A method, system and device for intelligent identification of navigation jamming of a distributed system under non-gaussian noise, and a medium

Abstract

A non-Gaussian noise distributed system navigation jamming intelligent identification method, system, device and medium, the method is: feature extraction is carried out on the navigation jamming signal received by the distributed system, fractional order smooth pseudo-Wigner-Ville distribution, fractional order Cauchy-William distribution and fractional order wavelet transform are constructed, and the three are combined as three-dimensional intelligent representation of the received signal, and the three-dimensional intelligent representation is used to construct the local interference data set; A step-by-step navigation jamming identification system is constructed, and the Swin-Transformer network is used as the local network of the distributed system sub-identification node; Local parameter updating is performed, and the global optimal network parameter is obtained through iteration; The global optimal network parameter is loaded into the global network for identification, realizing the in-class identification of the suppression type navigation jamming signal and the detection and identification of the deception type navigation jamming signal; The global network model is detected and identified as the distributed system received signal of the deception type navigation jamming signal, the separated deception type navigation jamming and navigation signal are demodulated to obtain the symbol sequence and construct the identification statistic and identification threshold, the size of the identification statistic and identification threshold is compared, and the in-class identification of the deception type navigation jamming is realized; Finally, the obtained navigation jamming identification result is merged to obtain the step-by-step system navigation jamming intelligent identification result under non-Gaussian noise.

Description

Technical Field

[0001] This invention belongs to the field of navigation interference monitoring technology, and particularly relates to a method, system, device and medium for intelligent identification of navigation interference in distributed systems under non-Gaussian noise. Background Technology

[0002] Satellite navigation signals offer numerous advantages, including wide coverage, long communication distances, independence from geographical limitations, high reliability, and flexibility, greatly enriching our lives. However, satellite navigation also faces many challenges in practical applications. During transmission, navigation signals are highly susceptible to various types of interference due to their long distances and weak signal power. This interference can range from affecting satellite positioning performance to causing complete signal failure. Therefore, researching fast and accurate interference type identification algorithms to guide subsequent anti-interference algorithm development and improve satellite navigation performance is of great significance.

[0003] Interference signal type identification is actually a typical application of pattern recognition, which can be summarized into three steps: data preprocessing, feature extraction, and classification decision. Currently, many methods for interference identification have been studied, which can be divided into machine learning-based decision-making algorithms and deep learning-based classification decision-making algorithms based on the type of classifier.

[0004] Machine learning-based algorithms include: G.-S. Wang, Q.-H. Ren and Y.-Z. Su, "The interference classification and recognition based on SF-SVM algorithm," 2017 IEEE 9th International Conference on Communication Software and Networks (ICCSN), Guangzhou, China, 2017, pp. 835-841, doi:10.1109 / ICCSN.2017.8230229. This paper proposes an interference classification and recognition algorithm based on signal feature space and SVM, establishes a signal feature space, and has good training speed and classification accuracy. J. Shu, Y. Liao and X. Luan, "An Interference Recognition Method Based on Improved Genetic Algorithm," 2021 7th International Conference on Computer and Communications (ICCC), Chengdu, China, 2021, pp. 496-500, doi:10.1109 / ICCC54389.2021.9674374. This paper improves the problems of premature convergence and insufficient population diversity in genetic algorithms by utilizing the method of crossover and mutation probabilities varying with population diversity. It enhances the performance of genetic algorithms, significantly improves population diversity, and improves the classification performance of the algorithm. Z. Xu and Z.-y. Wu, "Compound Model of Navigation Interference Recognition Based on Deep Sparse Denoising Auto-encoder," 2020 IEEE 3rd International Conference on Information Communication and Signal Processing (ICICSP), Shanghai, China, 2020, pp. 430-435, doi:10.1109 / ICICSP50920.2020.9232127. This paper proposes a composite classification model algorithm based on a deep sparse denoising autoencoder network, which features fast convergence speed and high recognition accuracy.Xu, J., Ying, S. & Li, H. GPS Interference Signal Recognition Based on Machine Learning. Mobile Network Appl 25, 2336–2350 (2020). The algorithm extracts four different entropy features, including power spectral entropy, to establish a mixed entropy dataset. It compares the Support Vector Machine (SVM) and Random Forest (RF) methods. The algorithm is simple and has good recognition accuracy. The disadvantage is that there are too few types of interference signals and it can only satisfy the recognition of specific types of interference. Algorithms based on deep learning include: X. Chen, D. He, X. Yan, W. Yu and T.-K. Truong, "GNSS Interference Type Recognition With Fingerprint Spectrum DNN Method," in IEEE Transactions on Aerospace and Electronic Systems, vol. 58, no. 5, pp. 4745-4760, Oct. 2022, doi:10.1109 / TAES.2022.3167985. This paper utilizes interference fingerprint spectral features (FPS) combined with a deep convolutional neural network (FPS deep convolutional neural network, DNN) to identify navigation interference signals. The proposed FPS makes it easier to distinguish different interference signals, and the proposed deep convolutional neural network greatly improves the recognition accuracy, while still maintaining good recognition performance under low signal-to-noise ratio conditions. J.Yu, M.Alhassoun and RMBuehrer, "Interference Classification Using Deep Neural Networks," 2020 IEEE 92nd Vehicular Technology Conference (VTC2020-Fall), Victoria, BC, Canada, 2020, pp.1-6, doi:10.1109 / VTC2020-Fall49728.2020.9348658. This paper utilizes supervised learning to identify six different types of interference signals, using power spectral density (PSD) and cyclic spectrum as input features of the network. Experimental results show that the accuracy surpasses other neural networks that utilize spectral density features.L.Xie, Y.Zhuang, X.Zhuang and X.Zeng, "Satellite Interference Signal Classification Using Improved MLP Model," 2022 IEEE International Conference on Unmanned Systems (ICUS), Guangzhou, China, 2022, pp.635-640, doi:10.1109 / ICUS55513.2022.9986568. This paper proposes a deception interference identification algorithm that uses an improved multilayer perceptron (MLP) model to identify and classify interference sources by using the rate of change of satellite incident angle as a feature. Experimental results show a higher recognition rate. However, the algorithm can only identify deception interference in specific situations and cannot be generalized to engineering implementation. Guo, Chengjun, Tu, Weijuan, "GNSS Interference Signal Recognition Based on Deep Learning and Fusion Time-Frequency Features," Proceedings of the 34th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+2021), St. Louis, Missouri, September 2021, pp. 855-863. The proposed algorithm combines short-time Fourier transform and WVD transform to input time-frequency features into a CNN for training. Experimental results show that compared to a single time-frequency transform, it improves the accuracy of interference recognition and achieves better recognition results.

[0005] The aforementioned navigation interference identification algorithm can only identify deceptive or suppressive navigation interference signals. It fails when both types of interference exist. Furthermore, in distributed systems, identification algorithms based on a single neural network require local data transmission to the central node, significantly increasing data transmission time and compromising the privacy of the local dataset.

[0006] Based on the above analysis, the problems and shortcomings of existing navigation interference identification algorithms are as follows:

[0007] (1) Existing navigation interference identification algorithms can only classify deceptive navigation interference or suppression interference, and cannot simultaneously identify interference when both types of interference exist.

[0008] (2) Existing research on deceptive navigation interference identification is mostly based on certain pre-existing information. Such deceptive navigation interference identification algorithms cannot be used in real-world scenarios, and most deceptive interference identification algorithms are aimed at generative deceptive interference, with less research on forwarding deceptive interference.

[0009] (3) Existing algorithms based on a single neural network need to transmit local data to the central node in a distributed system. This not only increases the transmission pressure on the network, but also fails to protect the privacy of local data. Summary of the Invention

[0010] To overcome the problems of existing technologies, the present invention aims to provide a method, system, device, and medium for intelligent identification of navigation interference in distributed systems under non-Gaussian noise. The invention loads globally optimal network parameters into a global network for identification, enabling intra-class identification of suppressed navigation interference signals and detection and identification of deceptive navigation interference signals. After separating the distributed system receiver signal identified as deceptive navigation interference by the global network model, the deceptive navigation interference and the navigation signal are obtained. Bandwidth and carrier frequency estimations are performed on the separated deceptive navigation interference and navigation signal, respectively. The resulting symbol sequences are demodulated, and identification statistics and thresholds are constructed. By comparing the magnitudes of the identification statistics and thresholds, intra-class identification of deceptive navigation interference is achieved. Finally, the obtained interference identification results are merged, resulting in intelligent identification of navigation interference in distributed systems under non-Gaussian noise, which effectively weakens the influence of non-Gaussian noise. This invention protects the privacy of the local dataset, enables rapid updating of sub-identification node parameters, and does not increase the data transmission volume of the central node. It can perform intra-class identification of deceptive navigation signals without prior knowledge of the data, and features high efficiency and accuracy in identification.

[0011] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0012] A method for intelligent identification of navigation interference in distributed systems under non-Gaussian noise is proposed. First, features are extracted from the navigation interference signal at the receiver of the distributed system, constructing a fractional-order smooth pseudo-Wigner-Ville distribution, a fractional-order Czesley-Williams distribution, and a fractional-order wavelet transform. These three are combined as a three-dimensional intelligent representation of the receiver signal, and a local interference dataset is constructed using this three-dimensional intelligent representation. Then, a distributed navigation interference identification system is constructed, with a Swin-Transformer network serving as the local network for the sub-identification nodes of the distributed system. Next, local parameter updates are performed, and the globally optimal network parameters are obtained through iteration. The globally optimal network parameters are loaded into the global network used for identification to achieve intra-class identification of suppressed navigation interference signals and detection and identification of deceptive navigation interference signals. Further, the distributed system receiver signal detected and identified as a deceptive navigation interference signal by the global network model is separated. The separated deceptive navigation interference and navigation signal are demodulated to obtain symbol sequences, and identification statistics and identification thresholds are constructed. By comparing the magnitudes of the identification statistics and identification thresholds, intra-class identification of deceptive navigation interference is achieved. Finally, the obtained navigation interference identification results are merged to obtain the intelligent identification result of navigation interference in a distributed system under non-Gaussian noise.

[0013] A method for intelligent identification of navigation interference in a distributed system under non-Gaussian noise includes the following steps:

[0014] Step 1: Extract features from the navigation interference signal x(t) received at the distributed system receiver and construct a fractional-order smooth pseudo-Wigner-Vell distribution W. r Fractional-Williams distribution C r and fractional wavelet transform T r The three elements are combined to form a three-dimensional intelligent representation F of the received signal. r A local interference dataset is constructed using three-dimensional intelligent representation;

[0015] Step 2: Construct a distributed navigation interference identification system, and use the Swin-Transformer network as the local network of the sub-identification nodes of the distributed system to train the local interference dataset of each sub-node locally.

