotes the detection probability under H i ; the decision threshold of the statistic T is denoted by γ; the detection probability P d and the false alarm probability P f are defined as P d = P{T > γ | H1} and P f = P{T > γ | H0}, respectively, where P denotes probability. denotes the minimum acceptable probability value in the interference detection.

[0171] S53, developing a detector with length L;

[0172] The output of the training sample after passing through the offline learning network is written as a 2x1 score vector, and the expression is as follows:

[0173]

[0174] where h θ (·) and denote the expected distribution of the offline learning network output under θ parameters and the expected output under decision H i , respectively.

[0175] The probability expressions of the two hypotheses are as follows:

[0176]

[0177] where P(d (k) = 1 | x (k) , y (k) ; θ) denotes the probability that the kth sample output of the offline learning network output under θ parameters satisfies the H1 criterion, and P(d (k) = 0 | x (k) , y (k) ; θ) denotes the probability that the kth sample output of the offline learning network output under θ parameters satisfies the H0 criterion.

[0178] The target of DNN (deep neural network) training is to maximize the likelihood, and the expression is as follows:

[0179]

[0180] where J(θ) denotes the objective function.

[0181] The solution θ * of the optimal parameter under the maximum a posteriori (MAP) criterion is as follows:

[0182]

[0183] Unlike existing deep networks, the method of the present application significantly enhances the adaptability of the network in the ultra-wideband radar environment. In the present embodiment, the step S6 is specifically as follows:

[0184] S61, input the interference echo to be detected to the online test network;

[0185] The echo data in the test set generated in step S21 is subjected to the operation in step S1, and the echo matrix expression is as follows:

[0186]

[0187] wherein, represents the echo matrix of the kth sample, represents the echo signal of the Lth point target in the kth sample signal.

[0188] S62, using the detector to perform sliding detection on the echo signal processed in step S61;

[0189] After obtaining the detector of length L trained by step S5, the MWP-SAR data is quickly processed in a simplified manner, and the detection criterion expression is as follows:

[0190]

[0191] wherein, represents the criterion formed by the output of the kth sample after the online test network. When , it indicates that the interference is detected, and when , it indicates that the interference is not detected. γ represents the discrimination threshold required in the detection process, which is derived from the false alarm constraint.

[0192] After being determined by the online test network, the interference components in the spectrum are accurately labeled.

[0193] In the present embodiment, the step S7 is specifically as follows:

[0194] In order to prevent subsequent imaging of the interference signal, the interference part in the interference echo frequency domain model is set to 0, and the interference signal is eliminated, and the expression is as follows:

[0195]

[0196] wherein, represents the N*M-dimensional echo matrix after removing the interference in one collection of the ultra-wideband radar.

[0197] To demonstrate the advantage of the detection network proposed by the method of the present application, various existing spectrum sensing algorithms, including CNN network, DNN network and LSTM network, are compared. At the same time, the detection probability of various sensing networks for different interference sets is evaluated, and the effect of each network in the training set and test set is analyzed, and the results are shown in Figure 4 , and the detection probability of different networks under false alarm is shown in Figure 5 . Among them, Figure 4 (a) is the detection probability of the first kind of interference; Figure 4 (b) is the detection probability of the second kind of interference; Figure 4 (c) is the detection probability of the third kind of interference. The specific detection probability values are shown in Table 2.

[0198] Table 2

[0199]

[0200] From the detection results, it can be seen that in the face of complex interference forms, the sensing network proposed by the method of the present application can obtain a higher detection probability, and can better tolerate the false alarm rate.

[0201] In addition, the change mode of the detection probability of each network with the batch of training samples is shown in Figure 6 (i.e. network learning speed diagram), through the analysis of the learning performance of the network, it can be seen that the network proposed by the method of the present application can still obtain good performance in the case of a small amount of training samples. This facilitates the deployment of the network proposed in this paper on the hardware platform. Therefore, the method of the present application can realize the spectrum sensing based ultra-wideband radar signal interference detection.

[0202] In summary, the method of the present application improves the LSTM network in view of its good performance, and proposes a new spectrum sensing network which can better adapt to the complex electromagnetic environment possessed by the ultra-wideband radar system, and realizes a detection probability of 99% under various interference scenarios. Even under strong electromagnetic interference conditions, it shows good robustness, has the advantages of high interference detection probability, fast network learning speed, and is easy to deploy on a hardware platform.

[0203] Those skilled in the art will realize that the embodiments described herein are for the purpose of helping the reader to understand the principles of the present application, and should be understood as not limiting the scope of protection of the present application to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations according to the technical inspiration disclosed in the present application without departing from the essence of the present application, and these modifications and combinations are still within the scope of protection of the present application.

