Robust distributed signal level fusion positioning method

Through sparse recovery and distribution robust optimization theory, the signal source positioning problem under non-uniform arrays and random error models is solved, and the robust positioning of arbitrary structural arrays is achieved, which reduces the dependence on the array error model and improves the positioning results. Accuracy and practicality.

CN118914970AActive Publication Date: 2024-11-08SICHUAN UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411113815.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-08
Estimated Expiration
2044-08-14

AI Technical Summary

Technical Problem

The existing signal source positioning methods have limitations in dealing with array amplitude phase errors, especially the insufficient processing capability of non-uniform linear arrays and random error models, resulting in limited positioning results deviations and practicality.

Method used

Using sparse recovery method combined with distribution robust optimization theory, signals are collected through multiple antennas, near-field signal reception model are constructed, SVD decomposition is performed, phase error model is established, and distribution robust optimization algorithm is used to transform it into convex problems to solve the target signal location information.

Benefits of technology

This method is suitable for receiving arrays of any structure. The error model established is more general. It does not require a complete understanding of all models of array amplitude phase error. It has a small calculation amount and can respond quickly, which significantly improves the robustness and practicality of the positioning result. sex.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118914970A_ABST
    Figure CN118914970A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of space signal source positioning, and particularly relates to a robust distributed signal level fusion positioning method. According to the method, on the basis of a traditional norm model, robust peer-to-peer transformation in a distributed robust optimization theory is used for converting a sparse signal recovery problem under disturbance into a solvable convex problem, and a credible solution when a receiving phase error exists is obtained. The method is theoretically suitable for receiving arrays of any structures, such as non-uniform linear arrays, so that the method has relatively high practicability; compared with an ideal signal receiving model, the error model established by the method is more general in form, and all models of the array amplitude-phase error do not need to be completely known, so that the positioning process of the signal source is more stable.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of spatial signal source positioning, and in particular relates to a robust distributed signal level fusion positioning method. Background Art

[0002] Positioning the signal source from space is an important part of distributed array signal processing. Engineering feasible positioning methods have great application value in radar systems, navigation, communications and other fields. However, in practical applications, due to the influence of measurement errors and electromagnetic fields, the ideal receiving model commonly used in the source direction estimation will deviate from the real scene, which is manifested as errors in the signal amplitude and phase output by different receiving array elements, resulting in mismatch between the steering vector and the real steering vector. This mismatch causes deviations in the results of the traditional method for source direction estimation.

[0003] The main method for dealing with array amplitude and phase errors in current research is to estimate the amplitude and phase errors from known signals and then correct them in the original model. Under the condition of external auxiliary sources, some scholars have proposed a method based on weighted least squares to jointly correct amplitude and phase errors and mutual coupling errors, but this method is applicable to uniform linear arrays. Under the condition of not relying on auxiliary sources, Friedlander et al. proposed a method for array self-correction, which corrects and compensates array errors by combining signal processing and optimization theory. This method is applicable to arrays of arbitrary structures, but there is a problem of convergence to local extreme values. Based on their work, Liu et al. proposed a combined strategy to compensate for amplitude errors; then independently estimate azimuth and phase errors, and finally weaken residual errors. This method solves the problem of local convergence, but is no longer applicable to linear arrays. However, the self-correction methods mentioned above all have obvious disadvantages: 1) They are only effective for arrays with specific regular structures, such as uniform linear arrays, which limits their practicality to a certain extent. 2) They all rely on an accurate model of array amplitude and phase errors, but this model is difficult to fully obtain and difficult to satisfy the random error model. Summary of the invention

[0004] In view of the above problems, the present invention considers the positioning problem under the condition of receiving phase error, uses sparse recovery means for positioning, and proposes a method for solving the target direction based on distributed robust optimization theory.