[0016] Step 3: Update the local parameter set obtained after training in Step 2, and transmit the local loss function, local data, and scale to the global network. After iteration, obtain the globally optimal network parameters w. o ;

[0017] Step four, use the globally optimal network parameters w obtained in step three. o Loaded into a global network for identification, it enables intra-class identification of suppressed navigation jamming signals and detection and identification of deceptive navigation jamming signals;

[0018] Step 5: Separate the distributed system receiver signal identified as a deceptive navigation interference signal in Step 4 using the global network model, obtaining the separated deceptive navigation interference and navigation signal; then estimate the bandwidth B and carrier frequency fc for the separated deceptive navigation interference and navigation signal respectively. i The code sequence is obtained by estimating and demodulating the separated deceptive navigation interference and navigation signal. A recognition statistic T and a recognition threshold η are constructed for the code sequence of the separated deceptive navigation interference and navigation signal. By comparing the magnitude of the recognition statistic T and the recognition threshold η, intra-class recognition of deceptive navigation interference is achieved.

[0019] Step six: Combine the interference identification results from steps four and five to obtain the intelligent identification results of navigation interference in the distributed system under non-Gaussian noise.

[0020] The specific process of step one is as follows:

[0021] 1.1 Modeling the navigation jamming signal x(t) at the receiver of the distributed system. The navigation jamming signal is divided into two categories: suppression jamming and deception jamming. Typical suppression jamming types include single-tone jamming, multi-tone jamming, partial frequency band jamming and linear frequency sweeping jamming. Deception jamming is divided into generation deception jamming and forwarding deception jamming.

[0022] A single-tone jamming signal is essentially a sinusoidal signal, and its frequency is typically set within the frequency band of navigation signals; the time-domain expression of a single-tone jamming signal is...

[0023] J(t) = A*cos(2πf c t+θ)

[0024] Where A is the amplitude of the single-tone interference signal, f c The carrier frequency, θ is the initial phase;

[0025] Multi-tone interference signals are composed of the superposition of multiple single-tone interference signal waveforms of different frequencies, and their time-domain expression is as follows:

[0026]

[0027] Among them, A m f m and θ m These represent the amplitude, carrier frequency, and initial phase of the m-th single-tone interference signal that constitutes the multi-factor interference;

[0028] Partial frequency band interference signals concentrate the energy of noise within the frequency band where the navigation signal is located. Their time-domain expression is:

[0029] J(t)=U n (t)cos(2πf ct+θ)

[0030] Among them, U n (t) has a mean of zero and a variance of . White noise, f c θ is the carrier frequency, and θ is the initial phase;

[0031] Linear sweep interference is a signal whose frequency has a linear relationship with time. At any given moment, the interference contains only one frequency, but it has a continuous bandwidth over a period of time. Its expression is as follows:

[0032] J(t)=Acos(2πf c t+πkt 2 +θ)

[0033] Where A is the interference amplitude, f c The carrier frequency is the interference frequency, k is the frequency modulation slope, and θ is the initial phase.

[0034] The expression for the repeater-type deception jamming signal is:

[0035] J(t)=A3b(t-τ)c(t-τ)cos(2πf c (t-τ)+θ)

[0036] Where A3 is the amplitude of the navigation signal, b(t) is the navigation message, and c(t) represents the pseudo-random ranging code; f c Let θ be the carrier frequency and θ be the initial phase.

[0037] Generative deception interference can be represented as:

[0038] J(t)=A2b2(t)c2(t)cos(2πf2t+θ)

[0039] Where A2 represents the amplitude of the generative spoofing jamming signal, b2(t) represents the message information of the generative spoofing jamming signal, c2(t) represents the spread spectrum pseudo-random code of the generative spoofing jamming signal, f2 is the carrier frequency of the generative spoofing jamming signal, and θ is the initial phase of the generative spoofing jamming signal.

[0040] For a distributed system, suppose the system consists of one navigation signal transmitter, one interference signal transmitter, and M receiving nodes. Each receiving node is a single-antenna system. The mathematical representation of the signal received by the i-th receiving node is:

[0041] X i (t)=h i *S(t)+J(t)+V(t)

[0042] Where S(t) represents the navigation signal, J(t) represents the navigation interference signal, V(t) represents the non-Gaussian noise, and hi This represents the fading coefficient of the channel between the navigation signal and the i-th receiver;

[0043] The navigation signal S(t) uses BPSK modulation, and its signal form is represented as follows:

[0044] S(t)=A I b(t)c(t)cos(2πf c t+θ)

[0045] Among them, A I Let f be the amplitude of the navigation signal, b(t) be the navigation message, and c(t) be the pseudo-random ranging code; c Let θ be the carrier frequency and θ be the initial phase.

[0046] The non-Gaussian noise V(t) is represented by an alpha-stable distribution, and its characteristic function expression is as follows:

[0047]

[0048] in:

[0049]

[0050]

[0051] Where α represents the characteristic index, with a value of 0 < α ≤ 2, and the impulse rate of the alpha stable distribution is determined by this parameter; β represents the symmetry parameter, with a value of -1 ≤ β ≤ 1, and the degree of symmetry of the alpha stable distribution is determined by this parameter; γ represents the scale parameter, with a value of γ ≥ 0, and the degree of dispersion of the sample mean relative to the mean can be represented by this parameter, which is equivalent to the definition of variance in the Gaussian distribution; ε is called the location parameter, with a value of -∞ < ε < ∞; sgn(u) represents the sign function;

[0052] 1.2 Calculate the fractional-order smoothed pseudo-Wigner-Vell distribution W of the signal received at the distributed system's receiver. r Fractional-Williams distribution C r and fractional wavelet transform T r Fractional-order smooth pseudo-Wigner-Vell distribution W r It can distinguish the characteristics of deceptive interference and single-tone interference signals, and uses the fractional-order C-Williams distribution. r It can distinguish the characteristics of partial frequency band interference, fractional wavelet transform T r It can distinguish the characteristics of multi-tone interference and linear frequency sweep interference;

[0053] Calculate the fractional smooth pseudo-Wigner-Vell distribution (SPWVD) of the signal received by the distributed system. r :

[0054]

[0055] Calculate the fractional-order C-Williams distribution C of the signal received at the distributed system. r :

[0056]

[0057] Calculate the fractional wavelet transform (WT) of the signal received at the distributed system. r

[0058]

[0059] in, |·| represents the absolute value operation, ψ(t) is the basis function of wavelet transform, h(τ-t) is the analysis window function of short-time Fourier transform, τ represents the scale, s represents the translation amount, and g(u) and h(r) are real even functions.

[0060] The fractional-order smoothed pseudo-Wigner-Vell distribution W r Fractional-Williams distribution C r and fractional wavelet transform T r After dimensional expansion, the components are stitched together to form a three-dimensional intelligent representation F. r ;

[0061] 1.3. Through modeling six types of navigation interference signals—single-tone interference, multi-tone interference, partial-band interference, linear sweep interference, generative spoofing interference, and repeater spoofing interference—it can be seen that both repeater spoofing interference and generative spoofing interference signals use BPSK modulation. The repeater spoofing interference signal and the generative spoofing interference signal are placed in the same local dataset and labeled as spoofing navigation interference. Single-tone interference, multi-tone interference, partial-band interference, and linear sweep interference generate their own independent local datasets for subsequent training of distributed network sub-nodes.

[0062] The specific method for step two is as follows:

[0063] 2.1 A distributed navigation interference identification system based on the ring-fully reduced gradient transfer algorithm is constructed. The distributed network based on the ring-fully reduced gradient transfer algorithm consists of multiple sub-identification nodes and a central fusion node. The central fusion node first initializes the global parameters w. i Then the network parameter w i Assigned to all child recognition nodes; after one round of iteration update using the local dataset, all child nodes transmit the local gradient update amount Δ to other child nodes through the ring-full reduction gradient transfer algorithm, and at the same time, the local loss function loss(w) is also transmitted. i ) and local data size d kThe data is transmitted to the central fusion node, and the global loss function loss(w) is calculated. After all child nodes have received the local gradient update amounts Δ from other child nodes, they take the arithmetic mean of their own gradient update amount and the local gradient update amounts from other child nodes, update their own network parameters using stochastic gradient descent, and begin the next iteration. The formula for calculating the global loss function loss(w) is as follows:

[0064]

[0065] Wherein, loss(w i ) is the loss function of the i-th sub-detection node on the local sample set, d i Let N be the size of the i-th local dataset and N be the number of sub-detection nodes. As can be seen from the above formula, the global loss function is obtained by taking a weighted average after each sub-network is sent to the central node through the transmission network.

[0066] 2.2. The Swin-Transformer network is used as the recognition network for the sub-recognition nodes of the distributed system. The local interference dataset of each sub-node in step one is trained locally. The vector dimension representing the output classification result in all Swin-Transformers is 5.

[0067] The specific method for step three is as follows:

[0068] 3.1 In the distributed navigation interference intelligent identification system based on the ring-full reduction gradient transfer algorithm constructed in step two, the initial network parameters of the central fusion node are w. i , network parameter w i Send to all child identification nodes;

[0069] 3.2 Number the M sub-identification nodes as 1, 2, ..., M. It is stipulated that all sub-identification nodes can only receive data from sub-nodes with a sequence number one greater than their own and can only send data to sub-nodes with a sequence number one less than their own. Sub-identification node No. 1 can only receive data from sub-node No. 2 and can only send data to sub-identification node No. M.

[0070] 3.3. Local training of the sub-recognition nodes is performed. After all sub-recognition nodes have completed one iteration on their local data, a stochastic gradient summation algorithm is used to calculate the local gradient sequentially. After all sub-nodes have updated their local gradients, the local gradient updates of all sub-nodes are divided into M groups in the same order, denoted as... in Let M be the local gradient update group for the i-th sub-identification node, and let L be the number of network parameters. Next, each sub-identification node performs M-1 gradient transmissions. In each iteration, each sub-identification node sends a local gradient update of length L / M to its neighboring sub-nodes with index 1 less than its own, and simultaneously receives a local gradient update of length L / M from its neighboring sub-nodes with index 1 greater than its own, adding it to its corresponding local gradient update group. The local gradient update groups sent and received by each sub-identification node are different. Let k be the local gradient update group sent by the m-th sub-identification node and k-1 be the local gradient update group received. Then, starting from the first iteration... Initially, the local gradient update packet sent in each iteration is the local gradient update packet received by the m-th sub-identification node in the previous iteration. After M-1 gradient transmissions, each sub-detection node has a packet containing the sum of the local gradient update packets corresponding to all sub-identification nodes in its M local gradient update packets. After another M-1 iterations, the iteration operation is as follows: the m-th sub-identification node first sends the (m+1)-th local gradient update packet and receives the m-th local gradient update packet, overwriting the local gradient update packet with the received local gradient update packet, and then continues to send the local gradient update packet obtained by the m-th sub-identification node in the previous iteration.

[0071] 3.4 After each iteration of the ring-to-full reduction algorithm, the local gradient update amount group on all sub-identification nodes is the sum of the local gradient update amount group corresponding to all sub-identification nodes. The updated local gradient update amount is arithmetically averaged, and the updated local parameters of the sub-identification nodes are obtained by using the stochastic gradient descent method, and the next ring-to-full reduction algorithm is performed.