Claims

1. A spectrum sensing-based ultra-wideband radar signal interference detection method, the specific steps being as follows: S1, modeling the ultra-wideband radar echo signal and the interference signal; S2, based on step S1, generating a large number of interfered data sets for network learning; S3, based on step S2, transforming the interfered echo matrix to the frequency domain, dividing it into pulse vectors along the azimuth direction, and constructing a pulse spectrum sensing model; S4, establishing an interference sensing network model including an offline learning network and an online test network; S5, offline training the offline learning network of step S4 to develop a detector with a length of L; S6, using the detector developed in step S5 in the online test network of step S4 to perform sliding detection on the echo signal processed in step S3; S7, setting the interference part in the echo detected in step S6 to 0 to complete the ultra-wideband radar signal interference detection.

2. The method of claim 1, wherein the method is based on spectrum sensing. The step S1 is specifically as follows: S11, modeling the ultra-wideband radar echo signal model based on the ASC model; The echo signal model expression is as follows: where η and f represent azimuth time and range frequency, respectively, I represents that the echo includes I scatterers, A i represents the scattering coefficient of the i-th scatterer, α i represents the reflection coefficient of the i-th reflection point, represents the additional scattering phase of the i-th scatterer, ω r (·) and ω a (·) represent the rectangular window of the echo signal in the range direction and the azimuth direction, η ci represents the imaging center in the azimuth direction, R i (η) represents the distance history of the i-th scattering center, k r represents the frequency modulation of the transmitted signal, c and f c represent the speed of light and the center frequency of the transmitted signal, respectively; According to the principle of stationary phase POSP, the echo expression in two-dimensional time domain is as follows: Where t represents the distance sampling time, and λ represents the radar wavelength; Then the echo signal expression along the azimuth direction in the ultra-wideband radar signal echo is as follows: S12, modeling the interference signal; The expression of the frequency modulation wideband interference is as follows: where Q denotes the total number of widebands, B n (η) denotes the complex amplitude of the qth wideband interference over η pulses, K q and f cq denote the frequency modulation and the center frequency of the qth wideband interference, respectively. The expression of the sinusoidal modulation SM interference is as follows: where β q and denote the modulation index and initial phase of the qth wideband jammer at azimuth sample η, U q denotes the amplitude of the qth sinusoidal jammer, f cq denotes the center frequency of the qth wideband jammer; The noise blocking interference is obtained by adding a rectangular window to the Gaussian white noise in the frequency domain, and the power spectral density expression is as follows: where B denotes an interference bandwidth, a denotes an energy intensity of noise blocking interference, f v denotes a frequency band of an interference signal.

3. The method of claim 2, wherein the method is based on spectrum sensing. The step S2 is specifically as follows: S21, simulating three forms of interference; (1) wideband suppression interference from a single interference source, the interference bandwidth accounts for p% of the signal frequency band; (2) joint interference from multiple sources within the same bandwidth, the interference frequency band accounts for q% of the signal; (3) interference caused by multiple sources distributed in different frequency bands, the interfered signal accounts for r% of the signal; Where the values of p, q, and r cannot exceed 50, that is, the proportion of the interference frequency band to the total signal frequency band cannot exceed 50%; for each signal, the center frequency and frequency modulation slope of the interference are randomly determined when generated; for the three forms of interference, U samples are generated, 70% of which are used as the training set and 30% of which are used as the test set; The expression of the interfered echo signal is as follows: S r (t) = S t (t) + WBI CM (t) + WBI SM (t) + Ψ(G(f)) where Ψ(·) denotes the transform mapping the power spectrum to the time domain signal, S t (t) denotes the expression of the echo signal sampled along the azimuth direction in the ultra-wideband radar signal echo given in step S11, WBI CM (t), WBI SM (t) and G(f) denote the expression of the frequency-modulated wideband jamming signal, the expression of the sinusoidal modulated jamming signal and the expression of the power spectral density of the noise blocking jamming signal given in step S12, respectively.

4. The method of claim 3, wherein the method further comprises: The step S3 is specifically as follows: S31, transforming the interfered echo matrix to the frequency domain and dividing it into pulse vectors along the azimuth direction; The distance Fourier transform of the interfered echo is performed to obtain the frequency spectrum expression of the nth pulse signal of the echo as follows: where N represents the number of pulses in the echo that need to be interfered and detected, S r_n (t) represents the n-th interfered pulse echo, obtained by step S22, represents the Fourier transform of the n-th pulse echo, f r represents the distance sampling frequency; Let the interfered echo collected by the ultra-wideband radar once be an N*M matrix, and the discretization of the matrix is: where (·) T denotes the transpose of a matrix, M denotes the length of each pulse sequence; S32, constructing a pulse spectrum sensing model; Combined with the signal processing model of the spectrum sensing algorithm, the spectrum sensing model of N pulses is expressed as follows: where x n (m) represents the echo signal transmitted by the MWP-SAR in the n-th pulse, ξ n (m) represents other sources or interferences present in the same pulse, v n (m) represents a noise vector specific to the n-th pulse.