[0005] The technical solution adopted by the present invention is:

[0006] A robust distributed signal-level fusion positioning method comprises the following steps:

[0007] S1. Collect signal source signals through multiple antennas and build a near-field signal reception model:

[0008] y(t)=A(θ)s(t)+n(t)

[0009] Where A(θ) is the array manifold matrix, s(t) and n(t) are the signal and noise vectors at time t, respectively;

[0010] S2, reconstruct the received signal matrix, complete the SVD decomposition of the matrix, construct a sparse dictionary matrix, and establish a phase error model, specifically including:

[0011] Reconstruct y(t) as:

[0012]

[0013] in, M is the number of receiving array elements, T is the number of sampling snapshots, and N θ is the size of the overcomplete set. The overcomplete set is a set of directional grids divided in the spatial domain. The corresponding sparse signal matrix is:

[0014] X=[x1…x T ]

[0015] in Represents a sparse signal, converting the positioning problem into the x in the restored signal vector t The indices of the non-zero elements;

[0016] Perform SVD decomposition on matrix Y:

[0017] Y=ULV H

[0018] Among them, U and V are standard orthogonal matrices, and the columns of U and V are respectively composed of matrices YY T and the matrix Y T Y is composed of the normalized eigenvectors; L is the singular value matrix, and the diagonal elements of L are composed of Y T The singular values ​​of Y are arranged from large to small; retain the signal subspace, set the number of sources to K, and obtain the retained matrix:

[0019] Y SV =ULD K =YVD K

[0020] Where the matrix D is selected K =[I K 0] T , I K is the identity matrix of order K, and 0 is the zero matrix of K×(TK);

[0021] With the goal of minimizing the number of non-zero items in x to ensure the sparsity of the signal, the phase error model is established as:

[0022]

[0023] stvec(I K ) H vec(Δ)≤η

[0024] Among them, X SV =XVD K , is a sparse direction matrix, Γ is a diagonal matrix whose diagonal elements correspond to the phase errors of the corresponding array elements, ||X SV || 2,1 is to calculate X SV l 2,1 norm, η is the threshold of the constraint inequality;

[0025] S3. Using the distributed robust optimization algorithm, the phase error model is transformed into a convex problem and solved to obtain the position information of the target signal, specifically:

[0026] Using the distributed robust optimization algorithm, the phase error model is transformed into the following convex problem:

[0027]

[0028] Among them, β is an intermediate variable, which is a constant, and p is the probability that the constraint condition is satisfied. is the set of all possible probability distributions that δ obeys, δ = d(Γ), d(Γ) is the column vector consisting of the diagonal elements of the matrix Γ, The first-order moment of all probability distribution statistics in is μ, and the second-order moment is Σ.

[0029] By solving the convex problem, the location information of the target signal is obtained.

[0030] Beneficial effects of the present invention:

[0031] 1) This method is theoretically applicable to receiving arrays of arbitrary structures, such as non-uniform linear arrays, and therefore has strong practicality.

[0032] 2) Compared with the ideal signal receiving model, the error model established by the method of the present invention is more general in form and does not require full knowledge of the entire model of the array amplitude and phase errors, making the signal source positioning process more robust.

[0033] 3) This method involves fewer formulas and less computation, and can respond quickly to positioning results. At the same time, when processing multiple snapshot data, it has significant advantages over other methods because it decomposes and reduces the dimensionality of the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the overall process of the method of the present invention. DETAILED DESCRIPTION

[0035] The technical solution of the present invention is described in detail below with reference to the accompanying drawings.

[0036] like Figure 1 As shown, the method of the present invention mainly comprises the following steps:

[0037] S1. Collect signal source signals through multiple antennas and build a near-field signal reception model;

[0038] S2, signal preprocessing stage: reconstruct the received signal matrix, complete the SVD decomposition of the matrix, construct the sparse dictionary matrix, and establish the phase error model

[0039] S3. Using the distributed robust optimization algorithm, the phase error model is transformed into a convex problem and solved to obtain the position information of the target signal.