[0072] 3.5, the local loss function loss(w) of the sub-identification node after each round of update. i ) and data size d k The data is transmitted to the central global network to obtain the global loss function loss(w);

[0073] 3.6 According to the definition of the global loss function loss(w) in step two, the global network parameters make w o The goal should be to minimize the global loss function loss(w), i.e.:

[0074] w o =argmin loss(w)

[0075] Due to the unique nature of the ring-total-reduction gradient transfer algorithm, the local parameters of each sub-recognition network are identical after each iteration of the ring-total-reduction gradient transfer algorithm. Therefore, the goal of the algorithm is to find a local network parameter w. iThis causes the global loss function loss(w) to converge;

[0076] Assuming that after multiple iterations of the loop-fully reduced gradient transfer algorithm, the global loss function loss(w) of the distributed network converges, yielding the final globally optimal parameters w. o It should be w i .

[0077] The specific method for step five is as follows:

[0078] First, the fast independent component method is used to perform blind source separation on the mixed signals identified as deceptive navigation interference by the global network model in step three. The specific steps are as follows:

[0079] 5.1 After transposing, normalizing, and centering the signal X(L) identified as deceptive interference at the receiver of the distributed system, matrix Y(L) is obtained; the matrix representation of X(L) is as follows:

[0080]

[0081] 5.2, whiten Y(L) to obtain Z(L);

[0082] 5.3 Select the number of signal categories to be estimated, N=2. Let S(L) be the matrix composed of the deceptive navigation interference and the navigation signal after separation. The matrix representation of S(L) is as follows:

[0083]

[0084] 5.4, ​​Select a randomly initialized weight vector W p ∈R N×M ;

[0085] Version 5.5, updated W p The updated formula is as follows

[0086] in

[0087] 5.6, regarding the updated W in step 5.5 p Perform normalization, let

[0088] 5.7, the normalized W from step 5.6 p To remove the correlation, the calculation formula is:

[0089]

[0090] 5.8, Repeat steps 5.6 to 5.7 until W. p convergence;

[0091] 5.9, using the converged W p Calculate S(L), and let S(L) = W p ×X(L), the matrix S(L) composed of the deceptive navigation interference and the navigation signal after separation is obtained;

[0092] The normalization and centering formulas in step 5.1 are as follows:

[0093] Normalization:

[0094] Centralization:

[0095] The whitening operation in step 5.2 is as follows:

[0096] Find the covariance matrix C of Y(N).

[0097] Perform eigenvalue decomposition on the covariance matrix C, CV = VΛ, where V = [v1, v2, ..., vm] is the matrix composed of eigenvectors by columns, and Λ = diag(λ1, λ2, ..., λm) is the diagonal matrix composed of eigenvalues;

[0098] The whitening data Z(L) of Y(L) is calculated using the following formula:

[0099] Z(L)=Λ (-1 / 2) (M*M)V T (M*M)Y(L)

[0100] Then, using the Welch bandwidth estimation algorithm, the carrier frequencies fc of the two components S1(L) and S2(L) of the matrix S(L) composed of the deceptive navigation interference and the navigation signal after separation are estimated. i and bandwidth B i (i = 1, 2), and then perform linear frequency modulation Z-transformation to further refine the carrier frequency to obtain the precise carrier frequency fc. i The experimental steps are as follows:

[0101] 5.10. Divide the components S1(L) of the matrix S(L) composed of the deceptive navigation interference and the navigation signal after separation into K parts, each part containing P data points. Then, the i-th segment of data can be represented as:

[0102] S i 1(n)=S1(n+iP-P),1≤i≤K,0≤n<P

[0103] 5.11. Load the window function w(n) (Kaiser function) onto all the data segments from step 5.10, and calculate the periodogram for each segment. The periodogram for the i-th data segment is:

[0104]

[0105] In the formula, U is called the normalization factor, and its expression is:

[0106]

[0107] 5.12, Calculate the arithmetic mean of the periodograms of each segment from step 5.11, and finally obtain the power spectrum estimate of S1(L) as follows:

[0108]

[0109] 5.13, find the value in step 5.12 that makes the power spectrum estimate P1(e jw The frequency corresponding to the maximum point is the carrier frequency fc1. P1(e) is estimated using the power spectrum. jw The maximum value of the bandwidth cutoff threshold T is set. hr ;

[0110] 5.14, Find all function values ​​in step 5.12 that are greater than the power spectral density estimate P1(e jw The bandwidth cutoff threshold T set in step 5.13 hr The frequency point, using the maximum frequency point f max Subtract the minimum frequency f min That is, the bandwidth B1 of the estimated signal S1(L) is obtained;

[0111] 5.15. The carrier frequency fc1 in step 5.13 is accurately calculated using the linear swept-frequency Z-transform (CZT). The implementation method of the linear swept-frequency Z-transform algorithm is as follows:

[0112] Command: z k =S1(L)AW -k

[0113] in, Indicates the starting sampling point Let A0 ≤ 1; otherwise, z0 will be outside the unit circle |z| = 1; θ0 represents the phase angle of the initial sampling point z0, and φ0 represents the angle difference between two adjacent sampling points. When φ0 is positive, the sampling point z... k The path rotates counterclockwise; φ0 is a negative number, and the sampling point z k The path rotates clockwise; the magnitude of W0 represents the extension rate of the spiral path. When W0 < 1, the spiral extends outward as the sampling point index k increases; when W0 > 1, the spiral extends inward as the sampling point index k increases; when W0 = 1, the spiral is a circular arc.

[0114] Linear sweep frequency Z-transform (CZT) X(z) k The expression for ) is as follows:

[0115]

[0116] Where M represents the number of points in the linear sweep frequency Z-transform (CZI);

[0117] 5.16, in step 5.15, X(z) k The expression is simplified to obtain

[0118]

[0119] in, * indicates a convolution operation, and the expression for g(n)*h[n] is as follows:

[0120]

[0121] 5.17, Combine the carrier frequency fc1 obtained in step 5.13 with the minimum frequency f obtained in step 5.14. min Midpoint frequency As the starting frequency of the linear sweep frequency Z-transform, the carrier frequency fc1 is compared with the maximum frequency f obtained in (14). max Midpoint frequency As the termination frequency of the linear sweep Z-transform; that is, z in step 5.15 k The expression for the start and end frequencies is:

[0122]

[0123] 5.18. Pad the g(n) sequence obtained in step 5.16 with zeros to obtain a sequence of length H, where H ≥ L + M - 1, and the length of H is an integer power of 2. Therefore:

[0124]

[0125] 5.19. The spectrum G(r) of g(n) padded with zeros in step 5.18 is obtained using the Fast Fourier Transform (FFT) as follows:

[0126]

[0127] 5.20, Pad the h(n) sequence obtained in step 5.16 with zeros to obtain a sequence of length H, we have

[0128]

[0129] 5.21. The spectrum H(r) of h(n) after zero-padding in step 5.20 is obtained using Fast Fourier Transform (FFT) as follows:

[0130]

[0131] 5.22, Multiply G(r) from step 5.19 by H(r) from step 5.21 to obtain Q(r), as follows:

[0132] Q (r) =H(r)G(r)

[0133] 5.23, regarding Q (r) The circular convolution q(n) of h(n) and g(n) obtained by performing an inverse fast Fourier transform is:

[0134]

[0135] 5.24, Take the first M points of q(n) from step 5.23 to calculate X(z). k The calculation formula is as follows:

[0136]

[0137] 5.25, according to step 5.24 X(z) k The expression for X(z) is used to obtain X(z). k The frequency point with the largest absolute value is the precise carrier frequency f′. c1 ;

[0138] 5.26. Repeat steps 5.10 to 5.25 above to obtain the bandwidth B2 and precise carrier frequency f′ of the component S2(L) of the matrix S(L) composed of the deceptive navigation interference and navigation signal after separation in step 5.9. c2 ;

[0139] 5.27. Using the bandwidth and precise carrier frequency of the components S1(L) and S2(L) of the matrix S(L) composed of the deceptive navigation interference and the navigation signal after separation, the navigation signal symbol sequence b2 and the deceptive interference signal symbol sequence b1 are obtained by demodulation.

[0140] 5.28. Utilizing the generation characteristics of two types of deceptive interference, the generative deceptive interference signal is a symbol sequence generated by the interference source itself, and the symbol sequence is different from the navigation signal. On the other hand, the forwarding deceptive interference does not change the navigation symbol sequence, and its symbol sequence is delayed in time but consistent in content with the navigation signal. Therefore, the detection statistic T and the detection threshold η can be constructed by correlating the symbol sequence of the demodulated real navigation signal and the deceptive interference.

[0141] The detection statistic T is:

[0142]

[0143] Where M is the number of symbols obtained by demodulation, b1 is the symbol sequence demodulated from the navigation signal, b2 is the symbol sequence demodulated from the spoofing interference signal, and the detection threshold η is set to 0.8; when T≥η, it is determined that the existing spoofing interference signal is a forwarding spoofing interference; otherwise, it is indicated that the existing spoofing interference signal is a generating spoofing interference.

[0144] The specific method for step six is ​​as follows:

[0145] The identification probability of the deceptive interference obtained in step four is multiplied by the identification probabilities of the two types of deceptive interference obtained in step five. Then, the intra-class identification probabilities of the suppressive navigation interference signal obtained in step four are combined to obtain the final distributed system navigation interference identification result.

[0146] A system for identifying navigation interference in a distributed system under non-Gaussian noise includes:

[0147] Intelligent Representation and Local Interference Dataset Construction Module 1: Extract features from the navigation interference signal x(t) at the receiver of the distributed system and construct a fractional-order smooth pseudo-Wigner-Vell distribution W. r Fractional-Williams distribution C r and fractional wavelet transform T r The three elements are combined to form a three-dimensional intelligent representation F of the received signal. r A local interference dataset is constructed using three-dimensional intelligent representation, which is used to complete the construction of the three-dimensional intelligent representation F in step one. r And the task of generating a local interference dataset;

[0148] Distributed Network Construction and Local Training Module 2 for Sub-Identification Nodes: Construct a step-by-step navigation interference identification system based on the ring-full reduction gradient transfer algorithm. Simultaneously, use the Swin-Transformer network as the local network for the sub-identification nodes of the distributed system to train the local interference datasets of each sub-node locally. This is used to complete the training of the local interference datasets in step one and the construction of the distributed network in step two.

[0149] Module 3 of Distributed Network Parameter Iteration and Global Optimal Network Parameter Training: This module uses the ring-full reduction algorithm to update local parameters and transmits the local loss function, local data, and scale to the global network. Through iteration, the globally optimal network parameters w are obtained. o This is used to complete the training of the distributed network in step two and obtain the globally optimal network parameters w in step three. o ;

[0150] Module 4 for identifying both suppression-type and deceptive navigation interference: This module identifies the globally optimal network parameters w. oLoaded into the global network used for identification, it enables intra-class identification of suppressed navigation jamming signals and identification of deceptive navigation jamming signals, and is used to obtain the globally optimal network parameters w obtained in step three. o Load it into the global network and use the global network to perform the recognition work in step four;

[0151] Module 5 for identifying deceptive navigation interference within a class: This module uses Fast Independent Component Analysis (HICOM) to separate the distributed system receiver signals identified as deceptive navigation interference signals in the global network model of step four, obtaining the separated deceptive navigation interference and navigation signals. The Welch algorithm is then applied to the separated deceptive navigation interference and navigation signals to estimate the bandwidth B and carrier frequency fc, respectively. i The estimated carrier frequency is then refined using the linear frequency modulated Z-transform to obtain the accurate carrier frequency fc. i Using the estimated bandwidth B and the precise carrier frequency fc i The deceptive navigation interference and the navigation signal are demodulated to obtain the symbol sequence. The symbol sequences of the deceptive navigation interference and the navigation signal are correlated and the identification statistics T and the identification threshold η are constructed. By comparing the magnitudes of the identification statistics T and the identification threshold η, the intra-class identification of deceptive navigation interference is achieved.