5. The method of claim 4, wherein the method is based on spectrum sensing. The step S4 is specifically as follows: First, construct the network architecture of interference sensing, which includes two convolution Conv layers and one dense layer, two long short-term memory LSTM layers, and finally two dense layers; The softmax activation function is used in the last two dense layers, and the ReLu activation function is applied in the previous layers to deal with nonlinear data; Then, Dropout is added after each layer of the network to prevent overfitting and enhance the generalization ability of the model, and the network hyperparameters are optimized through extensive cross-validation to ensure the best performance of the interference detection task. The network hyperparameters include the kernel size, filter size, stride, padding size of each Conv layer, the number of units of each LSTM layer, the number of neurons of each Dense, the batch size, and the dropout rate.

6. The method of claim 5, wherein the method is based on spectrum sensing. The step S5 is specifically as follows: S51, input the training set samples generated in step S21 into the offline learning network in step S4; Let each sample be a complex number with length L, and the sample set expression is as follows: {X,Y,D} = { (x (1) ,y (1) ,d (1) ),...,(x (k) ,y (k) ,d (k) )} where k = 1...K, K represents the number of samples, {X, Y, D} represents a set containing K samples, (x (k) ,y (k) ,d (k) ) represents the kth sample set, x (k) represents the real part of the echo signal with length L, y (k) represents the imaginary part of the echo signal with length L, both are 1xL vectors; x (k) ,y (k) together form the input to the training network, d (k) represents the label of the kth sample, d (k) = 0 and d (k) = 1 represent H0 and H1 two criteria described in step S32, respectively; S52, based on step S51, perform offline training to optimize the probability of correct interference detection; The goal of offline training is to optimize the probability of correct interference detection, and the optimization objective expression is as follows: where H i denotes the detection state under H i , f T|Hi (·) denotes the detection probability under H i ; the decision threshold of the statistic T is denoted by γ; the detection probability P d and the false alarm probability P f of the interference are defined as P d = P{T > γ | H1} and P f = P{T > γ | H0}, respectively, P denoting probability; denotes the lowest acceptable probability value in the detection of interference; S53, develop a detector with length L; The output obtained after the training sample passes through the offline learning network is written as a 2x1 score vector, and the expression is as follows: where h θ (·) and respectively denote the expected distribution of the offline learning network output and the decision H i under the θ parameter; Then the probability expressions of the two hypotheses are as follows: where P(d (k) = 1 | x (k) ,y (k) ; θ) denotes the probability that the kth sample output of the offline learning network satisfies the H1 criterion under the θ parameter, and P(d (k) = 0 | x (k) ,y (k) ; θ) denotes the probability that the kth sample output of the offline learning network satisfies the H0 criterion under the θ parameter. The goal of DNN training is to maximize the likelihood, and the expression is as follows: Where J(θ) represents the objective function. solutions of optimal parameters under maximum a posteriori (MAP) criterion * The expression is as follows:

7. The method of claim 6, wherein the method further comprises: The step S6 is specifically as follows: S61, input the interference echo to be detected into the online test network; Perform the operation in step S1 on the echo data in the test set generated in step S21 to obtain the echo matrix expression as follows: wherein, represents the echo matrix of the kth sample, represents the target echo signal of the Lth point in the kth sample signal; S62, use the detector to perform sliding detection on the echo signal processed in step S61; After obtaining the length L detector trained in step S5, the MWP-SAR data is quickly processed in a simplified manner, and the detection criterion expression is as follows: wherein represents a criterion formed by the output of the kth sample after passing through the online testing network; when represents that interference is detected, when represents that no interference is detected; γ represents a discrimination threshold to be used in the detection process, which is derived from the false alarm constraint; After being determined by the online test network, the interfered components in the spectrum are accurately labeled.

8. The method of claim 7, wherein the method further comprises: The step S7 is specifically as follows: The interference part in the interference echo frequency domain model is set to 0, and the interference signal is eliminated, and the expression is as follows: wherein, represents the N*M dimensional echo matrix after interference removal for one collection of ultra-wideband radar.

Images