[0040] The specific implementation methods of each step are as follows:

[0041] Step S1: In general, the signal model can be divided into near-field and far-field models according to the distance between the receiving array and the signal source. Here, the near-field signal hypothesis is considered. Assuming that a narrowband signal is incident on a linear receiving array at a distance, the array samples the received signal at time t and is expressed as

[0042] y(t)=A(θ)s(t)+n(t) (1)

[0043] Where A(θ) is the array manifold matrix. s(t) and n(t) are the signal and noise vectors at time t, respectively. It is usually assumed that the signal and noise are independent of each other, the signals of each channel are incoherent, and the noise satisfies the Gaussian white noise assumption.

[0044]

[0045] Array manifold A=[a1(x1,y1)a2(x2,y2)…a K (x K ,y K )], where K is the number of target signal sources, a i (x i ,y i ) is the steering vector of the receiving array. The scenario studied in this paper considers a random linear array, and its steering vector is constructed as follows

[0046]

[0047] Where M is the number of receiving array elements, λ is the wavelength of the incident signal, (x,y) is the coordinate of the signal source, (x M ,y M) are the coordinates of the Mth array element.

[0048] Step S2: From the perspective of sparseness, equation (1) can be rewritten as

[0049]

[0050] in, T is the number of sampling snapshots, where N θ is the size of the overcomplete set. The overcomplete set can be considered as a set of directional grids divided by the spatial domain. Its corresponding sparse signal matrix

[0051] X=[x1…x T ] (4)

[0052] Among them represents a sparse signal, most of whose elements are 0, so we can use x in the restored signal vector t The index of non-zero elements replaces the positioning problem.

[0053] Due to the interference of noise terms, the signal x t There are additional non-zero items in . If only a single snapshot is considered, the sparse signal matrix X degenerates into a vector x. In the theory of compressed sensing, the goal of positioning should be to minimize the number of non-zero items in x to ensure the sparsity of the signal. This problem can be expressed by the l0 norm. Since the l0 norm is a non-convex function, there is no global extreme value, only local extreme values, so the l1 norm is used instead of the l0 norm. From this, the loss function of this problem can be written

[0054]

[0055] Where y is the array output under a single snapshot. Formula (5) can be written as a constraint form and solved using second-order cone programming:

[0056]

[0057] Where ε is the threshold of the similarity constraint, which is usually set to the noise power.

[0058] However, when there are multiple snapshots, the sparse signal X is sparse in space but not in time, so the vector norm cannot be used directly. In order to require the recovered signal to be sparse in space but not in time, we can calculate the l of X 2,1 norm, denoted by ||X|| 2,1 . Make the signal of multiple quick shots First, for a sparse signal under a specific spatial index Calculate the l2 norm of all its time samples (number of snapshots).

[0059]

[0060] Then calculate Similarly, the constraints in (6) can be rewritten as

[0061]

[0062] Here the Frobenius norm Therefore, the sparse reconstruction problem model of the received signal of multiple snapshots is:

[0063]

[0064] Where X is the sparse signal and ε is the threshold of noise power. Solving equation (9) yields the sparse signal solution From this, the direction of the target signal source can be obtained.

[0065] However, when the number of snapshots is large, directly solving equation (9) will take a long time due to the large data dimension. The SVD decomposition method can be used to reduce the dimension of the data. The basic idea is to decompose the data matrix into two subspaces, signal and noise, retain only the signal subspace, and then substitute the reduced-dimensional data into the previous problem.