[0152] Navigation interference identification module 6: merges the interference identification results in steps four and five to obtain the intelligent identification result of navigation interference in the distributed system under non-Gaussian noise, which is used to integrate the interference identification results in steps four and five to obtain the final identification result of the present invention.

[0153] A smart detection device for navigation interference in a non-Gaussian noise distributed system includes:

[0154] Memory, used to store computer programs;

[0155] A processor, configured to implement the intelligent detection method for navigation interference in a distributed system under non-Gaussian noise according to any one of claims 1 to 6 when executing the computer program.

[0156] A computer-readable storage medium includes a computer program stored thereon, and a computer program-readable storage medium storing a computer program that, when executed by a processor, is capable of intelligently identifying navigation interference in a distributed system under non-Gaussian noise.

[0157] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0158] 1. This invention employs a ring-to-full programming gradient transmission algorithm, which, compared to other distributed networks, reduces the communication throughput of the central node and improves the parameter output speed. Furthermore, it eliminates the need to transmit local data and parameters of child nodes to the central node, thus protecting the privacy of local data.

[0159] 2. This invention uses cross-entropy as the loss function and uses the ring-full reduction algorithm to update the weights of the sub-identification nodes. Due to the characteristics of the ring-full reduction algorithm, the weights of all sub-identification nodes are the same after the All-gather operation. After multiple iterations, the globally optimal network parameters are obtained, realizing intra-class identification of suppressed navigation interference.

[0160] 3. The features proposed in this invention can effectively suppress the adverse effects of noise on the signal. By using the generated three-dimensional intelligent representation as network input, the huge amount of computation required to directly process the signal sequence is avoided. The recognition performance improves with the increase of signal-to-noise ratio.

[0161] 4. This invention classifies and identifies deceptive interference by classifying the signal sequences corresponding to the intelligent representations identified as deceptive interference; ultimately realizing an intelligent identification method for navigation interference in distributed systems under non-Gaussian noise. Compared to the limitations of other navigation interference identification algorithms that can only identify either suppressive or deceptive navigation interference, this invention can simultaneously identify both deceptive and suppressive navigation interference signals, increasing the types of identifiable interference.

[0162] In summary, compared with existing technologies, this invention effectively reduces the impact of non-Gaussian noise; protects the privacy of the local dataset; enables rapid updates of sub-identification node parameters without increasing the data transmission volume of the central node; allows for intra-class identification of deceptive navigation signals without prior knowledge of the data; and has the advantages of simple operation, fast identification speed, and high accuracy. Attached Figure Description

[0163] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the invention will be briefly introduced below. Obviously, the drawings shown below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0164] Figure 1 This is a system flowchart of the intelligent identification method for interference in distributed systems under non-Gaussian noise provided in an embodiment of the present invention.

[0165] Figure 2 This is a schematic diagram of the system structure for intelligent identification of interference in a distributed system under non-Gaussian noise, provided in an embodiment of the present invention.

[0166] Figure 3 is a schematic diagram of the intelligent interference identification performance of the distributed system under non-Gaussian noise provided by the embodiment of the present invention. Figure 3(a) shows the performance comparison of the method of the present invention with other single network model algorithms (transformer, resnet50). Figure 3(b) shows the identification performance of the method of the present invention for six types of interference under different signal-to-noise ratio conditions. Detailed Implementation

[0167] The present invention will now be described in detail with reference to the accompanying drawings:

[0168] S101, feature extraction is performed on the navigation interference signal x(t) at the receiver of the distributed system, and a fractional-order smooth pseudo-Wigner-Vell distribution W is constructed. r Fractional-Williams distribution C r and fractional wavelet transform T r The three elements are combined to form a three-dimensional intelligent representation F of the received signal. r A local interference dataset is constructed using three-dimensional intelligent representation;

[0169] S102, Construct a step-by-step navigation interference identification system based on the ring-full reduction gradient transfer algorithm, and use the Swin-Transformer network as the local network of the sub-identification nodes of the distributed system for local training on the local interference dataset of each sub-node.

[0170] S103 uses the ring-to-full reduction algorithm to update local parameters and transmits the local loss function, local data, and scale to the global network. After iteration, the globally optimal network parameters w are obtained. o ;

[0171] S104, the globally optimal network parameters w o Loaded into a global network for identification, it enables intra-class identification of suppressed navigation jamming signals and identification of deceptive navigation jamming signals;

[0172] S105, using the fast independent component analysis method, the distributed system receiver signal identified as a deceptive navigation interference signal in the global network model of the previous step is separated to obtain the separated deceptive navigation interference and navigation signal; the Welch algorithm is used to estimate the bandwidth and carrier frequency of the separated deceptive navigation interference and navigation signal respectively; the estimated carrier frequency is estimated using linear frequency modulation Z-transform to obtain the accurate carrier frequency; the estimated bandwidth and scarce carrier frequency are used to demodulate the separated deceptive navigation interference and navigation signal to obtain the symbol sequence; and identification statistics and identification thresholds are constructed for the symbol sequences of the separated deceptive navigation interference and navigation signal. By comparing the magnitude of the identification statistics and identification thresholds, intra-class identification of deceptive navigation interference is achieved.

[0173] S106. The interference identification results in S104 and S105 are merged to obtain the intelligent identification result of navigation interference in the distributed system under non-Gaussian noise.

[0174] like Figure 2 As shown in the embodiment of the present invention, the intelligent navigation interference system under non-Gaussian noise includes:

[0175] Intelligent Representation and Local Interference Dataset Construction Module 1: Extract features from the navigation interference signal x(t) at the receiver of the distributed system and construct a fractional-order smooth pseudo-Wigner-Vell distribution W. r Fractional-Williams distribution C r and fractional wavelet transform T r The three elements are combined to form a three-dimensional intelligent representation F of the received signal. r A local interference dataset is constructed using three-dimensional intelligent representation, which is used to complete the construction of three-dimensional intelligent representation F in S101. r And the task of generating a local interference dataset;

[0176] Distributed Network Construction and Sub-Identification Node Local Training Module 2: Constructs a step-by-step navigation interference identification system based on the ring-full reduction gradient transfer algorithm, and uses the Swin-Transformer network as the local network of the sub-identification nodes of the distributed system for local training of the local interference datasets of each sub-node; used to complete the training of the local interference dataset in S101 and the construction of the distributed network in S102.

[0177] Module 3 of Distributed Network Parameter Iteration and Global Optimal Network Parameter Training: This module uses the ring-full reduction algorithm to update local parameters and transmits the local loss function, local data, and scale to the global network. Through iteration, the globally optimal network parameters w are obtained. o This is used to complete the training of the distributed network in S102 and obtain the globally optimal network parameters w in S103. o ;

[0178] Module 4 for identifying both suppression-type and deceptive navigation interference: This module identifies the globally optimal network parameters w. o Loaded into the global network used for identification, it enables intra-class identification of suppressed navigation jamming signals and identification of deceptive navigation jamming signals, and is used to obtain the globally optimal network parameters w in S103. o Load it into the global network and use the global network to perform the recognition work in S104;

[0179] Module 5 for identifying deceptive navigation interference within a class: This module uses Fast Independent Component Analysis (HICOM) to separate the distributed system receiver signals identified as deceptive navigation interference signals in the global network model of the previous step, obtaining the separated deceptive navigation interference and navigation signals. The Welch algorithm is then applied to the separated deceptive navigation interference and navigation signals to estimate the bandwidth B and carrier frequency fc, respectively. i The estimated carrier frequency is then refined using the linear frequency modulated Z-transform to obtain the accurate carrier frequency fc. i Using the estimated bandwidth B and the precise carrier frequency fc i The deceptive navigation interference and the navigation signal are demodulated to obtain the symbol sequence. The symbol sequences of the deceptive navigation interference and the navigation signal are correlated and the identification statistics T and the identification threshold η are constructed. By comparing the magnitudes of the identification statistics T and the identification threshold η, the intra-class identification of deceptive navigation interference is achieved.

[0180] Navigation interference identification module 6: merges the interference identification results in S104 and S105 to obtain the intelligent identification result of navigation interference in the distributed system under non-Gaussian noise, which is used to integrate the interference identification results in S104 and S105 to obtain the final identification result of the present invention.

[0181] The invention will be further described below with reference to embodiments:

[0182] like Figure 2 As shown in the embodiment of the present invention, the intelligent navigation interference system under non-Gaussian noise includes:

[0183] Intelligent Representation and Local Interference Dataset Construction Module 1: Extract features from the navigation interference signal x(t) at the receiver of the distributed system and construct a fractional-order smooth pseudo-Wigner-Vell distribution W. r Fractional-Williams distribution C r and fractional wavelet transform T r The three elements are combined to form a three-dimensional intelligent representation F of the received signal. r A local interference dataset is constructed using three-dimensional intelligent representation, which is used to complete the construction of the three-dimensional intelligent representation F in step one. r And the task of generating a local interference dataset;

[0184] Distributed Network Construction and Local Training Module 2 for Sub-Identification Nodes: Construct a step-by-step navigation interference identification system based on the ring-full reduction gradient transfer algorithm. Simultaneously, use the Swin-Transformer network as the local network for the sub-identification nodes of the distributed system to train the local interference datasets of each sub-node locally. This is used to complete the training of the local interference datasets in step one and the construction of the distributed network in step two.

[0185] Module 3 of Distributed Network Parameter Iteration and Global Optimal Network Parameter Training: This module uses the ring-full reduction algorithm to update local parameters and transmits the local loss function, local data, and scale to the global network. Through iteration, the globally optimal network parameters w are obtained. o This is used to complete the training of the distributed network in step two and obtain the globally optimal network parameters w in step three. o ;

[0186] Module 4 for identifying both suppression-type and deceptive navigation interference: This module identifies the globally optimal network parameters w. o Loaded into the global network used for identification, it enables intra-class identification of suppressed navigation jamming signals and identification of deceptive navigation jamming signals, and is used to obtain the globally optimal network parameters w obtained in step three. o Load it into the global network and use the global network to perform the recognition work in step four;

[0187] Module 5 for identifying deceptive navigation interference within a class: This module uses Fast Independent Component Analysis (HICOM) to separate the distributed system receiver signals identified as deceptive navigation interference signals in the global network model of the previous step, obtaining the separated deceptive navigation interference and navigation signals. The Welch algorithm is then applied to the separated deceptive navigation interference and navigation signals to estimate the bandwidth B and carrier frequency fc, respectively. i The estimated carrier frequency is then refined using the linear frequency modulated Z-transform to obtain the accurate carrier frequency fc. i Using the estimated bandwidth B and the precise carrier frequency fc i The deceptive navigation interference and the navigation signal are demodulated to obtain the symbol sequence. The symbol sequences of the deceptive navigation interference and the navigation signal are correlated and the identification statistics T and the identification threshold η are constructed. By comparing the magnitudes of the identification statistics T and the identification threshold η, the intra-class identification of deceptive navigation interference is achieved.

[0188] Navigation interference identification module 6: merges the interference identification results in steps four and five to obtain the intelligent identification result of navigation interference in the distributed system under non-Gaussian noise, which is used to integrate the interference identification results in steps four and five to obtain the final identification result of the present invention.

[0189] The invention will be further described below with reference to embodiments:

[0190] The present invention provides a method for intelligent identification of navigation interference in a distributed system under non-Gaussian noise, comprising the following steps:

[0191] S101, feature extraction is performed on the navigation interference signal x(t) at the receiver of the distributed system, and a fractional-order smooth pseudo-Wigner-Vell distribution W is constructed. r Fractional-Williams distribution C rand fractional wavelet transform T r The three elements are combined to form a three-dimensional intelligent representation F of the received signal. r A local interference dataset is constructed using three-dimensional intelligent representation;

[0192] First, the navigation jamming signal at the receiver of the distributed system is modeled. The navigation jamming signal is divided into two categories: suppression jamming and deception jamming. Typical suppression jamming types include single-tone jamming, multi-tone jamming, partial frequency band jamming and linear frequency sweeping jamming. Deception jamming is divided into generation deception jamming and forwarding deception jamming.

[0193] A single-tone jamming signal is essentially a sinusoidal signal, and its frequency is typically set within the frequency band of navigation signals. The time-domain expression of a single-tone jamming signal is:

[0194] J(t) = A*cos(2πf c t+θ)

[0195] Where A is the amplitude of the single-tone interference signal, f c The carrier frequency, θ is the initial phase;

[0196] Multi-tone interference signals are composed of the superposition of multiple single-tone interference signal waveforms of different frequencies, and their time-domain expression is as follows:

[0197]

[0198] Among them, A m f m and θ m These represent the amplitude, carrier frequency, and initial phase of the m-th single-tone interference signal that constitutes the multi-factor interference;

[0199] Partial frequency band interference signals concentrate the energy of noise within the frequency band where the navigation signal is located. Their time-domain expression is:

[0200] J(t)=U n (t)cos(2πf c t+θ)

[0201] Among them, U n (t) has a mean of zero and a variance of . White noise, f c θ is the carrier frequency, and θ is the initial phase;

[0202] Linear sweep interference is a signal whose frequency has a linear relationship with time. At any given moment, the interference contains only one frequency, but it has a continuous bandwidth over a period of time. Its expression is as follows:

[0203] J(t)=Acos(2πf c t+πkt 2+θ)

[0204] Where A is the interference amplitude, f c The carrier frequency is the interference frequency, k is the frequency modulation slope, and θ is the initial phase.

[0205] The expression for the repeater-type deception jamming signal is:

[0206] J(t)=A3b(t-τ)c(t-τ)cos(2πf c (t-τ)+θ)

[0207] Where A3 is the amplitude of the navigation signal, b(t) is the navigation message, and c(t) represents the pseudo-random ranging code. c Let θ be the carrier frequency and θ be the initial phase.

[0208] Generative deception interference can be represented as:

[0209] J(t)=A2b2(t)c2(t)cos(2πf2t+θ)

[0210] Where A2 represents the amplitude of the generative spoofing jamming signal, b2(t) represents the message information of the generative spoofing jamming signal, c2(t) represents the spread spectrum pseudo-random code of the generative spoofing jamming signal, f2 is the carrier frequency of the generative spoofing jamming signal, and θ is the initial phase of the generative spoofing jamming signal.

[0211] For a distributed system, suppose the system consists of one navigation signal transmitter, one interference signal transmitter, and M receiving nodes. Each receiving node is a single-antenna system. The mathematical representation of the signal received by the i-th receiving node is:

[0212] X i (t)=h i *S(t)+J(t)+V(t)

[0213] Where S(t) represents the navigation signal, J(t) represents the navigation interference signal, V(t) represents the non-Gaussian noise, and h i This represents the fading coefficient of the channel between the navigation signal and the i-th receiver; navigation jamming signals are classified into two types: suppression jamming and deception jamming; typical suppression jamming types include single-tone jamming, multi-tone jamming, partial frequency band jamming, and linear frequency sweep jamming; deception jamming is divided into generation deception jamming and forwarding deception jamming.

[0214] The navigation signal S(t) uses BPSK modulation, and its signal form is represented as follows:

[0215] S(t)=A I b(t)c(t)cos(2πf c t+θ)

[0216] Among them, A I Let f be the amplitude of the navigation signal, b(t) be the navigation message, and c(t) be the pseudo-random ranging code. c Let θ be the carrier frequency and θ be the initial phase.

[0217] The non-Gaussian noise V(t) is represented by an alpha-stable distribution, and its characteristic function expression is as follows:

[0218]

[0219] in:

[0220]

[0221]

[0222] α represents the characteristic index, with a value of 0 < α ≤ 2. The impulse rate of the alpha stable distribution is determined by this parameter; β represents the symmetry parameter, with a value of -1 ≤ β ≤ 1. The degree of symmetry of the alpha stable distribution is determined by this parameter; γ represents the scale parameter, with a value of γ ≥ 0. The dispersion of the sample mean relative to the mean can be represented by this parameter, which is equivalent to the definition of variance in the Gaussian distribution; ε is called the location parameter, with a value of -∞ < ε < ∞; sgn(u) represents the sign function;

[0223] Then, the fractional-order smoothed pseudo-Wigner-Vell distribution W of the signal at the receiver of the distributed system is calculated. r Fractional-Williams distribution C r and fractional wavelet transform T r Fractional-order smooth pseudo-Wigner-Vell distribution W r It can distinguish the characteristics of deceptive interference and single-tone interference signals, and uses the fractional-order C-Williams distribution. r It can distinguish the characteristics of partial frequency band interference, fractional wavelet transform T r It can distinguish the characteristics of multi-tone interference and linear frequency sweep interference;

[0224] Calculate the fractional smooth pseudo-Wigner-Vell distribution (SPWVD) of the signal received by the distributed system. r :

[0225]

[0226] Calculate the fractional-order C-Williams distribution C of the signal received at the distributed system. r :

[0227]

[0228] Calculate the fractional wavelet transform (WT) of the signal received at the distributed system.r

[0229]

[0230] in, |·| represents the absolute value operation, ψ(t) is the basis function of wavelet transform, h(τ-t) is the analysis window function of short-time Fourier transform, τ represents the scale, s represents the translation amount, and g(u) and h(τ) are real even functions.

[0231] W r C r and T r After dimensional expansion, the components are stitched together to form a three-dimensional intelligent representation F. r ;

[0232] Finally, through modeling the above six types of navigation interference signals, it can be seen that both the forwarding navigation interference signal and the generating navigation interference signal adopt the BPSK modulation method. Therefore, the forwarding navigation interference signal and the generating navigation interference signal are placed in the same local dataset and labeled as deceptive navigation interference. Single-tone interference, multi-tone interference, partial frequency band interference and linear frequency sweep interference are generated into their own independent local datasets for subsequent training of distributed network sub-nodes.

[0233] S102, Construct a step-by-step navigation interference identification system based on the ring-full reduction gradient transfer algorithm, and use the Swin-Transformer network as the local network of the sub-identification nodes of the distributed system for local training on the local interference dataset of each sub-node.

[0234] First, a distributed navigation interference identification system based on the ring-fully reduced gradient transfer algorithm is constructed. The distributed network based on the ring-fully reduced gradient transfer algorithm consists of multiple sub-identification nodes and a central fusion node; the central fusion node first initializes the global parameters w. i Then the network parameter w i Assigned to all child recognition nodes; after one round of iteration update using the local dataset, all child nodes transmit the local gradient update amount Δ to other child nodes through the ring-full reduction gradient transfer algorithm, and at the same time, the local loss function loss(w) is also transmitted. i ) and local data size d k The data is transmitted to the central fusion node, and the global loss function loss(w) is calculated. After all child nodes have received the local gradient update amounts Δ from other child nodes, they take the arithmetic mean of their own gradient update amount and the local gradient update amounts from other child nodes, update their own network parameters using stochastic gradient descent, and begin the next iteration. The formula for calculating the global loss function loss(w) is as follows:

[0235]

[0236] Wherein, loss(w i ) is the loss function of the i-th sub-detection node on the local sample set, d i Let N be the size of the i-th local dataset and N be the number of sub-detection nodes. As can be seen from the above formula, the global loss function is obtained by taking a weighted average after each sub-network is sent to the central node via the transmission network.

[0237] Then, the Swin-Transformer network, as the recognition network of the sub-recognition nodes in the distributed system, is trained locally on the local interference dataset of each sub-node in step one; the vector dimension representing the output classification result in all Swin-Transformers in this invention is 5.

[0238] S103 uses the ring-to-full reduction algorithm to update local parameters and transmits the local loss function, local data, and scale to the global network. After iteration, the globally optimal network parameters w are obtained. o ;

[0239] First, in the distributed navigation interference intelligent identification system based on the ring-full reduction gradient transfer algorithm constructed using S102, the central fusion node initializes the network parameters to w. i , network parameter w i Send to all child identification nodes;

[0240] Then, the M sub-identification nodes are numbered 1, 2, ..., M. It is stipulated that all sub-identification nodes can only receive data from sub-nodes with a sequence number one greater than their own and can only send data to sub-nodes with a sequence number one less than their own. Sub-identification node 1 can only receive data from sub-node 2 and can only send data to sub-identification node M.

[0241] Next, local training of the sub-recognition nodes is performed. After all sub-recognition nodes have undergone one iteration on their local data, the local gradients are calculated sequentially using a stochastic gradient summation algorithm. Once all sub-nodes have updated their local gradients, the local gradient updates of all sub-nodes are divided into M groups in the same order, denoted as follows: in Let M be the local gradient update group for the i-th sub-identification node, and let L be the number of network parameters. Next, each sub-identification node performs M-1 gradient transmissions. In each iteration, each sub-identification node sends a local gradient update of length L / M to its neighboring sub-node with a smaller index (1), and simultaneously receives a local gradient update of length L / M from its neighboring sub-node with a larger index (1), adding these to its corresponding local gradient update group. The local gradient update groups sent and received by each sub-identification node are different. Let k be the local gradient update group sent by the m-th sub-identification node and k-1 be the local gradient update group received. Then, starting from the first iteration... Initially, the local gradient update packet sent in each iteration is the local gradient update packet received by the m-th sub-identification node in the previous iteration. After M-1 gradient transmissions, each sub-detection node will have a packet containing the sum of the local gradient update packets corresponding to all sub-identification nodes in its M local gradient update packets. After another M-1 iterations, the iteration operation is as follows: the m-th sub-identification node first sends the (m+1)-th local gradient update packet and receives the m-th local gradient update packet, overwriting the local gradient update packet with the received local gradient update packet, and then continues to send the local gradient update packet obtained by the m-th sub-identification node in the previous iteration.