[0066] Perform SVD decomposition on matrix Y

[0067] Y=ULV H

[0068] Among them, U and V are standard orthogonal matrices, and the columns of U and V are respectively composed of matrices YY T and the matrix Y T Y is composed of the normalized eigenvectors; L is the singular value matrix, and the diagonal elements of L are composed of Y T The singular values ​​of Y are arranged from large to small;

[0069] K is the number of information sources, so the retained matrix

[0070] Y SV =ULD K =YVD K (10)

[0071] Where the matrix D is selected K =[I K 0] T , I K is the identity matrix of order K, 0 is the zero matrix of K×(TK), Y SV In addition, let X SV =XVD K ,N SV =NVD SVIn order to construct

[0072]

[0073] Note that compared to X in (3), X SV The number of columns is reduced, but the spatial sparsity does not change, so it has no effect on positioning. In this case, (9) can be rewritten as

[0074]

[0075] Where η is the threshold of the constraint inequality, which is related to the noise power. Although the solution obtained in (12) is the signal after dimensionality reduction But it does not affect the direction information of the target signal source, because It is sparse, with only a few rows of data having large values, and the index i of these rows corresponds to the coordinates of the target signal.

[0076] Considering only the phase errors between different receiving array elements, let Γ be a diagonal matrix whose diagonal elements are the corresponding array element phase errors. According to the sparse signal recovery model (9), the azimuth problem under array error can be written as

[0077]

[0078] Use SVD to decompose (13) and reduce the dimension.

[0079]

[0080] in

[0081]

[0082] According to matrix theory knowledge

[0083]

[0084] Where K is the dimension of the data after dimensionality reduction, i.e., the number of information sources, I K is a K-order identity matrix, then (15) can be expanded to

[0085]

[0086] make Then equation (16) can be simplified as

[0087]

[0088] According to the transformation formula of Kronecker product

[0089]

[0090] Then formula (14) is equivalent to

[0091]

[0092] in

[0093]

[0094] The uncertain terms in the existing distributed robust optimization problem are all in vector form, and the error matrix Γ here is a diagonal matrix. In order to use the distributed robust optimization algorithm, all Γ need to be vectorized and the matrix conversion is used.

[0095]

[0096] where ⊙ represents the Khatri-Rao product, which is the Kronnecker product operation performed on each column, d(Γ) = [γ1γ2…γ M ] T is a column vector consisting of the diagonal elements of the matrix Γ. Let d(Γ) = δ. In this problem, δ is considered to be the uncertainty term (disturbance) in the distribution-robust problem. Therefore, after the expansion of equation (18), it can be written as

[0097]

[0098] Notice It is the signal X after SVD dimensionality reduction. SV The power of the received signal Y is equal to the power of the original signal in theory. Therefore, to simplify the calculation, it is assumed in this paper that the power of the original signal and the number of signal sources K are known. SV It is known in engineering that

[0099]

[0100] Then formula (19) can be written as

[0101]

[0102] make

[0103]

[0104] Then formula (20) can be written as

[0105]

[0106] Since it involves the operation of taking the real part, let Then formula (21) can be written as

[0107]

[0108] in

[0109]

[0110]

[0111] Disturbance d Contains the array phase error, which needs to be known in advance according to the theory of distributed rod optimization d The probability distribution of or the fuzzy set of probability distribution constructed based on some information. In this paper, it is assumed that the phase error matrix The phase term in Independently obey a certain uniform distribution. Under this assumption, let d The expectation of the first-order moment is μ, and the variance of the second-order moment is Σ.

[0112] Step S3: Write the original problem model of distributed robust optimization as

[0113]

[0114] where w is the beam weight vector, is an imaginary steering vector, and δ is the error of the steering vector caused by noise and interference, that is, the uncertain disturbance term. For δ, the distribution The collection Given the known situation, the above beamforming optimization problem can be written as the following distributed robust optimization form:

[0115]

[0116] Considering equation (22), the known perturbation d After the first-order moment μ and the second-order moment Σ, we can construct d The fuzzy set of all possible probability distributions of . Referring to the problem of distributed robust beamforming, the uncertain parameters in the problem are exactly the phase error d , so the constraint of (22) is written in the form of distributed robustness

[0117]