[0242] After each iteration of the ring-to-full reduction algorithm, the local gradient update amount group on all sub-identification nodes is the sum of the local gradient update amount group corresponding to all sub-identification nodes. The updated local gradient update amount is arithmetically averaged, and the updated local parameters of the sub-identification nodes are obtained by using the stochastic gradient descent method, and then the next ring-to-full reduction algorithm is performed.

[0243] Finally, the local loss function loss(w) of each updated sub-identification node is calculated. i ) and data size d k The data is transmitted to the central global network to obtain the global loss function loss(w);

[0244] According to the definition of the global loss function loss(w) in S102, the global network parameters make w o The goal should be to minimize the global loss function loss(w), i.e.:

[0245] w o =argmin loss(w)

[0246] Due to the unique nature of the ring-total-reduction gradient transfer algorithm, the local parameters of each sub-recognition network are identical after each iteration of the ring-total-reduction gradient transfer algorithm. Therefore, the goal of the algorithm is to find a local network parameter w. iThis causes the global loss function loss(w) to converge;

[0247] Assuming that after multiple iterations of the loop-fully reduced gradient transfer algorithm, the global loss function loss(w) of the distributed network converges, yielding the final globally optimal parameters w. o It should be w i .

[0248] S104, the globally optimal network parameters w o Loaded into a global network for identification, it enables intra-class identification of suppressed navigation jamming signals and identification of deceptive navigation jamming signals;

[0249] Using the global network parameters w obtained from S103 o This is then loaded onto the network model of the global node; enabling intra-class identification of suppressive navigation interference and identification of deceptive navigation interference.

[0250] S105, using the fast independent component analysis method, the distributed system receiver signal identified as a deceptive navigation interference signal in the global network model of the previous step is separated to obtain the separated deceptive navigation interference and navigation signal; the Welch algorithm is used to estimate the bandwidth and carrier frequency of the separated deceptive navigation interference and navigation signal respectively; the estimated carrier frequency is estimated using linear frequency modulated Z-transform to obtain the accurate carrier frequency; the estimated bandwidth and scarce carrier frequency are used to demodulate the separated deceptive navigation interference and navigation signal to obtain the symbol sequence; the symbol sequences of the separated deceptive navigation interference and navigation signal are correlated and identification statistics and identification thresholds are constructed; by comparing the magnitude of the identification statistics and identification thresholds, intra-class identification of deceptive navigation interference is achieved.

[0251] First, the fast independent component method is used to perform blind source separation on the mixed signals identified as deceptive navigation interference by the global network model in S103. The specific steps are as follows:

[0252] (1) The signal X(L) identified as deceptive interference by the receiver of the distributed system is transposed, normalized, and centered to obtain matrix Y(L); the matrix representation of X(L) is as follows:

[0253]

[0254] (2) Whiten Y(L) to obtain Z(L);

[0255] (3) Select the number of signal categories to be estimated as N=2. Let S(L) be the matrix composed of the deceptive navigation interference and the navigation signal after separation. The matrix representation of S(L) is as follows:

[0256]

[0257] (4) Select a randomly initialized weight vector W p ∈R N×M ;

[0258] (5) Update W p The updated formula is as follows

[0259] in

[0260] (6) Update W in (5) p Perform normalization, let

[0261] (7) Normalize W from (6) p To remove the correlation, the calculation formula is:

[0262]

[0263] (8) Repeat process (5)v(7) until W p convergence;

[0264] (9) Using the converged W p Calculate S(L), let S(L) = W p ×X(L), the matrix S(L) composed of the deceptive navigation interference and the navigation signal after separation is obtained;

[0265] The normalization and centering formulas for (1) are as follows:

[0266] Normalization:

[0267] Centralization:

[0268] The whitening operation in (2) is as follows:

[0269] Find the covariance matrix C of Y(N).

[0270] Perform eigenvalue decomposition on the covariance matrix C, CV = VΛ, where V = [v1, v2, ..., vm] is the matrix composed of eigenvectors by columns, and Λ = diag(λ1, λ2, ..., λm) is the diagonal matrix composed of eigenvalues;

[0271] The whitening data Z(L) of Y(L) is calculated using the following formula:

[0272] Z(L)=Λ (-1 / 2) (M;*;M)V T (M*M)Y(L)

[0273] Then, using the Welch bandwidth estimation algorithm, the carrier frequencies fc of the two components S1(L) and S2(L) of the matrix S(L) composed of the deceptive navigation interference and the navigation signal after separation are estimated. i and bandwidth B i (i = 1, 2), and then perform linear frequency modulation Z-transformation to further refine the carrier frequency to obtain the precise carrier frequency fc. i The experimental steps are as follows:

[0274] (10) Divide the components S1(L) of the matrix S(L) composed of the deceptive navigation interference and the navigation signal after separation into K parts, each part containing P data points. Then the i-th segment of data can be represented as:

[0275] S i 1(n)=S1(n+iP-P),1≤i≤K,0≤n<P

[0276] (11) Load the window function w(n) (Kaiser function) onto all the data segments in (10) and calculate the periodogram of each segment. The periodogram of the i-th data segment is:

[0277]

[0278] In the formula, U is called the normalization factor, and its expression is:

[0279]

[0280] (12) The periodicity of each segment in (11) is averaged to obtain the power spectrum estimate of S1(L):

[0281]

[0282] (13) Find the correct (12) so that the power spectrum estimate P1(e jw The frequency corresponding to the maximum point is the carrier frequency fc1. P1(e) is estimated using the power spectrum. jw The maximum value of the bandwidth cutoff threshold T is set. hr ;

[0283] (14) Find all function values ​​in (12) that are greater than the power spectral estimate P1(e jw The bandwidth cutoff threshold T set in (13) hr The frequency point, using the maximum frequency point f max Subtract the minimum frequency f min That is, the bandwidth B1 of the estimated signal S1(L) is obtained;

[0284] (15) The carrier frequency fc1 in (13) is accurately obtained by using linear sweep frequency Z-transform (CZT). The implementation method of the linear sweep frequency Z-transform algorithm is as follows:

[0285] Command: z k =S1(L)AW -k

[0286] in, Indicates the starting sampling point Let A0 ≤ 1; otherwise, z0 will be outside the unit circle |z| = 1; θ0 represents the phase angle of the initial sampling point z0, and φ0 represents the angle difference between two adjacent sampling points. When φ0 is positive, the sampling point z... k The path rotates counterclockwise; φ0 is a negative number, and the sampling point z k The path rotates clockwise; the magnitude of W0 represents the extension rate of the spiral path. When W0 < 1, the spiral extends outward as the sampling point index k increases; when W0 > 1, the spiral extends inward as the sampling point index k increases; when W0 = 1, the spiral is a circular arc.

[0287] Linear sweep frequency Z-transform (CZT) X(z) k The expression for ) is as follows:

[0288]

[0289] Where M represents the number of points in the linear sweep frequency Z-transform (CZI);

[0290] (16) X(z) in (15) k The expression is simplified to obtain

[0291]

[0292] in, * indicates a convolution operation. The expression for g(n)*h[n] is as follows:

[0293]

[0294] (17) Combine the carrier frequency fc1 obtained in (13) with the minimum frequency fc1 obtained in (14). min Midpoint frequency As the starting frequency of the linear sweep frequency Z-transform, the carrier frequency fc1 is compared with the maximum frequency f obtained in (14). max Midpoint frequency As the termination frequency of the linear sweep Z-transform; that is, z in (15) k The expression for the start and end frequencies is:

[0295]

[0296] (18) Pad the g(n) sequence obtained in (16) with zeros to obtain a sequence of length H, where H≥L+M-1, and the length of H is an integer power of 2. Therefore, we have:

[0297]

[0298] (19) The spectrum G(r) of g(n) after zero-padding in (18) is obtained by using Fast Fourier Transform (FFT) as follows:

[0299]

[0300] (20) By padding the h(n) sequence obtained in (16) with zeros, a sequence of length H is obtained, which has

[0301]

[0302] (21) The spectrum H(r) of h(n) after zero-padding in (20) is obtained by using Fast Fourier Transform (FFT) as follows:

[0303]

[0304] (22) Multiply G(r) in (19) by H(r) in (21) to obtain Q(r), as follows:

[0305] Q (r) =H(r)G(r)

[0306] (23) Q (r) The circular convolution q(n) of h(n) and g(n) obtained by performing an inverse fast Fourier transform is:

[0307]

[0308] (24) Take the first M points of q(n) from (23) to calculate X(z). k The calculation formula is as follows:

[0309]

[0310] (25) According to (24)X(z) k The expression for X(z) is used to obtain X(z). k The frequency point with the largest absolute value is the precise carrier frequency f′. c1 ;

[0311] (26) Repeat steps (10)-(25) above to obtain (9) the bandwidth B2 and the precise carrier frequency f′ of the component S2(L) of the matrix composed of the deceptive navigation interference and the navigation signal after separation. c2 ;

[0312] (27) Demodulate the components S1(L) and S2(L) of the matrix S(L) composed of the deceptive navigation interference and the navigation signal after separation, and obtain the navigation signal symbol sequence b2 and the deceptive interference signal symbol sequence b1.

[0313] (28) Utilizing the generation characteristics of two types of deceptive interference, the generative deceptive interference signal is a symbol sequence generated by the interference source itself. The symbol sequence is different from the navigation signal. The forwarding deceptive interference does not change the navigation symbol sequence. Its symbol sequence is delayed in time and consistent in content with the navigation signal. Therefore, the correlation between the real navigation signal after demodulation and the symbol sequence of the deceptive interference can be used to construct the detection statistic T and the detection threshold η.

[0314] The detection statistic T is:

[0315]

[0316] Where M is the number of symbols obtained by demodulation, b1 is the symbol sequence demodulated from the navigation signal, b2 is the symbol sequence demodulated from the spoofing interference signal, and the detection threshold η is set to 0.8; when T≥η, it is determined that the existing spoofing interference signal is a forwarding spoofing interference; otherwise, it is indicated that the existing spoofing interference signal is a generating spoofing interference.

[0317] S106. The interference identification results in S104 and S105 are merged to obtain the intelligent identification result of navigation interference in the distributed system under non-Gaussian noise.

[0318] The identification probability of the deceptive interference obtained in S104 is multiplied by the identification probabilities of the two types of deceptive interference obtained in S105, and then the intra-class identification probabilities of the suppressive navigation interference signal obtained in S104 are combined to obtain the final distributed system navigation interference identification result.