[0118] Where η is the threshold of the constraint inequality, which is related to the noise power, and p is the probability that the constraint condition is satisfied. is the probability of all possible results of δ obeying The set of distributions, that is The first-order moment of the statistic for all probability distributions in is μ, and the second-order moment is Σ. The phase error of the receiving array element is When the independent and identically distributed assumption is met, the distributional robust optimization problem of formula (26) can be transformed into a solvable convex problem using robust equivalence transformation, that is, formula (26) is equivalent to formula (27)

[0119]

[0120] in, μ and Σ are The mean and variance of , β is an intermediate variable, which is a constant, is a continuous loss function, L can be a concave function or a non-concave function, and L is also a quadratic function of ξ.

[0121] In summary, the method of the present invention is based on the traditional norm model, and uses the robust equivalence transformation in the distributional robust optimization theory to transform the sparse signal recovery problem under disturbance into a solvable convex problem, and obtains a credible solution in the presence of a received phase error. Through simulation comparison of traditional signal reconstruction algorithms, traditional subspace methods (MUSIC) and IAA algorithms under a small number of snapshots, the method of the present invention still maintains a high accuracy under sufficient phase errors, proving that the method has good robustness to array reception errors.

Claims

1. A robust distributed signal-level fusion positioning method, characterized in that: The following steps are involved: S1. Collect signal source signals through multiple antennas and build a near-field signal reception model: y(t)=A(θ)s(t)+n(t) Where A(θ) is the array manifold matrix, s(t) and n(t) are the signal and noise vectors at time t, respectively; S2, reconstruct the received signal matrix, complete the SVD decomposition of the matrix, construct a sparse dictionary matrix, and establish a phase error model, specifically including: Reconstruct y(t) as: in, M is the number of receiving array elements, T is the number of sampling snapshots, and N θ is the size of the overcomplete set. The overcomplete set is a set of directional grids divided in the spatial domain. The corresponding sparse signal matrix is: X=[x1…x T ] in Represents a sparse signal, converting the positioning problem into the x in the restored signal vector t The indices of the non-zero elements; Perform SVD decomposition on matrix Y: Y=WOLF H Among them, U and V are standard orthogonal matrices, and the columns of U and V are respectively composed of matrices YY T and the matrix Y T Y is composed of the normalized eigenvectors; L is the singular value matrix, and the diagonal elements of L are composed of Y T The singular values ​​of Y are arranged from large to small; retain the signal subspace, set the number of sources to K, and obtain the retained matrix: Y SV =ULD K =YVD K Where the matrix D is selected K =[I K 0] T , I K is the identity matrix of order K, and 0 is the zero matrix of K×(TK); With the goal of minimizing the number of non-zero items in x to ensure the sparsity of the signal, the phase error model is established as: stvec(I K ) H thing(Δ)≤η Among them, X SV =XVD K , is a sparse direction matrix, Γ is a diagonal matrix whose diagonal elements are the phase errors of the corresponding array elements, ||X SV || 2,1 is to calculate X SV l 2,1 norm, η is the threshold of the constraint inequality; S3. Using the distributed robust optimization algorithm, the phase error model is transformed into a convex problem and solved to obtain the position information of the target signal, specifically: Using the distributed robust optimization algorithm, the phase error model is transformed into the following convex problem: Among them, β is an intermediate variable, which is a constant, and p is the probability that the constraint condition is satisfied. is the set of all possible probability distributions that δ obeys, δ = d(Γ), d(Γ) is the column vector consisting of the diagonal elements of the matrix Γ, The first-order moment of all probability distribution statistics in is μ, and the second-order moment is Σ. By solving the convex problem, the location information of the target signal is obtained.

Citation Information

Patent Citations

  • Distributed type sparse optimization detecting method of water regime information

    CN103533557A

  • Robust direction of arrival (DOA) estimation method based on sparse and low-rank recovery

    CN110045321A

  • Inpainting based signal gap reconstruction with sparse representation

    EP3940696A1