[0319] The technical effects of the present invention will be described in detail below with reference to simulation experiments;

[0320] To evaluate the performance of this invention, simulation verification was conducted. In the simulation experiment, MIMO signals were simulated using Matlab software. The navigation signal was modeled as a BSPK intermediate frequency signal, the BPSK signal symbol rate was 1250Hz, the carrier frequency was 10000Hz, and the sampling frequency was 50000Hz. Interference signals are classified into two types based on their interference mechanisms: suppression interference and deception interference. Suppression interference is further subdivided into: single-tone interference, multi-factor interference, pre-sweep interference, and partial-band interference. Deception interference is divided into: repeater-type deception interference and generator-type deception interference. Forwarding spoofing jamming involves receiving, enhancing, and forwarding navigation signals. The message information is consistent with the navigation signal. To maintain experimental consistency, the navigation message has 16 symbols. The symbols of both forwarding and generating spoofing jamming signals lag behind the navigation signal by two symbols. The non-Gaussian noise is alpha-stable distributed noise, and the channel is a flat fading channel. The signal-to-noise ratio (SNR) range is set from -10dB to 10dB. 1000 data pairs are generated for each SNR, with 256 sampling points per pair. As mentioned in the experiment, after generating the 3D intelligent representation, the two types of spoofing jamming are placed in the same dataset, representing spoofing jamming. A total of 12000 training samples and 3000 test samples are generated. The local sub-identification node network model uses cross-entropy as the loss function and updates the weights of the sub-identification nodes using the ring-full reduction algorithm. Due to the characteristics of the ring-full reduction algorithm, all sub-identification nodes have the same weights after the All-gather operation. After multiple iterations, the globally optimal network parameters are obtained, achieving intra-class recognition of suppressed navigation jamming. The signal sequences corresponding to the intelligent representations identified as deceptive interference are used for classification and identification of deceptive interference; ultimately, an intelligent identification method for navigation interference in distributed systems under non-Gaussian noise is realized. The simulation results of this invention are shown in Figure 3. Figure 3(a) shows a performance comparison between the method of this invention and other single network model algorithms (transformer, resnet50). Figure 3(b) shows the identification performance of the method of this invention for six types of interference under different signal-to-noise ratios. As can be seen from Figure 3(a), compared with existing transformer-based and resnet50-based networks, this network has higher identification performance under the same signal-to-noise ratio. As can be seen from Figure 3(b), the method of this invention has good identification performance for navigation interference under low signal-to-noise ratio conditions and has high robustness.

[0321] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent identification of navigation interference in a distributed system under non-Gaussian noise, characterized in that, Includes the following steps: Step 1: Navigation interference signal at the receiver of the distributed system. Feature extraction was performed, and a fractional-order smooth pseudo-Wigner-Vell distribution was constructed. Fractional-Williams distribution and fractional wavelet transform The three elements are combined to form a three-dimensional intelligent representation of the received signal. A local interference dataset is constructed using three-dimensional intelligent representation; Step 2: Construct a distributed navigation interference identification system, and use the Swin-Transformer network as the local network of the sub-identification nodes of the distributed system to train the local interference dataset of each sub-node locally. 2.1 A distributed navigation interference identification system based on the ring-fully reduced gradient transfer algorithm is constructed. The distributed network based on the ring-fully reduced gradient transfer algorithm consists of multiple sub-identification nodes and a central fusion node. The central fusion node first performs global parameter initialization. Then network parameters Assigned to all child recognition nodes; all child nodes update their local gradients after one iteration using the local dataset. The gradient is transmitted to other child nodes using the ring-fully reduced gradient transfer algorithm, while the local loss function is also transferred. and local data scale Transmitted to the central fusion node and the global loss function is calculated. When all child nodes receive local gradient updates from other child nodes besides themselves. Then, the gradient update of the node itself is arithmetically averaged with the local gradient updates of other child nodes, and the network parameters are updated using stochastic gradient descent. The next iteration then begins. The global loss function... The calculation formula is as follows: in, It is the first The loss function of each sub-detection node on the local sample set For the first The size of a local dataset, The number of sub-detection nodes; as can be seen from the above formula, the global loss function is obtained by taking a weighted average after each sub-network is sent to the central node through the transmission network; 2.

2. The Swing-Transformer network is used as the recognition network for the sub-recognition nodes of the distributed system. The local interference dataset of each sub-node in step one is trained locally. The vector dimension representing the output classification result in all Swing-Transformers is 5. Step 3: Update the local parameter set obtained after training in Step 2, and transmit the local loss function, local data, and scale to the global network. After iteration, obtain the globally optimal network parameters. ; Step four, apply the globally optimal network parameters obtained in step three. Loaded into a global network for identification, it enables intra-class identification of suppressed navigation jamming signals and detection and identification of deceptive navigation jamming signals; Step 5: Separate the distributed system receiver signal identified as a deceptive navigation interference signal in the global network model of Step 4 to obtain the separated deceptive navigation interference and navigation signal; then perform bandwidth calculations on the separated deceptive navigation interference and navigation signal respectively. Estimation and carrier frequency The symbol sequences of the separated deceptive navigation interference and navigation signal are estimated and demodulated. Recognition statistics are then constructed from the symbol sequences of the separated deceptive navigation interference and navigation signal. With recognition threshold Comparison of identification statistics With recognition threshold The size of the target value determines the intra-class identification of deceptive navigation interference. Step six: Combine the interference identification results from steps four and five to obtain the intelligent identification results of navigation interference in the distributed system under non-Gaussian noise.

2. The method as described in claim 1, wherein the specific process of step one is as follows: 1.1 Navigation interference signals at the receiver of a distributed system Modeling reveals that navigation jamming signals can be categorized into two types: suppression jamming and deception jamming. Typical suppression jamming types include single-tone jamming, multi-tone jamming, partial frequency band jamming, and linear frequency sweeping jamming. Deception jamming is divided into generation deception jamming and relay deception jamming. A single-tone jamming signal is essentially a sinusoidal signal, and its frequency is typically set within the frequency band of navigation signals; the time-domain expression of a single-tone jamming signal is... J(t)=A*cos(2πf c t+θ) in, A represents the amplitude of the single-tone interference signal. The frequency of the carrier wave, This is the initial phase; Multi-tone interference signals are composed of the superposition of multiple single-tone interference signal waveforms of different frequencies, and their time-domain expression is as follows: in, , and These represent the amplitude, carrier frequency, and initial phase of the m-th single-tone interference signal that constitutes the multi-factor interference; Partial band interference signals concentrate the energy of noise within the frequency band where the navigation signal is located; its time-domain expression is: J(t)=U n (t)cos(2πf c t+θ) in, With a mean of zero and a variance of , White noise, For carrier frequency, This is the initial phase; Linear sweep interference is a signal whose frequency has a linear relationship with time. At any given moment, the interference contains only one frequency, but it has a continuous bandwidth over a period of time. Its expression is as follows: J(t)=Acos(2πf c t+πkt 2 +i) Where A is the interference amplitude. To interfere with the carrier frequency, For frequency modulation slope, This is the initial phase; The expression for the repeater-type deception jamming signal is: J(t)=A3b(t-τ)c(t-τ)cos(2πf c (t-τ)+θ) in, The amplitude of the navigation signal. Navigation messages, This represents a pseudo-random ranging code; For carrier frequency, This is the initial phase; Generative deception interference can be represented as: J(t)=A2b2(t)c2(t)cos(2πf2t+θ) in, This represents the amplitude of the generative deception interference signal. The message information representing the generative deception jamming signal, The spread spectrum pseudo-random code representing the generative spoofing interference signal. The carrier frequency of the generative deception jamming signal. It is the initial phase of the generative deception interference signal; For a distributed system, suppose the system consists of one navigation signal transmitter and one interference signal transmitter. It consists of several receiving nodes, each of which is a single-antenna system. The mathematical representation of the signal received by each receiving node is: in, Indicates navigation signals, Indicates navigation interference signal, Indicates non-Gaussian noise. Indicates navigation signal to the The fading coefficient of the channel between receivers; Navigation signals The signal form using BPSK modulation is as follows: in, The amplitude of the navigation signal. Navigation messages, This represents a pseudo-random ranging code; For carrier frequency, This is the initial phase; Non-Gaussian noise Represented by the alpha-stable distribution, its characteristic function expression is as follows: in: in, Represents the characteristic index, with a value of The number of pulses with a stable alpha distribution is determined by this parameter; Denotes the symmetry parameter, which takes the value of The degree of symmetry of the alpha stable distribution is determined by this parameter; This represents the scale parameter, with values ​​ranging from 1 to 2. The degree of dispersion of the sample mean relative to the mean can be represented by this parameter, which is equivalent to the definition of variance in a Gaussian distribution; These are called position parameters; their values ​​are... ; Represents a symbolic function; 1.2 Calculate the fractional-order smoothed pseudo-Wigner-Weil distribution of the signal at the receiver of the distributed system. Fractional-Williams distribution and fractional wavelet transform Fractional-order smooth pseudo-Wigner-Vell distribution It can distinguish the characteristics of deceptive interference and single-tone interference signals, using the fractional-order CWilliams distribution. It can distinguish the characteristics of partial frequency band interference using fractional wavelet transform. It can distinguish the characteristics of multi-tone interference and linear frequency sweep interference; Calculate the fractional-order smoothed pseudo-Wigner-Vell distribution of the signal at the receiver of a distributed system. : Calculate the fractional-order Cham-Williams distribution of the signal at the receiver of a distributed system. : Calculate the fractional wavelet transform (WT) of the signal received at the distributed system's receiver. in, , For the absolute value operation, These are the basis functions of the wavelet transform. It is the analysis window function of the short-time Fourier transform. Representative scale, Represents the translation amount. , It is a real even function; Fractional smooth pseudo-Wigner-Vell distribution Fractional-Williams distribution and fractional wavelet transform T r After dimensional expansion, the components are stitched together to form a three-dimensional intelligent representation F. r ; 1.

3. Through modeling six types of navigation interference signals—single-tone interference, multi-tone interference, partial-band interference, linear sweep interference, generative spoofing interference, and repeater spoofing interference—it can be seen that both repeater spoofing interference and generative spoofing interference signals use BPSK modulation. The repeater spoofing interference signal and the generative spoofing interference signal are placed in the same local dataset and labeled as spoofing navigation interference. Single-tone interference, multi-tone interference, partial-band interference, and linear sweep interference generate their own independent local datasets for subsequent training of distributed network sub-nodes.

3. The method as described in claim 1, characterized in that, The specific method for step three is as follows: 3.1 In the distributed navigation interference intelligent identification system based on the ring-full reduction gradient transfer algorithm constructed in step two, the initial network parameters of the central fusion node are as follows: network parameters Send to all child identification nodes; 3.2, regarding Sub-identification nodes are numbered as The rules stipulate that all child identification nodes can only receive data from child nodes with a sequence number one greater than their own, and can only send data to child nodes with a sequence number one less than their own. Child identification node number 1 can only receive data from child node number 2, and child identification node number 1 can only send data to child nodes with a sequence number one less than their own. Number identification node; 3.

3. Local training of the sub-recognition nodes is performed. After all sub-recognition nodes have completed one iteration on their local data, a stochastic gradient summation algorithm is used to calculate the local gradient sequentially. Once all sub-nodes have updated their local gradients, the local gradient updates of all sub-nodes are divided into... Group, represented as ,in For the first The number of sub-identification nodes Grouping local gradient update values ​​into groups, assuming the number of network parameters is . Next, each sub-identification node performs... Subgradient transmission, in each iteration, each sub-identifier node sends a block of length 1 to its neighboring sub-nodes whose index is 1 less than its own. The local gradient update amount, while receiving a block of length of adjacent child nodes with an index one greater than its own. The local gradient update amount is added to the corresponding local gradient update amount group; the local gradient update amount groups sent and received by each sub-identification node are different; let the first... The local gradient update amounts sent by each sub-identification node are grouped as follows: The received local gradient update amounts are grouped as follows: Then, starting from the first iteration, the local gradient update group sent in each iteration is the first one. Each sub-recognition node is grouped based on the local gradient update received in the previous iteration; after... After the sub-gradient transmission, each sub-detection node's Each local gradient update group contains a group that includes the sum of the local gradient update groups corresponding to all sub-identification nodes; then after... The iteration operation is as follows: [Number]... The first sub-identification node sends the first... Grouping local gradient update values ​​and receiving the first... The first local gradient update packet is overwritten with the received local gradient update packet, and then the second local gradient update packet is sent continuously. Grouping of local gradient update amounts obtained by each sub-identification node in the previous iteration; 3.4 After each iteration of the ring-to-full reduction algorithm, the local gradient update amount group on all sub-identification nodes is the sum of the local gradient update amount group corresponding to all sub-identification nodes. The updated local gradient update amount is arithmetically averaged, and the updated local parameters of the sub-identification nodes are obtained by using the stochastic gradient descent method, and the next ring-to-full reduction algorithm is performed. 3.5, the local loss function of the sub-identification nodes after each round of updates. and data scale The data is transmitted to the central global network to obtain the global loss function. ; 3.6 Based on the global loss function in step two The definition of global network parameters makes The global loss function should be made so that Minimum, that is: Due to the unique nature of the ring-to-fully-reduced gradient transfer algorithm, the local parameters of each sub-recognition network are identical after each iteration of the ring-to-fully-reduced gradient transfer algorithm. Therefore, the goal of the algorithm is to find a set of local network parameters. This makes the global loss function convergence; Assuming that after multiple iterations of the loop-fully reduced gradient transfer algorithm, the global loss function of the distributed network... Convergence is achieved, and the final globally optimal parameters are obtained. Should be .

4. The method as described in claim 1, characterized in that, The specific method for step five is as follows: First, the fast independent component method is used to perform blind source separation on the mixed signals identified as deceptive navigation interference by the global network model in step three. The specific steps are as follows: 5.1 Signals identified as deceptive interference at the receiver of a distributed system After transposing, normalizing, and centering, the matrix is ​​obtained. ; The matrix representation is as follows: 5.2, regarding whitening ; 5.3 Select the number of signal categories to be estimated. Let the matrix composed of the deceptive navigation interference and the navigation signal after separation be... , The matrix representation is as follows: 5.4, ​​Select a randomly initialized weight vector ; Version 5.5, Update The updated formula is as follows ; in ; 5.6, Updated in step 5.5 Perform normalization, let ; 5.7, the normalized result from step 5.6 To remove the correlation, the calculation formula is: ； 5.8, Repeat steps 5.6 to 5.7 until... convergence; 5.9, using the converged... calculate ,make The matrix obtained after separation of deceptive navigation interference and navigation signal is as follows: ; The normalization and centering formulas in step 5.1 are as follows: Normalization: ; Centralization: ; The whitening operation in step 5.2 is as follows: beg covariance matrix , ; For covariance matrix Perform feature decomposition. ,in It is a matrix composed of eigenvectors column by column. It is a diagonal matrix composed of eigenvalues; Calculate using the following formula Whitening data : Then, the matrix composed of the deceptive navigation interference and the navigation signal after separation is estimated using the Welch bandwidth estimation algorithm. Two components and carrier frequency and bandwidth The carrier frequency is further refined by linear frequency modulation Z-transform to obtain the precise carrier frequency. The experimental steps are as follows: 5.10, the matrix composed of the separated deceptive navigation interference and navigation signal. The amount Divided into Servings, each serving contains The first data point, then the second... Segment data can be represented as: 5.11, Window Function The Kaiser function loads all the data segments from step 5.10 and calculates the periodogram for each segment. The periodic chart of the segment data is as follows: In the formula, This is called the normalization factor, and its expression is: 5.12, Calculate the arithmetic mean of the periodic graphs for each segment in step 5.11, and finally obtain... The power spectrum is estimated as follows: 5.13, Find the step in step 5.12 that makes the power spectrum estimation... The frequency corresponding to the maximum point is the carrier frequency. Using power spectrum estimation The maximum value sets the bandwidth cutoff threshold. ; 5.14, Find all function values ​​in step 5.12 that are greater than the power spectrum estimate. The bandwidth cutoff threshold set in step 5.13 The frequency point, using the maximum frequency point. Subtract the minimum frequency That is, the estimated signal is obtained. bandwidth ; 5.15, Apply the linear swept-frequency Z-transform (CZT) to the carrier frequency in step 5.

13. The precise, linearly swept-frequency Z-transform algorithm is implemented as follows: make: in, Indicates the starting sampling point Let the length of the vector radius be . ;otherwise Will be located on the unit circle The exterior; Indicates the starting sampling point phase angle, This represents the angle difference between two adjacent sampling points. When it is a positive number, the sampling points The path rotates counterclockwise; For negative numbers, sampling points The path rotates clockwise; The magnitude of represents the elongation of the helical path, when As the index k of the sampling point increases, the spiral extends outward. At that time, with the index of the sampling point The spiral extends inward as the spiral lengthens. At that time, the spiral is a circular arc; Linear sweep frequency Z-transform (CZT) The expression is as follows: in, This represents the number of points in the linear sweep frequency Z-transform CZI; 5.16, in step 5.15 The expression is simplified to obtain in, , This represents the convolution operation. The expression is as follows: 5.17, the carrier frequency obtained in step 5.13 The minimum frequency obtained in step 5.14 Midpoint frequency As the starting frequency of the linear sweep Z-transform, the carrier frequency The maximum frequency obtained in (14) Midpoint frequency As the termination frequency of the linear sweep Z-transform; that is, in step 5.

15. The expression for the start and end frequencies is: 5.18, take the result obtained in step 5.16 Padding the sequence with zeros yields a sequence of length . The sequence, where ,and The length is an integer power of 2, therefore: 5.19, Use Fast Fourier Transform (FFT) to process the zero-padding in step 5.18 Find the spectrum as follows: 5.20, take the result obtained in step 5.16 Padding the sequence with zeros yields a sequence of length . The sequence has 5.21, Use Fast Fourier Transform (FFT) to process the zero-padding in step 5.

20. Find the spectrum as follows: 5.22, in step 5.19 With step 5.21 Multiply to get ,as follows: 5.23, for Performing inverse fast Fourier transform yields and Circular convolution for: 5.24, take the result from step 5.

23. The first M points are used to calculate The calculation formula is as follows: 5.25, according to step 5.24 The expression that makes The frequency point with the largest absolute value is the precise carrier frequency. ; 5.

26. Repeat steps 5.10 to 5.25 above to obtain the matrix composed of the deceptive navigation interference and the navigation signal after separation in step 5.

9. The amount bandwidth and precise carrier frequency ; 5.27, the matrix composed of the separated deceptive navigation interference and navigation signal is as follows: The amount and The respective bandwidths and precise carrier frequencies are demodulated to obtain the navigation signal symbol sequence. With deceptive interference signal symbol sequence ; 5.

28. Utilizing the generation characteristics of two types of deceptive interference, generative deceptive interference signals are symbol sequences generated by the interference source itself, and these symbol sequences differ from the navigation signal. In contrast, forwarding deceptive interference does not alter the navigation symbol sequence; its symbol sequence is delayed in time but identical in content to the navigation signal. Therefore, correlation can be performed between the demodulated real navigation signal and the symbol sequences of the deceptive interference to construct detection statistics. With detection threshold ; Detection statistics for: in, The number of symbols obtained by demodulation. The symbol sequence demodulated from the navigation signal. To deceive, the symbol sequence demodulated from the interference signal is used to detect the threshold. Set to 0.8; when If the existing deceptive interference signal is determined to be a forwarding deceptive interference signal, then the existing deceptive interference signal is a generating deceptive interference signal.

5. The method as described in claim 1, characterized in that, The specific method for step six is ​​as follows: The identification probability of the deceptive interference obtained in step four is multiplied by the identification probabilities of the two types of deceptive interference obtained in step five. Then, the intra-class identification probabilities of the suppressive navigation interference signal obtained in step four are combined to obtain the final distributed system navigation interference identification result.

6. An intelligent recognition system based on the method of any one of claims 1-5, characterized in that, include: Intelligent Representation and Local Interference Dataset Construction Module 1: Navigation Interference Signals for Distributed System Receivers Feature extraction was performed, and a fractional-order smooth pseudo-Wigner-Vell distribution was constructed. Fractional-Williams distribution and fractional wavelet transform The three elements are combined to form a three-dimensional intelligent representation of the received signal. A local interference dataset is constructed using three-dimensional intelligent representation, which is used to complete the construction of three-dimensional intelligent representation in step one. And the task of generating a local interference dataset; Distributed Network Construction and Local Training Module 2 for Sub-Identification Nodes: Construct a step-by-step navigation interference identification system based on the ring-full reduction gradient transfer algorithm, and use the Swin-Transformer network as the local network of the sub-identification nodes in the distributed system for local training on the local interference dataset of each sub-node. Used to complete the training of the local interference dataset in step one and the construction of the distributed network in step two. Module 3 of Distributed Network Parameter Iteration and Global Optimal Parameter Training: This module uses the ring-full reduction algorithm to update local parameters and transmits the local loss function, local data, and scale to the global network. Through iteration, the globally optimal network parameters are obtained. This is used to complete the training of the distributed network in step two and obtain the globally optimal network parameters in step three. ; Module 4 for identifying both suppressive and deceptive navigation interference: This module identifies the globally optimal network parameters. Loaded into the global network used for identification, it enables intra-class identification of suppressed navigation jamming signals and identification of deceptive navigation jamming signals, and is used to obtain the globally optimal network parameters in step three. Load it into the global network and use the global network to perform the recognition work in step four; Module 5 for identifying deceptive navigation interference within a class: This module uses Fast Independent Component Analysis (HICOM) to separate the distributed system receiver signals identified as deceptive navigation interference signals in the global network model of step four, obtaining the separated deceptive navigation interference and navigation signals. The bandwidth of the separated deceptive navigation interference and navigation signals is then calculated using the Welch algorithm. Estimation and carrier frequency The estimated carrier frequency is then refined using the linear frequency modulated Z-transform to obtain the accurate carrier frequency. Using the estimated bandwidth and precise carrier frequency The separated deceptive navigation interference and navigation signal are demodulated to obtain symbol sequences. Correlation is then performed between the symbol sequences of the separated deceptive navigation interference and navigation signal to construct identification statistics. With recognition threshold Comparison of identification statistics With recognition threshold The size of the target value determines the intra-class identification of deceptive navigation interference. Navigation interference identification module 6: merges the interference identification results in steps four and five to obtain the intelligent identification result of navigation interference in the distributed system under non-Gaussian noise, which is used to integrate the interference identification results in steps four and five to obtain the final identification result.

7. An intelligent identification device based on the method of any one of claims 1-5, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the intelligent detection method for navigation interference in a distributed system under non-Gaussian noise according to any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, comprising, wherein the computer-readable storage medium stores a computer program, characterized in that... The computer program readable storage medium stores a computer program that, when executed by a processor, is capable of intelligently identifying navigation interference in a distributed system under non-Gaussian noise based on the method described in any one of claims 1-5.

Images